240 24 33MB
English Pages 357 Year 2015
Hispanic Linguistics at the Crossroads
Issues in Hispanic and Lusophone Linguistics (IHLL) issn 2213-3887
IHLL aims to provide a single home for the highest quality monographs and edited volumes pertaining to Hispanic and Lusophone linguistics. In an effort to be as inclusive as possible, the series includes volumes that represent the many sub-fields and paradigms of linguistics that do high quality research targeting Iberian Romance languages. IHLL considers proposals that focus on formal syntax, semantics, morphology, phonetics/phonology, pragmatics from any established research paradigm, as well as psycholinguistics, language acquisition, historical linguistics, applied linguistics and sociolinguistics. The editorial board is comprised of experts in all of the aforementioned fields.
For an overview of all books published in this series, please see http://benjamins.com/catalog/ihll
Editors Jason Rothman
University of Reading
Jennifer Cabrelli Amaro
University of Illinois at Chicago
Editorial Board Sonia Colina
University of Arizona
João Costa
Universidade Nova de Lisboa
Inês Duarte
Universidade de Lisboa
Sónia Frota
Universidade de Lisboa
Ángel J. Gallego
University of Barcelona
María del Pilar García Mayo Universidad del País Vasco
Anna Gavarró
Universitat Autònoma de Barcelona
Kimberly L. Geeslin Indiana University
Michael Iverson
Liliana Sánchez
Paula Kempchinsky
Ana Lúcia Santos
Juana M. Liceras
Carmen Silva-Corvalán
Macquarie University University of Iowa
University of Ottawa
Rutgers University Universidade de Lisboa
John M. Lipski
University of Southern California
Gillian Lord
University of Maryland
Jairo Nunes
University of Ottawa
Acrisio Pires
Michigan State University
Pennsylvania State University University of Florida Universidade de São Paulo
Juan Uriagereka
Elena Valenzuela Bill VanPatten
University of Michigan, Ann Arbor
Pilar Prieto
Universitat Pampeu Fabra
Volume 4 Hispanic Linguistics at the Crossroads. Theoretical linguistics, language acquisition and language contact. Proceedings of the Hispanic Linguistics Symposium 2013 Edited by Rachel Klassen, Juana M. Liceras and Elena Valenzuela
Hispanic Linguistics at the Crossroads Theoretical linguistics, language acquisition and language contact Proceedings of the Hispanic Linguistics Symposium 2013 Edited by
Rachel Klassen Juana M. Liceras Elena Valenzuela University of Ottawa
John Benjamins Publishing Company Amsterdam / Philadelphia
8
TM
The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.
doi 10.1075/ihll.4 Cataloging-in-Publication Data available from Library of Congress: lccn 2015004262 (print) / 2015011030 (e-book) isbn 978 90 272 5803 8 (Hb) isbn 978 90 272 6860 0 (e-book)
© 2015 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
Table of contents Preface Introduction
vii xi
part i. Theoretical and descriptive approaches No superiority, no intervention effects: The Spanish puzzle Lara Reglero & Emma Ticio
3
Overt PRO in Romance: Towards a unification of PRO and pro Peter Herbeck
25
The semantics and pragmatics of andar and venir + gerund Lorena Sainz-Maza Lecanda & Meagan E. Horn
49
Sequence of tenses in complementation structures: Lexical restrictions and effects on language acquisition Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos Fue muerto: Suppletion in Spanish analytic passives Grant Armstrong & Jason Doroga Temporal and spectral dependencies in the processing of Spanish and English stop consonant voicing Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz
69 89
113
Segmental and prosodic conditionings on gradient voicing assimilation in Spanish 127 Rebeka Campos-Astorkiza part ii. Language acquisition The sum is more than its parts: Acquisition of clitic clusters in Spanish María Cristina Cuervo & Ana Teresa Pérez-Leroux Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals? José Camacho & Alena Kirova Perfecting the past Joanne Markle LaMontagne & María Cristina Cuervo
147
169 191
Hispanic Linguistics at the Crossroads
The protracted acquisition of past tense aspectual values in child heritage Spanish Alejandro Cuza & Lauren Miller
211
part iii. Language contact and language variation Morphological adjectival intensifier variation in Lima, Peru Álvaro Cerrón-Palomino
233
An experimental approach to hypercorrection in Dominican Spanish Barbara E. Bullock & Almeida Jacqueline Toribio
251
Dialect identification and listener attributes: Do you hear la tonada? Jennifer Lang-Rigal
269
Sociophonetic analysis of young Peninsular Spanish women’s voice quality Meghan E. Armstrong, Nicholas Henriksen & Christian DiCanio
293
A sociophonetic analysis of trill production in Panamanian Spanish Delano S. Lamy
313
Index
337
Preface The selection of refereed chapters included in this volume were presented at the 17th Hispanic Linguistics Symposium hosted by the Language Acquisition Research Laboratory of the Department of Modern Languages at the University of Ottawa in October 2013 in Ottawa, Canada. Following the tradition that was established in 2006, this Symposium is the merge of what started as two different conferences: the Conference on the Acquisition of Spanish and Portuguese as First and Second Languages (first held at Penn State in 1995) and the Hispanic Linguistic Symposium (established in 1997 at Miami University in Oxford, OH). Although these two conferences were celebrated together from 1999 to 2005, it wasn’t until 2006 that they actually merged under the umbrella of the Hispanic Linguistic Symposium. The 17th edition of the Hispanic Linguistics Symposium featured four plenary speakers – Ana Arregui, Eulàlia Bonet, Jason Rothman, and Liliana Sánchez – who represented a wide range of the areas traditionally covered in the Symposium including semantics, syntax, morphology, phonology, language contact and acquisition. Specifically, Ana Arregui provided an analysis of a large body of examples in an attempt to show how to derive and cancel counterfactuality implicatures in conditionals. Eulàlia Bonet argued that even though in its origins the el/la alternation of Spanish had a (partially) phonological cause and could be analyzed using a single underlying morph, in some varieties of Spanish the phenomenon has expanded and been reanalyzed by speakers as a case of lack of concord in prenominal position, a phenomenon found in other languages. In Spanish, she further argued, the concord asymmetry is triggered by a specific set of lexical items. Jason Rothman’s talk was primarily focused on identifying the idiosyncrasies of L3 acquisition in relation to L2 acquisition. He argued that L3 acquisition differs from L2 acquisition in that bilinguals bring to L3 acquisition: (i) superior metalinguistic knowledge; (ii) the fact that L3 learners have access to more grammatical options; (iii) cognitive benefits to executive functions and attention that enable bilinguals to ignore irrelevant information, resolve conflict among competing alternatives and minimize costs associated with task switching. Liliana Sánchez discussed the emergence of new morphological patterns in the L2/Bilingual Spanish of native speakers of morphologically diverse languages such as Quechua, Mandarin and Shipibo. She proposed that crosslinguistic influence is favoured by the existence of highly divergent morphological patterns that guide the organization of the lexicon in the L1 and the L2.
Hispanic Linguistics at the Crossroads
There were 77 oral presentations and 28 posters presented, covering topics including diachronic change, classroom learning, language acquisition, language contact and variation, morphology, phonology, semantics and pragmatics, and syntax. Scholars came from different parts of Europe, Latin America and North America. We are grateful to the institutions that provided funding for this edition of the Symposium, including the Faculty of Arts and the Department of Modern Languages at the University of Ottawa, as well as the Embassy of Spain in Ottawa. We would also like to thank all the abstract reviewers: Jessi Aaron, Lourdes A guilar, Anahí Alba de la Fuente, Gabriela Alfaraz, Scott Alvord, Patrícia A maral, K arlos Arregi, Julie Auger, Marc Authier, Isabelle Barriere, Aurora Bel, Ricardo Bermúdez- Otero, Eulàlia Bonet, Travis Bradley, Joyce Bruhn de Garavito, Monica Cabrera, José Camacho, Richard Cameron, Walcir Cardoso, Ana Carvalho, Laura Colantoni, Sonia Colina, João Costa, Marie-Hélène Côté, María Cristina Cuervo, Alejandro Cuza, Verónica González López, Grant Goodall, John Grinstead, Aarnes Gudmestad, Pedro Guijarro-Fuentes, Javier Gutiérrez-Rexach, David Heap, José Esteban Hernández, Jonathan Holmquist, Jose Ignacio Hualde, Michael Iverson, Tiffany Judy, Cynthia Kauffeld, Paula Kempchinsky, Carol Klee, Alexei Kochetov, Karen Lahousse, Manuel Leonetti, Juana Liceras, John Lipski, Ruth Lopes, Gillian Lord, Fernando MartínezGil, Ana-Maria Martins, Stefania Marzo, Jaume Mateu, Silvia Perpiñán, Pierre Pica, Carme Picallo, Carlos Piera, Carlos Eduardo Piñeros, Acrisio Pires, Cecilia Poletto, Shana Poplack, Chris Pountain, Fernando Ramallo, Rajiv Rao, Lisa Reed, Lara Reglero, Yves Roberge, Ian Roberts, Celeste Rodríguez Louro, Maribel Romero, Jason Rothman, Cinzia Russi, Nuria Sagarra, Liliana Sánchez, Teresa Satterfield, Leonardo Savoia, Cristina Schmitt, Sarah Sinnott, and Petra Sleeman. Last but not least, we would like to express our gratitude to the faculty, administrative personnel, graduate students, and volunteers from the Departments of Modern Languages and Linguistics at the University of Ottawa. The selected proceedings of the Conference on the Acquisition of Spanish and Portuguese as First and Second Languages and the Hispanic Linguistic Symposium from 1999 until 2012 have been published with Cascadilla Press. This volume is the first that will appear in the series Issues in Hispanic and Lusophone Linguistics published by the John Benjamins Publishing Company. We are grateful to the following reviewers who contributed their expertise to the evaluation process for the proceedings: Mark Amengual, Jose Amenós, P atricia Andueza, Ana Arregui, Aurora Bel, Beatriz Blecua, Eulàlia Bonet, Marda C. Rose, Jennifer Cabrielli Amaro, João Costa, María Cristina Cuervo, Alejandro Cuza, Lourdes Díaz, Miriam Díaz, Manuel Díaz-Campos, John Esling, Antonio Fábregas, Olga Fernández Soriano, Raquel Fernández-Fuertes, Gerardo Fernández-Salgueiro, Christoph Gabriel, Juana Gil, Michael Gradoville, Lourdes Güell, Javier GutiérrezRexach, David Heap, Nieves Hernandez-Flores, Michol Hoffman, Jose I gnacio Hualde,
Preface
Paula Kempchinsky, Miran Kim, John Lipski, Conxita Lleó, Mireia Llinas Grau, Gillian Lord, Cristóbal Lozano, Oana Lungu, Jonathan MacDonald, Jeff MacSwan, Shiti Malhotra, Roberto Mayoral Hernandez, Sandra Montserrat, Rafael NuñezCedeño, Daniel Olson, Francisco Ordóñez, Erin O’Rourke, Rafael Orozco, Ricardo Otheguy, Diego Pascual y Cabo, Ana Teresa Pérez-Leroux, Carmen Picallo, Jeffrey Punske, Rajiv Rao, Liliana Sanchez, Esther Torrego, Paul Toth, Annie Tremblay, and Julio Villa-García.
Introduction This volume presents a collection of articles that address core issues in three different domains of Hispanic linguistics: theoretical linguistics, language acquisition and language contact. Together these papers provide an overview of how the analysis of Spanish can contribute to current formal and experimental linguistics, while on an individual level offering fine-grained analyses and innovative proposals covering a wide range of areas such as semantics and pragmatics, syntax, morphology, phonology, prosody, dialectal variation, first, second and bilingual language acquisition, as well as sociolinguistics. The first part of the volume consists of seven chapters. In the first chapter, No superiority, no intervention effects: The Spanish puzzle, Reglero and Ticio deal with Superiority and Intervention Effects in wh-phrases that are said to be in complementary distribution in natural languages: a given language has either Superiority Effects or Intervention Effects but not both. The authors show that such a complementary distribution does not hold in Spanish in that in situ whphrases show neither of these effects. This implies that ¿Quién compró qué?, where there is no Superiority Effect, and ¿Qué compró quién?, where there is a Superiority Effect since qué proceeds quién, are both possible in Spanish. It also implies that both ¿Quién ha comprado qué?, where there is no Intervention Effect, and ¿Qué no ha comprado quién?, where there is an Intervention Effect given that the negative operator no precedes the in situ wh-phrase quién, are grammatical. The authors posit that it is the different nature of the Force Phrase and the binding mechanisms in Spanish compared to those present in other languages (such as English and German) that account for the distinct behaviour of in situ wh-phrases in Spanish. In the second chapter, Overt PRO in Romance: Towards a unification of PRO and pro, Herbeck shows that the availability and distribution of nominative overt subjects in both obligatory and non-obligatory control contexts in Spanish (and Romance) provide evidence for the unification of the categories PRO and pro. He bases his proposal on the assumption that null subjects are minimally-specified nominal elements whose phonological emptiness derives from a lack of a phi-specification at PF. After showing that overt pronouns have a different distribution than DPs and R-expressions, the author proposes that overt ‘pronominal’ subjects in infinitives are licensed by [π] (a discourse-sensitive feature like [focus], [emphasis] or [contrast]), and not by Case (they are not sensitive to finiteness). He further argues that the overt/covert alternation of subjects in pro-drop and control depends on the derivational point at which a subject interfaces with external systems and that the ordering of insertion and valuation is crucially mediated by the pragmatic interface.
Hispanic Linguistics at the Crossroads
In the third chapter, The semantics and pragmatics of andar and venir + gerund, Sainz-Maza Lecanda and Horn show that, while in terms of syntax the aspectual periphrasis andar + gerund and venir + gerund can occur in monoclausal and biclausal contexts, it is only the monoclausal contexts that are relevant for the different semantic and pragmatic characteristics they display. Specifically, the authors show that both periphrases behave as frequentative pluractional operators (they denote a multiplicity of events) but differ in two respects: first with respect to the frequency of the event repetitions and second with regards to two conversational implicatures. Regarding the frequency of event repetitions, venir + gerund entails a higher degree of event frequency whereas with andar + gerund the event repetitions are less frequent and occur less regularly. As for conversational implicatures, the andar + gerund periphrasis implies more indifference and less involvement towards the agent’s accomplishment of the goal than venir + gerund. Furthermore, the level of the agent’s involvement is low in the case of andar + gerund but high in the case of venir + gerund. In the fourth chapter, Sequence of tenses in complementation structures: Lexical restrictions and effects on language acquisition, Marques, Silvano, Gonçalves and Santos argue that in European Portuguese lexical restrictions on tenses in embedded clauses are primarily semantic in nature and, consequently, independent of mood. They show that it is the type of verb that imposes restrictions in complementation structures since a main clause in the present with querer (to want) requires a subordinate clause in the present, while a verb like dizer (to say) does not impose restrictions on the sequence of tenses. This provides evidence that even though temporal dependence is a relevant mechanism in the sequence of tenses, semantic lexical restrictions must also be taken into consideration. They specifically argue that the ungrammaticality of present under past structures with volition, doxastic (belief) and negative verbs illustrate the importance of this in that the sequence is only possible when the set of worlds denoted by the complement clause includes the real world. The authors also analyze the spontaneous production of three children (Santos, 2006 corpus in CHILDES) in order to determine whether children produce structures of verbal complementation where main and embedded clauses form a single temporal domain and whether children are aware of the semantic lexical restrictions on the sequence of tenses that characterize European Portuguese. They conclude that at the initial stages there is an avoidance of temporal dependence in the production of finite complement clauses and a total absence of lexically-restricted combinations of tenses of the type pres+past, past+past (perf) or past+pres. This absence, according to the authors, should be attributed to the fact that lexical restrictions are acquired later than the general possibilities of the grammar. In the fifth chapter, Fue muerto: Suppletion in Spanish analytic passives, taking as a point of departure the morphological and syntactic behaviour of Spanish analytic passives such as ser muerto, Armstrong and Doroga argue that the roots of mat-ar
Introduction
(to kill) and mor-ir (to die) should be considered instances of root allomorphy. To this end they use the grammatically-conditioned alternation between the words typically associated with these two Spanish verbs to argue that suppletion or root allomorphy is an empirical reality in natural languages. Specifically, using the tools of Distributed Morphology, they posit that the Spanish root mat- is specified to appear in agentive verbal environments while mor- is an elsewhere morpheme. They show that there is a much wider range of insertion possibilities for mor- than for mat-. The authors’ analysis leads them to join the broad view of grammar, also put forward by Distributed Morphology, the exo-skeletal model and Nanosyntax, according to which words are not atomic units that are opaque to syntactic operations but are products of syntactic operations interpreted at the interfaces. In the sixth chapter, Temporal and spectral dependencies in the processing of Spanish and English stop consonant voicing, Llanos, Dmitrieva, Francis and Shultz examine two competing explanations (a temporal-based account and a spectralbased account) of the relative contribution of Voice Onset Time (VOT) and onset f0 to the perceptual categorization of stop consonant voicing in Spanish and English. The temporal-based account states that the perceptual contribution of onset f0 is lower because the voicing decision has been previously biased by the presence of prevoicing that precedes the occurrence of onset f0 in the signal. The spectral-based account assumes that onset f0 has no effect because its contribution to low frequency energy is insignificant compared to that of VOT in voicing lead. To test both accounts the authors analyzed a perceptual database (Llanos, 2013) of voicing decisions made by 40 participants (20 native speakers of Peninsular Spanish and 20 native speakers of American English). The task consisted of perceptual responses to a binary forcedchoice task. The authors conclude that the results support the spectral-based account, since the peak effect of onset f0 occurred in both languages at the VOT values where onset f0’s contribution to the low frequency energy would be most appreciable. In the seventh chapter, Segmental and prosodic conditionings on gradient voicing assimilation in Spanish, Campos-Astorkiza presents an experimental analysis of assimilation in the voicing of /s/ before a voiced consonant. She analyzes data recorded from 8 native speakers of Northern Peninsular Spanish who were asked to repeat sentences containing voicing during /s/. She examines the percent voicing and the voicing category, on the one hand, and the stress patterns on the degree of voicing on the other. The results show that the variable degree of sibilant voicing is dependent upon the location of stress, the manner of articulation of the triggering consonant, and the type of prosodic boundaries. The author argues that this gradient assimilatory process is modeled as the result of blending of the glottal gestures for the adjacent voiceless and voiced sounds, which is the reason why the degree of voicing is subject to influence by segmental and supra-segmental information. The author further argues that it is only by exploring the effects of both prosodic and
Hispanic Linguistics at the Crossroads
s egmental material, and their interplay, that the nature of the reported assimilation can be properly accounted for. The second part of the volume includes four chapters that are devoted to discussing the L1 and the L2 acquisition of Spanish. In the eighth chapter, The sum is more than its parts: Acquisition of clitic clusters in Spanish, Cuervo and PérezLeroux provide an analysis of the L1 acquisition of clitic clusters by two children (María and Emilio) exposed to Peninsular Spanish. The data analyzed from these two children, which is available in CHILDES, covers the ages of 1;09–2;11 and 1;10–3;00, respectively. After determining the number of contexts that could potentially contain a clitic cluster and comparing those with the contexts in which a cluster was actually produced, the cluster contexts were classified as dative/accusative (ditransitive constructions consisting of an accusative theme and a dative recipient); inchoative/dative (inchoative constructions appearing with a dative argument); or aspectual/accusative (a reflexive construction with aspectual se, an optional intensifier common with verbs of consumption). Specifically, the authors show that even though both children produce clitic clusters in dative/accusative contexts first, the rate of cluster production in this context is lower than in inchoative/dative and aspectual/accusative contexts. The data also show that once these two children move past a short no-cluster stage, they produce clitic clusters at adult-like productivity rates and almost without errors (only an 8% error rate which is only slightly higher than their error rate of 1% to 4% with individual clitics). Furthermore, those errors are more often errors of omission than errors of commission. In the ninth chapter, Does agreeement affect the syntax of bare nominal subjects in Russian-Spanish bilinguals?, Camacho and Kirova discuss experimental data intended to investigate the status of bare nominal subjects in the L2 Spanish of L1 Russian speakers. They specifically aim at determining whether grammatical gender, which is available in both Russian and Spanish, facilitates the acquisition of the structure and distribution of bare nominals in Spanish. After showing that Russian and Spanish differ in the distribution of bare NPs and overt full DPs but display similar agreement and concord systems, the authors hypothesize that the two structures would be dealt with differently and that the different gender markings (–a, -o, and -e) would not have the same effect on the nominal structure and the distribution of bare nominals. The results of an acceptability judgment task administered to 36 L1 Russian learners of Spanish and 6 native speakers of Spanish show that the two groups treat bare NPs and full overt DPs differently, though the difference is much higher for the L1 Spanish speakers. This indicates that the structure and distribution of bare nominals in Spanish constitutes an area of difficulty for L1 Russian speakers. The results also show that there is a facilitating effect of the -a marking on the acquisition of the nominal structure in Spanish which cannot be explained under the traditional division between masculine/feminine–ending words (–o/–a) and words not overtly marked for gender
Introduction
(–e). The authors posit that it is the analysis of -a as a word marker which encodes an abstract gender feature in the syntax and of -o words as lacking a gender feature that can account for these results. In the tenth chapter, Perfecting the past, Markle LaMontagne and Cuervo investigate the L1 acquisition of the temporal and aspectual properties of Spanish present perfect (PP). Using spontaneous data from María (1;09–3;11) and Emilio (1;09–3;10) available in CHILDES, the authors show that these two children use the PP form before the preterit form. However, their use of the PP form does not entail the complex indirect referential meaning that this form conveys in adult Peninsular Spanish. That is, while four different uses of the PP have been identified in adult Peninsular Spanish (perfect of result, recent past, experiential perfect, and perfect of persistent situation), the analysis of the children’s longitudinal data reveals that while the PP accounts for 70% of the past reference in María and Emilio’s data, almost all of their uses (94%) correspond to the recent past and perfect of result. The authors argue that there is no correlation between the productive use of the PP form and the acquisition of the complex temporal and aspectual properties that it encodes in adult Spanish. In the eleventh chapter, The protracted acquisition of past tense aspectual values in child heritage Spanish, Cuza and Miller address the source of the difficulties Spanish heritage children in the U.S. encounter with the aspectual distinction between two past tenses: the preterit and the imperfect. The authors compare the production of 19 Spanish-English bilingual children born and raised in the U.S. with that of 12 of their parents (long-term Mexican immigrants) in an elicited production task. The results show that while both the older and the younger children display stable use of the preterit, they underproduce the imperfect, particularly with activity and accomplishment predicates. The fact that the parent group performed at ceiling implies that the input the children received was not reduced and would not lead to incomplete acquisition. As for age, the fact that there was no significant effect of the age of the children on their performance indicates that increased contact with English through the school system did not significantly affect the use of aspect in their Spanish and consequently underproduction of the imperfect cannot be attributed to attrition. Thus, since the children’s underproduction of imperfect can neither be attributed to incomplete acquisition nor to attrition, the authors conclude that the results are consistent with cross-linguistic influence from English in which this morphological aspectual distinction does not exist and thus the preterit is the default or unmarked form. The final part of the volume contains five chapters that examine Spanish in situations of language contact. In the twelfth chapter, Morphological adjectival intensifier variation in Lima, Peru, Cerrón-Palomino addresses the alternation between the standard adjectival intensifier -ísimo and the non-standard variant -azo in Limeño Spanish. Using a novel data collection technique, the author
Hispanic Linguistics at the Crossroads
extracted 1102 occurrences of -ísimo and -azo as adjectival intensifiers from online corpora, including written data from magazines, journals, and digitalized books as well as e-networks such as Facebook, Twitter, YouTube and other Peru-specific online forums and blogs. In his analysis of the factors responsible for the expansion of -azo over -ísimo in this variety of Spanish, he found that social factors were the strongest predictors of the use of -azo over -ísimo, namely, colloquial styles and male speakers. Other important predicting factors included adjectives with a negative value, predicative structures, and participial adjectives. Specifically, stative/ resultative constructions of the form estar + participle with an experiencer argument favour the use of -azo. The author concludes that -azo is a colloquial intensifier variant that appears to be more specialized than -ísimo, which is more formal and has a broader distribution. In the thirteenth chapter, An experimental approach to hypercorrection in Dominican Spanish, Bullock and Toribio investigate two types of [s] hypercorrection in rural Dominican Spanish: quantitative hypercorrection, namely, the overuse of etymological [s] in an attempt to emulate the speech of others; and qualitative hypercorrection, or the use of [s] in a linguistic context in which it is not historically warranted. In order to assess whether these speakers could perceive and correctly reproduce coda [s] to the same degree across different phonological contexts, the authors carried out an elicited repetition study in which 27 children and 24 adults from rural Cibao were asked to repeat novel names containing target [s]. The authors show that literate speakers correctly reproduced [s] (like etymological [s] in natural speech) at significantly higher rates than those who are semiliterate. With respect to following phonological environment, results show that voiceless stops are the environment most favourable for all speakers’ reproduction of [s]. In terms of cases in which [s] was moved from its original position (intrusive [s]), the authors found that literate and semiliterate speakers produced intrusive [s] at the same rates. Given that the speakers did not differ quantitatively or qualitatively in the production of intrusive [s], Bullock and Toribio conclude that [s] hypercorrection is not conditioned by literacy. They further suggest that intrusive [s] and etymological [s] may be entirely different entities given that they are subject to different linguistic and social conditioning. This implies that only the production of etymological [s] is socially conditioned and that the use of intrusive [s] is not the result of lack of knowledge about lexical /s/ distribution. In the fourteenth chapter, Dialect identification and listener attributes: Do you hear la tonada?, Lang Rigal investigates the perception of vowel lengthening that distinguishes the Spanish spoken in Córdoba, Argentina from other varieties of Argentinian Spanish. To this end, a matched-guise dialect perception task was created by extracting spontaneous speech samples from recordings of speakers from three different regions in Argentina (Córdoba, Tucumán, and Buenos Aires) and creating pairs of tokens in which one of the tokens was the original recording and the other
Introduction .
was manipulated with respect to the duration of the vowel in the pre-tonic syllable. In the identification task completed over the internet, 63 speakers of Argentinian Spanish from different regions indicated whether each token belonged to a speaker from Córdoba, Tucumán, or Buenos Aires. Results show that dialect identification accuracy was highest when speakers were listening to tokens from their own region. Furthermore, the accuracy rates for the identification of the manipulated tokens was significantly lower than accuracy rates for the tokens taken from the original recordings, thus indicating that manipulating the pre-tonic vowel length has a significant effect on the correct identification of the dialect. The author therefore concludes that pre-tonic vowel lengthening is associated with the Cordoba identity and that this cue alone is sufficient to influence the perception of a dialect. In the fifteenth chapter, Sociophonetic analysis of young Peninsular Spanish women’s voice quality, Armstrong, Henriksen and DiCanio examine the acoustic correlates of the impressionistic observation that some young Peninsular Spanish women’s voice quality seems more hoarse than others as well as the socio-indexical information associated with different types of voice quality. In a production experiment, they acoustically analyze narrated speech data from 3 modal and 3 non-modal female speakers of Peninsular Spanish between the ages of 18 and 23. Results show that the perception differences between the modal and non-modal speakers were likely due to the significantly higher H1-H2 values for modal than non-modal speakers. In order to investigate whether the differences in voice quality carried social meaning, the authors surveyed 50 participants aged 21–77 living in Spain, asking them what the speaker was like, where the person was from and whether the person smoked, based on listening to audio clips of the 6 speakers from the production experiment. While modal speakers were described as feminine and intelligent, non-modal speakers were judged to be apathetic and strong. Modal speakers were also more often said to be from an urban or city zone than non-modal speakers and non-modal speakers tended to be classified as smokers more often than modal speakers. Overall, the authors show that there are within-gender acoustic differences in voice quality and that listeners make use of these differences in their classification and judgment of speakers. In the sixteenth chapter, A sociophonetic analysis of trill production in Panamanian Spanish, Lamy presents a detailed account of variable trill production in Panamanian Spanish within a variationist sociolinguistic framework. While standard alveolar trills in Spanish have been defined as a segment consisting of two or three occlusions, many other non-standard trill realizations have been documented and shown to vary according to linguistic and extralinguistic factors. For this study, the author analyzes data from sociolinguistic interviews with 11 monolingual speakers of Panama City Spanish who represent different age groups and socioeconomic classes. An acoustic analysis based on the presence of occlusions, waveform amplitude reduction, transition in formant structure, and reduced intensity showed that there are eight trill
Hispanic Linguistics at the Crossroads
variants in Panama City Spanish. Lamy further shows that these eight variants are significantly conditioned by preceding segment type, syllable stress, number of syllables, and grammatical category, as well as speaker age, and speaker sex. Higher word frequency is also shown to favour non-normative variants (i.e. variants with fewer than two occlusions), leading the author to conclude that it is important to take frequency effects into account in future analyses of variable trill production.
part i
Theoretical and descriptive approaches
No superiority, no intervention effects The Spanish puzzle* Lara Reglero1 & Emma Ticio2 1Florida
State University / 2Syracuse University
Recent typological approaches (Cable, 2010; Pesetsky, 2000) explain the independently-noticed crosslinguistic differences between Superiority and Intervention Effects as the result of general lexical differences of wh-phrases. A pattern that seems to be recurrent crosslinguistically is that languages that display Superiority do not exhibit Intervention Effects and vice versa. In other words, Superiority and Intervention Effects appear to be in complementary distribution in natural languages. Surprisingly, Spanish lacks both Superiority and Intervention Effects. This paper accounts for this unexpected behavior in Spanish as the result of a lexical property of Spanish that allows some of its wh-phrases to enter the derivation as two dependent elements: a Q-particle and a dependent wh-variable. Keywords: Superiority Effects; Intervention Effects; wh-phrases; Q-particle; (un)selective binding
1. Introduction The fundamental premise of Minimalism allows us to accommodate intra- and crosslinguistic variation within a formal account of language (Adger & Smith, 2005). More concretely, the characterization of lexical items as bundles of features enables us to have a universally uniform grammatical component along with a language-specific lexicon, which allows for the account of variation facts that have not been possible to explain previously. Given this, it is not surprising that some well-known phenomena have been restated as typological properties, and that typological studies are common under Minimalism.
* Thanks are due to the Hispanic Linguistics Symposium audience at the University of Ottawa, and to two anonymous reviewers for their relevant questions and insights on this material. Errors are solely ours.
doi 10.1075/ihll.4.01reg © 2015 John Benjamins Publishing Company
Lara Reglero & Emma Ticio
In this paper, we examine the relation established between Superiority and Intervention Effects in a number of languages, with the goal of incorporating the behavior of Spanish into the current typological accounts. According to the descriptive generalizations put forward in the literature, languages that display Superiority do not exhibit Intervention Effects (Cable, 2010; Pesetsky, 2000). Both phenomena have been traditionally related to the properties of wh-phrases in different languages (see Beck (1996) for German, for example). More recently, these phenomena have been addressed from typological perspectives that straightforwardly explain the independently-noticed crosslinguistic differences as the result of lexical differences (Pesetsky, 2000; Cable, 2010). These typological approaches have considered unrelated and understudied languages, such as Tinglit or Navajo, but have not taken into consideration more common languages, such as Spanish, a language that lacks both Superiority and Intervention Effects. This paper is an attempt to evaluate and expand on these approaches by introducing the properties of Spanish wh-phrases into the typological discussion. The paper is organized as follows: in Section 2, we introduce the two main phenomena under examination: namely, Superiority and Intervention Effects. After a crosslinguistic discussion, we illustrate in detail the behavior of Spanish with regard to these phenomena. In order to clarify the properties of the constructions involved and the connection between them, in Section 3 we review previous attempts to connect Superiority and Intervention Effects (Cable, 2010; Pesetsky, 2000), and show their shortcomings when applied to the Spanish data. Focusing then on the Spanish properties, in Section 4 we introduce a new proposal to account for Superiority and Intervention Effects that retains the main elements of analyses argued for some other languages (such as English and German). Our proposal capitalizes on some language-specific properties of Spanish, such as its lack of wh-in-situ movement, to accommodate its behavior in the typology. In Section 5, we discuss the predictions of the analysis for some related constructions in Spanish that have not been previously addressed in the literature. Finally, in Section 6, we summarize the main findings of the paper.
2. Superiority and intervention effects 2.1 Crosslinguistic evidence Superiority Effects were first discussed in English, as illustrated in (1a–b): (1) a. Who bought what? b. *What did who buy? As argued by Chomsky (1973), when two wh-phrases take scope in the same CP, the superior (structurally higher) wh-phrase must move to Spec, CP. This is known as the Superiority Condition, as shown in (2):
No superiority, no intervention effects
(2) No rule can involve X, Y in the structure … X … [α … Z … -WYV …] Where the rule applies ambiguously to Z and Y and Z is superior to Y. Based on this approach to Superiority (1b) is ungrammatical in English because the structurally lower object ‘what’ moves over the structurally higher subject ‘who’. Note that (1a) is grammatical because the structurally higher subject ‘who’ moves before the structurally lower ‘what’, and therefore conforms to the Superiority Condition. Superiority Effects do not arise in languages such as German. As the grammaticality of the examples in (3a–b) illustrate, in German, the structurally higher wer can move over the structurally lower was (see (3a)), or the structurally lower was can move over the structurally higher wer, as in (3b): (3) a. Wer hat was gekauft? who has what bought ‘Who has bought what?’ b. Was hat wer gekauft? what has who bought ‘Who has bought what?’ However, as first noted by Beck (1996), Intervention Effects appear in languages such as German when some operators/quantifiers (typically, negation or Focus) linearly precede an in situ wh-phrase in those languages, as in (4): (4) *Wer hat nicht was gekauft? who has not what bought ‘Who didn’t buy what?’
(Beck, p.c.)
Beck’s pioneering work on Intervention Effects links them to the different properties of wh-movement crosslinguistically. More specifically, her proposal is based on the claim that Intervention Effects only affect covert wh-movement. Under Beck’s (1996) analysis, then, in situ expressions in German undergo movement at LF to a position above the operator. This covert movement leaves an LF trace, which is dominated by the Q while its restriction is not. This results in the illegitimate construction in (5):
(5) LF: *[…Xi…[ Q …[…tiLF…]]]
(Beck 1996: 1)
Even though Beck (1996) only discusses data from German, her treatment of Intervention Effects points to a typological approach as a way to explain crosslinguistic differences since covert movement is only possible in some languages. In later work based on insights by Pesetsky (2000), Beck (2006) notes that, typologically, a pattern that seems to be recurrent is that languages that display Superiority do not exhibit Intervention Effects. Thus, the English grammatical example in (6) is expected as the result of the different properties of English wh-phrases that enable
Lara Reglero & Emma Ticio
them to be placed in a position in which they are not c-commanded by a focusing or quantificational element.
(6) Who didn’t buy what?
In contrast, whenever a language is affected by Intervention Effects, it does not seem to conform to the Superiority Condition, as shown in the German examples in (3). This connection is observed in a wide variety of typologically unrelated languages (Beck, 2006; Cable, 2010; Pesetsky, 2000), such as English and German (as illustrated above), as well as Japanese, Korean, Tinglit, and Navajo.1 Neither the crosslinguistic studies on Superiority and Intervention effects mentioned above nor the broader syntactic studies on these phenomena have explored in detail the properties of these constructions in Spanish. Thus in the next section we introduce the corresponding structures in Spanish, and summarize the scarce discussion of these phenomena in the literature.
2.2 Superiority and intervention effects in Spanish To the best of our knowledge, Jaeggli (1982) was the first author to note the lack of Superiority Effects in Spanish, illustrated in the grammatical examples in (7a–b): (7) a. ¿Quién compró qué? who bought what ‘Who bought what?’ b. ¿Qué compró quién? what bought who ‘Who bought what?’ Jaeggli discusses the lack of Superiority Effects in languages such as Spanish or Italian as the result of the postverbal generation of the subject in these languages. According to this author, Superiority Effects in languages such as English are due to a violation of the Empty Category Principle (ECP). This ECP violation is caused by the LF trace of an LF movement rule which extracts the subject, left inside the sentence after wh-movement takes place, and attaches it to the Sentence (S’) node. The ECP is violated because the trace left in place by the subject cannot be properly identified (8b’), whereas the trace left by the object does not violate the ECP because it is identified by the V (8a’) (structures adapted from Jaeggli):
. In this paper, we focus mainly on Spanish, but we follow the descriptive generalization provided by Cable (2010), and Pesetsky (2000) for other languages. A more thorough crosslinguistic analysis of the phenomenon is beyond the scope of this paper.
No superiority, no intervention effects
(8) a. Who saw what? a’. [S’ what2 [S’ [COMP who1] [S [NP t1][VP [VP [V see] [NP t2]]]]]] b. *What did who see? b’. [S’ who1 [S’ [COMP what2] [S [NP t1][VP [VP [V see] [NP t2]]]]]]
[Dotted line: LF rule; solid line: wh-movement]
Assuming that Spanish and Italian subjects generate in the postverbal position, their traces do not violate the ECP when their traces are left behind after LF movement, since both t1 and t2 can be identified by the V. Therefore, Superiority Effects do not arise, as in (9). (9) a’. [S’ quién1 [S’ [COMP qué2] [S [NP PRO1][VP [VP [V compró] [NP t2]] [NP t1]]]]]
who
b’. [S’
qué2
[S’ [COMP
what quién1]
[S [NP
bought PRO1][VP
[VP [V compró] [NP t2]] [NP t1]]]]]
what
who
bought
Note that Jaeggli’s proposal does not account for the lack of Superiority Effects with adjuncts, as their traces violate the ECP but they do not trigger Superiority Effects in Spanish (see (10)): (10) a. ¿Dónde vive Juan con quién? where lives Juan with whom ‘Where does John live with whom?’ b. ¿Con quién vive Juan dónde? with whom lives Juan where ‘With whom does John live where? c. ¿Qué compró Juan cuándo? what bought Juan when ‘What did John buy when?’ d. ¿Cuándo compró Juan qué? when bought Juan what ‘When did John buy what?’ After Jaeggli (1982), there have been several mentions of the Spanish Superiority data in the literature (see for example Gallego, 2007; Reglero, 2007; Uribe-Etxebarria, 1992; Zubizarreta, 1998), but Bošković (1997) is the only author who provides a detailed
Lara Reglero & Emma Ticio
syntactic account for the lack of Superiority Effects in this language.2 More specifically, Bošković proposes an Economy account of Superiority and provides evidence for the existence of AgrO even in languages that lack overt object agreement. For Bošković, (7b) will have the derivation in (11): (11) ¿[CP Quéi comprój [AgrOP ti [VP quién [V’ tj ti]]]]? what bought who In (11), the subject quién is located in Spec VP. The object qué moves to Spec AgrOP before it moves to Spec CP. As the derivation shows, qué is higher that quién before wh-movement takes place. As Bošković points out, qué can now move to Spec CP without violating Superiority.3 Note that this line of analysis is based on the existence of an independently required movement that some argumental wh-phrases undergo prior to their movement to satisfy their wh-requirements in the CP position. Crucially, this line of analysis would not be able to explain the lack of Superiority Effects in cases such as (12), containing complements that in principle do not have to move to Spec AgrOP: (12) a. ¿Quién soñaba con qué? who dreamed with what ‘Who dreamed what?’ b. ¿Con qué soñaba quién? with what dreamed who ‘What did who dream?’ In the cases illustrated in (12), the lack of Superiority Effects cannot be explained as the result of the existence of a movement of the complement con qué prior to the whmovement, as qué receives its oblique Case from the P con. Based on this discussion, we can conclude that Bošković’s approach cannot account for all the data.
. Zubizarreta (1998) suggests that Superiority Effects can be derived from prosodic considerations. While prosodic factors may play a role, this is not an issue we will pursue further here. . Fitzpatrick (2002) discusses Bošković’s account in the context of minimalist approaches to the locality of movement. Based on Bošković’s proposal, Fitzpatrick argues that the object qué should be able to move to Spec CP of a higher clause. That is, (i) should be ungrammatical: (i) ¿Quéi dijo quién que Juan compra ti? what said who that Juan bought ‘What did who say that John bought?’ Fitzpatrick gives a * to this sentence. However, some native speakers find this sentence grammatical (Fitzpatrick himself acknowledges that the judgments are not clear). If (i) is grammatical, the problem for Bošković’s analysis no longer arises.
No superiority, no intervention effects
To the best of our knowledge, there is no detailed syntactic discussion of Intervention Effects in Spanish, although their absence has been occasionally documented in the wh-movement literature (Uribe-Etxebarria, 2002; Reglero, 2007). The grammatical examples in (13a–b) show that Intervention Effects do not arise in Spanish. In other words, the presence of an Operator, such as Negation, does not preclude the alleged covert wh-movement. (13) a. ¿Quién no compró qué? who neg bought what ‘Who didn’t buy what?’ b. ¿Qué no compró quién? what neg bought who ‘What did who not buy?’ Even though there is some discussion of the lack of Superiority Effects in Spanish, and a few mentions of the Intervention data, no attempt has been made to explain the lack of both Superiority and Intervention Effects in Spanish. This topic is crucial for the analyses of these phenomena crosslinguistically for two reasons. First, the absence of both phenomena in Spanish calls into question the apparent complementary distribution of Superiority and Intervention Effects in natural language and, as it will be shown in the next section, undermines the uniform analyses of these phenomena. Second, and more importantly, the lack of a uniform explanation for the absence of Superiority and Intervention Effects in Spanish excludes this language from the typological approaches intended to account for the properties of wh-phrases in natural language.
3. Uniform approaches to superiority and intervention effects There have been multiple approaches to derive the Superiority and Intervention Effects independently in the different languages where these phenomena have been attested. In this paper, we focus only on those approaches that aim to provide a uniform account of Superiority and Intervention Effects to propose a typological classification of languages with regards to their wh-properties. Taking as a point of departure Beck’s (1996) seminal work on Intervention Effects (which relates them to the movement of wh-phrases at LF, as discussed in Section 2.1.), authors such as Pesetsky (2000) and Cable (2010) have assumed a strong link between the two phenomena, which is stated in the universal descriptive generalization in (14). (14) The Complementarity of Superiority Effects and Intervention Effects In any language L, the in situ wh-words of a multiple wh-question of L are subject to Intervention Effects if and only if the multiple wh-questions of L are not subject to Superiority Effects. (Cable, 2010: 132; building on previous work by Beck, 1996 and Pesetsky, 2000)
Lara Reglero & Emma Ticio
In the following sections, we discuss Pesetsky’s (2000) and Cable’s (2010) proposals and highlight their shortcomings when applied to the Spanish paradigm.
3.1 Pesetsky (2000) Pesetsky’s (2000) approach is based on the assumption that, contrary to Beck’s (1996) analysis, what we know as Intervention Effects are not restrictions on a particular type of movement, but rather a semantic constraint. More specifically, he assumes that “a quantifier (including wh-) may not be separated from its variable by a scope-bearing element” (Pesetsky, 2000: 67), as illustrated in (15): (15) LF: *[… Xi …[ Z … [… ti …]]] Building on this assumption, Pesetsky makes a distinction between two types of movements, phrasal movement and feature movement. In the context of multiple questions (see examples in (16) and (17)), Pesetsky assumes that each type of movement obeys different movement constraints, Superiority being only applicable to phrasal movement. Pesetsky’s proposal is also based on the assumption that some languages have restrictions regarding the number of Specifiers in their CPs. Thus, some languages have the possibility of having more than one Specifier in the CP (Cm-spec), as English and Bulgarian, which allow for phrasal movement, whereas other languages, such as German for example, can only have a single Specifier in their CP (C1-spec), forcing feature movement. Given these assumptions, Pesetsky’s approach explains Superiority Effects in English as the result of the multiple-spec requirement displayed in the English Cmspec. The multiple-spec requirement forces both wh-phrases to undergo phrasal whmovement. Since both wh-phrases must move, the first instance of wh-movement to C is overt, (16b), and secondary instances of wh-phrase movement to C are also phrasal but covert, (16c). The relevant derivation for English is provided in (16) (steps and derivation adapted from Pesetsky). Note that given that phrasal movement is subject to Superiority, the only possible derivation is the one illustrated in (16), in which the subject must precede the object in its movement. (16) a. Structure before movement Cm-spec [who bought what] b. Step 1 Cm-spec attracts the wh-phrase who overtly Who [ _______ bought what] c. Step 2 Cm-spec attracts the wh-phrase what covertly Who [ _______ bought what] d. Pronounced result Who bought what?
No superiority, no intervention effects
As for German, Pesetsky assumes that, given the C1-spec in German, there is only one instance of phrasal wh-movement, which is overt, (17c). Any other instance of whmovement must involve feature movement, (17b), and is not subject to Superiority. Thus, the lack of Superiority Effects in German follows from the absence of more than one instance of phrasal wh-movement in this language. The relevant derivation for German is provided in (17) (steps and derivation from Pesetsky (2000: 75), but using the German example in (3b)): (17) a. Structure before movement C1-spec [wer hat was gekauft] who has what bought b. Step 1 C1-spec attracts the wh-feature of wer Fi-C1-spec [Fi-wer hat was gekauft] c. Step 2 C1-spec attracts the wh-phrase was was Fi-C1-spec [Fi-wer hat gekauft] d. Pronounced result Was hat wer gekauft? ‘What has who bought?’
As for Intervention Effects, Pesetsky resorts again to the difference between phrasal and non-phrasal movement, as only languages with non-phrasal movement will yield the LF conducive to Intervention Effects. This way, Intervention Effects are also seen as a diagnostic to distinguish wh-feature movement from phrasal wh-movement. In English (Cm-spec), the existence of phrasal wh-movement cannot trigger Intervention Effects, while in German (C1-spec) the feature movement of wh-in situ leads to Intervention Effects. As noted by many authors (Cable, 2010; Malhotra, 2011; Shields, 2008, among others), there are a number of empirical and conceptual problems with Pesetsky’s approach. For instance, Pesetsky’s approach has difficulties explaining reconstruction effects. As stated by Malhotra (2011), Pesetsky’s account would predict sentences such as (18) to be ungrammatical. In this example, the semantic restriction on the whelement (assumed to be in its base-generated position), would be separated from the wh-element by the scope-bearing element “everyone”.4 (18) Whose mother does everyone like? . This only becomes relevant if we assume that some type of reconstruction is taking place in these cases. Note that the role and properties of reconstruction has been under debate in the field (see Munn (1994) for an earlier discussion of reconstruction asymmetries within the Minimalist Program).
Lara Reglero & Emma Ticio
While these problems are real, in this paper we only focus on the impossibility of discussing feature movement under the current theoretical framework (see (17)), once we have adopted the Copy Theory of movement, currently standard in the Minimalist Program. More crucially, Pesetsky’s account is not empirically adequate, as it makes the wrong predictions for Spanish. Under Pesetsky’s analysis, Spanish should have a C with one Specifier (C1-spec) since there are no Superiority Effects. This assumption would predict Intervention Effects in Spanish, since the alleged LF movement of the wh-in situ in Spanish would trigger a representation that is intervened by the operator. This is clearly the wrong prediction, as the grammaticality of (13a–b) illustrates. In the next section, we summarize a more recent proposal that, grounding its analysis in Pesetsky’s (2000) approach, aims at overcoming some of the empirical issues that emerge from this type of analysis.
3.2 Cable (2010) For Cable, wh-questions are more intricate and consist of Force, a Q-particle, which is similar to any other operator, and the wh-phrase. Languages can vary depending on the requirements of each of these three elements. This way, the type of Force, the number of Q-particles and the relationship established between the Q-particle and the wh-phrase (agreement or not) will determine the properties of the wh-questions in a particular language. For English, Cable proposes that Force in this language requires establishing a relationship between multiple Q-particles associated with wh-phrases. Then, all Q-particles associated with wh-phrases must enter into a relationship with Force before the semantic component (SEM). This, along with the requirement that the Q-particle must agree with the wh-phrase, results in Superiority Effects in English. This is so because English Force attracts the closest Q-particle associated with the whphrase, which must move first. This requirement is fulfilled in the LF representation provided in (19) (FP stands for Force Phrase). In this case, ‘who’ moved before ‘what’. In contrast, the derivation crashes in (20), as ‘what’ moved before ‘who’. (19) [ FP [what-Q [who-Q [ Force-Q] [IP who [read what]]]]] (20) [ FP [who-Q [what-Q [ Force-Q] [IP who [[read what]]]]] In other words, according to Cable, English has a Force Q2, which is associated with multiple Q-particles. In this scenario, no Intervention Effects are expected because wh-phrases have been removed from the scope of the offending operator. The lack of Intervention Effects is illustrated in (21):
No superiority, no intervention effects
(21) [ FP [what-Q [who-Q [ Force-Q] [NEG [IP who [read what ]]]]]] The structure in (21) shows that both Q-particles have moved, along with their corresponding agreeing wh-phrases, to a position over Negation. This is a requirement imposed by the type of English Force, which must check its features against all the Q-particles present in the structure. Given the fact that both wh-phrases must move to the Force position, Intervention Effects are not possible since the wh-phrases move prior to SEM and no illicit configuration is created. To account for the different behavior attested in German, Cable (2010) assumes that the Force head only requires one Q-particle in this language (Force Q+). Based on this, he predicts that, in German multiple wh-questions, only one wh-phrase carries a Q-particle associated with it. This derives the lack of Superiority Effects because German Force can only attract one Q-particle associated with a wh-phrase, which is the only one available in the structure, and no more than one movement is possible. In (22), the Q-particle is associated with wer. As a consequence, this is the only element that is attracted by Force. This derives the wer before was order. In (23), the Q-particle is associated with was, which derives the was before wer order. Given the fact that one of the wh-phrases remains in a structurally lower position, Intervention Effects arise as another operator can bind the variable introduced by the wh-phrase, as shown in (24): (22) [FP [wer-Q] Force-Q hat [IP wer was gekauft ]] who has what bought (23) [FP [was-Q] Force-Q hat [IP wer was gekauft ]] what has who bought (24) [FP [wer-Q] Force-Q hat nicht [IP wer was gekauft]] who has not what bought Given that German has a Force Q+, which is associated with one Q-particle only, Intervention Effects are expected in this language because one of the wh-phrases is under the scope of the offending operator. Cable’s typological approach is able to accommodate a vast number of unrelated languages, and has proven useful to explain phenomena such as the existence of pied-piping in languages like English. This way, the amount of wh-related material displaced depends on the properties of the Q-particle selected in each particular language. More specifically, pied-piping depends on the need of some Q-particles to establish an agreement relationship with the wh-phrase being selected. Nevertheless, Cable’s approach cannot explain the properties of Spanish whphrases, since it predicts that Spanish would have one Force Q+ (similar to German), requiring the movement of only one Q-particle. This would account for Superiority, but it would predict that one of the wh-phrases will remain in situ under the scope of
Lara Reglero & Emma Ticio
the intervening operator. In turn, this element could bind the wh-phrase, leading to Intervention Effects. Again, this is the wrong prediction, as discussed above. In this section, we have reviewed some of the most prominent accounts of the main crosslinguistic differences regarding the behavior of wh-phrases in Superiority and Intervention Effect constructions. Assuming the previous accounts are intended to be applied universally, we would expect the typological schema to account for the overall properties of Spanish wh-phrases as well.
4. The proposal 4.1 Preliminary assumptions In order to accommodate the Spanish data and thus preserve the overall typological account, we propose our analysis that is based on three basic minimalist notions. First, linguistic variation is encoded in the lexicon. Second, no ‘ad hoc’ principles exist (only Merge and Move). From this perspective, Superiority must be a general principle (basically, an instantiation of the universal principle ‘Attract Closest’)5 but the lexical properties of languages can intervene to avoid a violation of this principle. Third, all syntactic derivations must fulfill the requirements of the interfaces. This way, although some authors have claimed that the different types of movement are the cause of ungrammaticality in Intervention Effects contexts (Beck, 1996), we propose that the types of movement are only relevant because they trigger a particular representation, and this representation is banned by SEM. This banned representation consists of an Operator and its variable separated by another Operator. Whenever that representation reaches SEM, the derivation is marked as ungrammatical, possibly as a general constraint against vacuous quantification à la Kratzer (1995).6 More specifically, these assumptions translate into the fact that Intervention Effects are ultimately representational constraints and that only languages displaying some types of non-phrasal wh-movement will display Intervention Effects, unless a lexical property of the language avoids that configuration.
. Attract Closest is a principle of grammar requiring that a head H that attracts a particular type of constituent must attract the closest constituent of the relevant type, which it c-commands (Radford, 2004). . This author’s Prohibition against vacuous quantification states the following: “For every quantifier Q, there must be a variable x such that Q binds an occurrence of x in both its restrictive clause and its nuclear scope” (Kratzer, 1995: 131).
No superiority, no intervention effects
4.2 Spanish The ultimate goal of our proposal is to preserve the link between Superiority and Intervention Effects established by previous analyses. This is motivated by the minimalist premise to account for the different crosslinguistic properties of wh-phrases as differences in the lexical properties of these elements across languages, rather than the result of various mechanisms that apply in different languages in narrow syntax. Obviously, the challenge now is how to account for the Spanish data. To explain the properties of Spanish, we turn to Reglero’s (2007) well-established generalization on wh-in situ: There is no wh-movement in wh-in situ constructions in Spanish. Reglero (2007) showed this extensively for the clausal domain with examples such as (25) where we see that the wh-phrase con quién is able to escape a syntactic island when it is in situ, as in (25a), but not when it has been fronted, as in (25b):7 (25) a. ¿Te has enamorado del hombre que vive con quién? you have fallen-in-love of-the man who lives with whom ‘You have fallen in love with the man who lives with whom?’ b. *¿Con quién te has enamorado del hombre que vive? with whom you have fallen-in-love of-the man who lives ‘*With whom have you fallen in love with the man who lives?’ The contrast between the two examples suggests that overt movement to Spec CP is only taking place in (25b). If overt movement also took place in (25a), as proposed by Uribe-Etxebarria (2002), the two examples in (25) should have the same grammatical status (that is, both sentences should be ungrammatical), contrary to fact.8
. Adjuncts such as dónde (where) and cuándo (when) are also allowed in situ in island contexts, as illustrated in (ia) with dónde: (i) a. ¿Te has enamorado del hombre que vive dónde? you have fallen-in-love of-the man who lives where ‘You have fallen in love with the man who lives where?’ b. *¿Dónde te has enamorado del hombre que vive? where you have fallen-in-love of-the man who lives ‘*Where have you fallen in love with the man who lives?’ . An anonymous reviewer wonders why we cannot assume that lack of island effects is due to the lack of overt movement. If we were to make this assumption, island effects in Japanese would be not be accounted for. As shown in (i) (example from Reglero & Ticio, 2013), Japanese exhibits island effects and no overt movement of the wh-phrase: (i) *Fumi-wa [Haruka-ga nani-o katta to] siritagatteimasu ka? Fumi-top [Haruka-nom what-acc bought C] know-want Q ‘What does Fumi want to know whether Haruka bought?’
Lara Reglero & Emma Ticio
Similarly, Reglero & Ticio (2013) also demonstrated that this property holds in the nominal domain. Their evidence is based on well-known properties of PP modifiers included in Spanish nominal expressions. As many authors have discussed (Ormazabal, 1991; Ticio, 2010; Torrego, 1987; among others), PPs within the Spanish nominal domain display blocking effects, with some PPs unable to move out of the nominal expression in the presence of certain PPs. These extraction restrictions have been linked to the existence of a thematic hierarchy (possessors>agents>objects) regulating movement within Spanish DPs. For instance, (26b) shows a well-documented extraction constraint, the ban of extraction of an object over an agent. Crucially, as illustrated in (26a), these blocking effects do not seem to be operative if the wh-phrase is in situ. (26) a. ¿Compraste un libro de Chomsky de qué? you-bought a book of Chomsky of what ‘You bought a book by Chomsky on what?’ b. *¿De qué compraste un libro de Chomsky? of what you-bought a book of Chomsky ‘*On what did you buy a book by Chomsky?’ In previous proposals, we have accounted for this difference in the behavior of the two types of wh-phrases in Spanish (i.e. fronted and in situ wh-phrases) as the result of lexical differences in the Force Phrase in Spanish. In this paper, we show that it is precisely this property of Spanish that is responsible for the absence of Superiority and Intervention Effects in this language. Therefore, a lexical difference will prevent the appearance of Superiority and Intervention Effects in Spanish, and enable us to accommodate the evidence from Spanish within the overall crosslinguistic paradigm. The gist of our proposal is that (some) wh-in situ phrases in Spanish do not move, and the reason is that they enter the derivation differently.9 More concretely, we are assuming that the traditionally-considered wh-in situ phrases in Spanish do not undergo any displacement (other than movement to a case-checking position). This differentiates them lexically from the wh-phrases that do show displacement properties (namely, fronted wh-phrases). Below, we show how this proposal can be implemented to account for the properties of Spanish. Given that all wh-phrase constructions under discussion trigger an interrogative meaning in Spanish, the assumption is that these constructions must contain the following universal components: a (silent) Q-particle, a Force head, and a wh-phrase
. D-linking affects the movement properties of wh-phrases in Spanish (as it does in English, see Pesestky (2000)).
No superiority, no intervention effects
(see Cable (2010), and Section 3.2. for additional discussion).10 In wh-fronting constructions, we assume that these three components appear as in (27a), where the Q-particle must be inserted at the point of the derivation in which the wh-phrase is entered in the construction. However, since we have evidence that Spanish wh-insitu phrases do not move (albeit their sentences are interpreted as true questions), we propose that they enter the derivation as a ‘discontinuous constituent’.11 Wh-in situ phrases in Spanish do not carry a Q-particle (i.e. driving force for movement), but enter the derivation as a wh-variable (represented in (27b) as an NP carrying a wh-interpretable feature). The structures of the two different wh-phrases in Spanish are given in (27):12 (27) a. [QP Q[uwh][NP ]] (wh-fronting phrases) b. [NP [iwh-]] (wh-in situ phrases) To secure the interrogative meaning of the wh-in situ constructions, a (silent) Q-particle must be introduced as a Last Resort, represented as QLex below, which is introduced when the derivation is in need of a binder for a wh-variable: (28) [QP q[uwh]] QLex carries an uninterpretable feature that probes the wh-interpretable feature present in the variable inserted in wh-in situ constructions in Spanish. Note that lack of insertion of QLex will result in the ungrammaticality of the sentence containing a wh-in situ phrase, since the variable introduced in the wh-in situ element needs to be bound to prevent it from being interpreted as a free variable. Crucially, an existential quantifier cannot bind the variable, as it must be selectively bound (see additional
. Echo questions, which do not carry interrogative meaning, exhibit different properties, as discussed extensively in Reglero & Ticio (2013). . Here we use the term ‘discontinuous constituent’ to refer to the components of whphrases (Q, wh-phrase) that are introduced via different merge operations in Spanish wh-in situ phrases. We understand that these components still belong to the same constituent since they must establish a dependency through the derivation. However, given their different generation sites, there is material appearing between them. Therefore, they do not have a continuous constituent status. . This avoids Pesetsky’s (2000) difficulties in explaining reconstruction effects. As discussed in Section 3.1, Pesetsky’s account predicts sentences such as (i) to be ungrammatical. In this example, the semantic restriction on the wh-element (assumed to be in its base-generated position), would be separated from the wh-element by the scope-bearing element ‘everyone’ under reconstruction approaches.
(i)
Whose mother does everyone like?
Note that this problem does not arise in (27b) given that the nature of the variable is different.
Lara Reglero & Emma Ticio
discussion on Intervention Effects below). The relevant derivations for Spanish wh-in situ constructions such as (29) are provided in (30).13 (29) ¿Juan compró qué? Juan bought what ‘John bought what?’ (30) a. [FP [ Juan [Force-Q] compró [IP Juan qué-[iwh] compró qué-[iwh]]]] b. [FP QLex [Juan [Force-Q] compró [IP Juan qué-[iwh] compró qué-[iwh]]]] Example (30a) shows the structure before QLex is inserted (the variable is not bound), and (30b) illustrates the derivation after QLex is inserted (the variable is selectively bound by the Q-particle). This proposal allows us to explain the true interrogative nature of wh-in situ questions in Spanish, along with the lack of Superiority and Intervention Effects. In addition, this analysis preserves the ban of movement exhibited by these constructions, and is consistent with the rightmost linear position of wh-in situ in Spanish (Reglero & Ticio (2013), and Footnote 13 below). The difference between Spanish and German is then that Spanish wh-in situ phrases are selectively bound. In minimalist terms this means that they carry an interpretable wh-feature that only a particular type of operator (namely, the operator lexically inserted in Force, QLex) can bind. Note that this makes German Force also different from Spanish Force. While the Force in German requires just one Q-particle, the Force in Spanish needs checking of all particles, but Spanish allows for their direct insertion at the Force Phrase (unlike English) as a last resort mechanism and only when they are silent.
. In situ wh-phrases in Spanish are subject to the Sentence Final Requirement (UribeEtxebarria, 2002); that is, they need to appear in final position, as illustrated by the ungrammaticality of (i) below, and the grammaticality of (29) and (ii) below: (i) *¿Juan compró qué el otro día? Juan bought what the other day ‘John bought what the other day?’ (ii) ¿Juan compró qué # el otro día? (# stands for pause) Juan bought what # the other day ‘John bought what # the other day?’ See Reglero (2007), and Reglero & Ticio (2013) for a syntax-phonology approach to the phenomenon, where it is argued that in situ wh-phrases in Spanish need to appear last to receive nuclear stress.
No superiority, no intervention effects
The lack of Superiority Effects is thus easily explained assuming that only one of the two Spanish wh-phrases carries a Q-particle. The Spanish Force probes the only wh-phrase that carries the Q-particle (regardless of its structural position), and attracts it. After that, and in order to avoid a derivation crash, the silent Q-particle is inserted in the upper position. Note that the other wh-phrase cannot be probed because Force probes for Q-particles, not for NPs. The relevant derivations are provided in (31) and (32): (31) [FP QLex [quién–Q [Force-Q] compró [IP quién compró qué-[iwh]]]]] who bought what (32) [FP QLex [qué-Q [Force-Q] compró [IP quién-[iwh] compró qué]]] what bought who As for the lack of Intervention Effects, we have proposed that they are the result of a configuration that is illicit in SEM, in which the variable is bound unselectively by an intervening operator. In the case of Spanish, we insert an operator that searches for a particular variable: the one carrying an interpretable wh-feature that cannot be bound by any other element. We are essentially coindexing the two items, thus preventing the intervening operator from binding the variable since it is specified for another operator and cannot receive unselective binding. Our proposal is in the spirit of proposals that have been put forward for Chinese wh-phrases in situ, such as the one by Aoun & Li (1993).14 The derivations for the Intervention data are provided in (33) and (34): (33) [FP QLex [quién–Q [Force-Q] [no compró [IP quién compró qué [iwh]]]]] who neg bought what (34) [FP QLex [qué–Q [Force-Q] [no compró [IP quién-[iwh] compró qué]]]] what neg bought who Assuming Cable’s (2010) typological proposal (see Section 3.2), the difference between Spanish and German is that Spanish wh-in situ are selectively bound, being coindexed or carrying an interpretable wh-feature that only the operator lexically inserted in Force can bind. Note that this also differentiates German Force from Spanish Force. The Force in German is satisfied with just one Q-particle, as illustrated in the structure
. Note that Aoun and Li’s proposal is based on the idea that Chinese wh-phrases are free variables, which is precisely the opposite of what we are claiming for Spanish. From this perspective, Aoun & Li’s hypothesis establishes an interesting parallelism between Chinese and German regarding the type of behavior expected with Intervention Effects in these languages. Both of them enter variables that can be freely bound by different operators and, as a consequence, Intervention Effects may arise.
Lara Reglero & Emma Ticio
in (35), while the Force in Spanish requires checking of all particles, just as with English Force. (35) [FP [wer-Q] Force-Q hat nicht [IP wer was gekauft]] who has not what bought The difference between English and Spanish is that Spanish allows the possibility to insert Q-particles directly at the Force Phrase, as a last resort mechanism. In this way, Spanish differs from English and German, but still adheres to the overall constraints applying to all languages.
5. Additional evidence: Separation structures in Spanish In the previous section, we proposed that there are different types of Spanish whphrases, some of them generated in the lexicon and others resulting from the syntactic derivation. A natural consequence of this proposal is that the different lexical choices should be evident when they interact with some of the other elements in the derivation. In this section, we explore some of those predictions and illustrate that they are borne out in the case of some previously unnoticed constructions in Spanish. We have proposed that the selective or unselective nature of wh-in situ phrases in Spanish is derived from the generation point of the Q-particle in Spanish. This leads to the prediction that whenever we have the Q-particle inserted along with the wh-phrase (i.e. when the wh-phrase is not selectively bound), any separation of the Q-particle from the wh-phrase during the derivation will trigger an Intervention Effect in Spanish. The literature on Intervention Effects has actually considered cases in which the members of the wh-phrase are apparently separated in these constructions. Consider the German examples in (36), taken from Pesestky (2000): (36) a. Wen von den Musikern hat Hans getroffen? whom of the musicians has Hans met ‘Who among the musicians has Hans met?’ b. Wen hat Hans von der Musikern getroffen? whom has Hans of the musicians met c. Wen von der Musikern hat keine Studentin getroffen? who of the musicians has no student met ‘Who among the musicians has no student met?’ d. ??Wen hat keine Studentin von der Musikern getroffen? whom has no student of the musicians met As discussed in Section 2.1, German displays Intervention Effects (as in (4)), which were originally explained in the literature as the result of the particular properties of covert wh-movement. The paradigm in (36c) and (36d) shows that the overt movement
No superiority, no intervention effects
of part of the wh-phrase also produces ungrammaticality in these languages, as illustrated by the ungrammaticality of (36d). These constructions, which display a similar pattern in other languages, such as French (as in (37), also from Pesetsky (2000)), are characterized by the overt movement of the wh-operator, which moves overtly to a higher position. These constructions are termed ‘separation structures’: (37) a. Combien de véhicules a-t-il conduit? how-many of cars did-he drive ‘How many cars did he drive?’ b. Combien a-t-il conduit de véhicules? how-many did-he drive of cars ‘How many cars did he drive?’ c. Combien de véhicules n’a-t-il pas conduit? how-many of cars did-he not drive ‘How many cars didn’t he drive?’ d. *Combien n’a-t-il pas conduit de véhicules? how-many did-he not drive of cars ‘*How many did he not drive cars?’ Separation structures are therefore ideal for testing our proposal for Spanish. More specifically, they allow us to test the prediction that Intervention Effects will occur anytime that the Q-particle inserted as part of the wh-element is moved overtly. This is so because we have to assume that there is no QLex in those cases. Otherwise, the insertion of a QLex at Force would trigger a violation of the general ban on vacuous quantification, as the variable entered along with the displaced Q-particle is already bound by this displaced Q-particle, and QLex will not be able to bind anything. Although separation structures are not widely productive in Spanish, there are some cases attested. Crucially, in these cases our prediction is borne out. This is exactly what happens in (38), where the difference in grammaticality between (38b) and (38d) shows an Intervention Effect: (38) a. ¿Cómo de caliente quieres la sopa? how of hot you-want the soup ‘How hot do you want the soup?’ b. ¿Cómo quieres la sopa de caliente? how you-want the soup of hot ‘How hot do you want the soup?’ c. ¿Cómo de caliente no quieres la sopa? how of hot neg you-want the soup ‘How hot do you not want the soup?’ d. *¿Cómo no quieres la sopa de caliente? how neg you-want the soup of hot ‘*How do you not want the soup hot?’
Lara Reglero & Emma Ticio
The derivation for the data in (38), illustrated in (39a–b) (for (38c–d)), generates cómo de caliente as a continuous constituent (i.e. Q and its variable are merged to each other and entered in the main derivation as a unit), and enables the Q element to move overtly (i.e. in narrow syntax) to Force without pied-piping the movement of de caliente, which remains in the base generation position. Given that Negation is placed in a lower position than Force in Spanish, it interferes with the relation between the displaced Q element and its variable, which explains the Intervention Effect in (38d). (39)
a. [FP [QP Cómo-Q (de) caliente] [Force-Q] [no quieres [IP (tú) [VP quieres [la sopa [QP cómo-Q (de) caliente]]]] b. [FP [QP Cómo-Q] [Force-Q] [no quieres [IP (tú) [VP quieres [la sopa [QP cómo-Q (de) caliente]]]]
Therefore, our approach to these phenomena also allows us to accommodate different grammaticality possibilities with respect to Superiority and Intervention Effects within the same language. This is so because the requirements of different constructions can favor a particular grammatical status over the other after lexical choices are made. Along the same lines, further research on the syntax of Spanish D-linked constructions (Reglero & Ticio, in progress) can shed light on our predictions and on the particular properties that the relevant literature (Beck, 2006; Cable, 2010; Pesetsky, 2000; Shields, 2008) has repeatedly discussed for these D-linked constructions. Ultimately, the unexpected behavior of some constructions against the overall behavior of the language with respect to Superiority and Intervention Effects must be the consequence of the separation of an Op-variable unit during the derivation.
6. Conclusions We have argued that lexical differences can explain the different crosslinguistic behavior regarding Superiority and Intervention Effects. Under this approach, Superiority is reduced to Attract closest, and Intervention is reduced to an interface condition. The lack of a complementary distribution between Superiority and Intervention Effects in Spanish is accounted for as the result of a lexical property of Spanish that allows some of its wh-phrases to enter the derivation as two dependent elements: a Q-particle and a dependent wh-variable.
References Adger, D., & Smith, J. (2005). Variation and the minimalist program. In L. Cornips & K. P. Corrigan (Eds.), Syntax and variation. Reconciling the biological and the social (pp. 149–178). Amsterdam: John Benjamins. DOI: 10.1075/cilt.265.10adg Aoun, J., & Li, Y.A. (1993). WH-elements in situ: Syntax or LF? Linguistic Inquiry, 24, 199–238.
No superiority, no intervention effects
Beck, S. (1996). Quantified structures as barriers for LF movement. Natural Language Semantics, 4, 1–56. DOI: 10.1007/bf00263536 Beck, S. (2006). Intervention effects follow from focus interpretation. Natural Language Semantics, 14, 1–56. DOI: 10.1007/s11050-005-4532-y Bošković, Ž. (1997). On certain violations of the Superiority Condition, AgrO, and economy of derivation. Journal of Linguistics, 33, 227–254. DOI: 10.1017/s0022226797006476 Cable, S. (2010). The grammar of Q: Q-particles, wh-movement and pied-piping. Oxford: Oxford University Press. DOI: 10.1093/acprof:oso/9780195392265.003.0004 Chomsky, N. (1973). Conditions on transformations. In S. Anderson, & P. Kiparsky (Eds.), A festschrift for Morris Halle (pp. 232–286). New York, NY: Holt, Reinhart & Winston. Fitzpatrick, J. (2002). On Minimalist approaches to the locality of movement. Linguistic Inquiry, 33 (3), 443–463. DOI: 10.1162/002438902760168563 Gallego, A. (2007). Phase theory and parametric variation. Unpublished Doctoral Dissertation. Universitat Autònoma de Barcelona. Jaeggli, O. (1982). Topics in Romance syntax. Dordrecht: Foris. DOI: 10.1075/li.9.1.11roo Kratzer, A. (1995). Stage-level and individual-level predicates as inherent generics. In G. N. Carlson & F. J. Pelletier (Eds.), The generic book (pp. 125–75). Chicago, IL: University of Chicago Press. Malhotra, S. (2011). Movement and Intervention Effects: Evidence from Hindi/Urdu. Unpublished Doctoral Dissertation. University of Maryland. Munn, A. (1994). A Minimalist account of reconstruction asymmetries. In M. Gonzalez (Ed.), Proceedings of NELS 24 (pp. 397–410). Amherst, MA: GLSA. Ormazabal, J. (1991). Asymmetries on wh-movement and some theoretical consequences. Ms., University of Connecticut: Storrs, CT. Pesetsky, D. (2000). Phrasal movement and its kin. Cambridge, MA: The MIT Press. DOI: 10.1017/s0022226702281980 Radford, A. (2004). Minimalist syntax: Exploring the structure of English. Cambridge: Cambridge University Press. DOI: 10.1017/s0022226706003963 Reglero, L. (2007). Wh-in-situ interrogatives in Spanish. Probus, 19(2), 267–297. DOI: 10.1515/probus.2007.009 Reglero, L., & Ticio, M.E. (2013). A unified analysis of wh-in-situ in Spanish. The Linguistic Review, 30(4), 1–47. DOI: 10.1515/tlr-2013-0018 Reglero, L., & Ticio, M.E. (in progress). D-linking in Spanish: A movement analysis. Ms. Florida State University, Tallahassee, FL, and Syracuse University, Syracuse, NY. Shields, R. (2008). Intervention and rescue: Arguments for representational constraints on syntactic dependencies. Unpublished Doctoral Dissertation. University of Wisconsin-Madison. Ticio, M.E. (2010). Locality domains in the Spanish determiner phrase. New York, NY: Springer. DOI: 10.1007/978-90-481-3398-7 Torrego, E. (1987). On empty categories in nominals. Ms. University of Massachusetts, Boston. Uribe-Etxebarria, M. (1992). On the nature of Spec/IP and its relevance for scope asymmetries in Spanish and English. In J. Amastae, G. Goodall, M. Montalbetti, & M. Phinney (Eds.), Contemporary research in Romance linguistics (pp. 355–366). Amsterdam: John Benjamins. Uribe-Etxebarria, M. (2002). In situ questions and masked movement. Linguistic Variation Yearbook, 2, 259–303. DOI: 10.1075/livy.2.09uri Zubizarreta, M.L. (1998). Prosody, focus and word order. Cambridge, MA: The MIT Press.
Overt PRO in Romance Towards a unification of PRO and pro* Peter Herbeck
University of Salzburg Several types of infinitives in Spanish, Italian, and Catalan can have overt nominative subjects. These elements are morphologically pronominal although they are syntactically anaphoric in Obligatory Control contexts and pronominal in Nonobligatory Control contexts. The data will be taken as evidence for a unification of PRO and pro to an externally merged D head with fully variable phi-features (in the spirit of Sigurðsson’s (2008) notion of ‘reference variable’), whose status with respect to Binding Theory is determined by interpretable features on T, implementing a version of Borer’s (1989) and Landau’s (2000, 2004) anaphoric AGR. In this approach, overt subjects in infinitives are the result of post-syntactic realization of D[φ:_], triggered by discourse-related rather than Case-related needs. Keywords: control; infinitives; overt subjects; pro-drop; Romance languages
1. Introduction A traditional diagnostic for the difference between the two types of null subjects PRO and pro has been the obligatory emptiness of the former and the optional overt realization of the latter. However, Suñer (1986), Piera (1987), Fernández-Lagunilla (1987), Hernanz (1999), Mensching (2000), and Pöll (2007), among others, have shown that overt subjects are licit in non-finite control configurations in Spanish, even though the verb is not equipped with overt agreement markings:
* I would like to thank Luis López for helpful discussions during my time at the University of Wuppertal. I also thank the members of the Centre de Lingüística Teòrica (CLT; Universitat Autònoma de Barcelona) for discussions during my stay as a visiting scholar – especially Gemma Rigau for her time and patience. My stay at the CLT was supported by a grant from the DAAD. Furthermore, I am grateful to Idan Landau, Bernhard Pöll, Valentina Bianchi and two anonymous reviewers for comments. All errors are fully my own.
doi 10.1075/ihll.4.02her © 2015 John Benjamins Publishing Company
Peter Herbeck
(1) Juliai quería [telefonear (ellai/*x)].(Piera, 1987: 160) Julia wanted-3sg phone.inf she ‘Julia wanted to phone herself.’1 (2) [Salir (túx)] fue un error. go.out.inf you was an error ‘It was a mistake that you went out.’
(Piera, 1987: 165)
One possible analysis considers these elements as genuine instances of ‘overt PRO’ (see Cardinaletti, 1999; Mensching, 2000: 61f; Alonso-Ovalle & D’Introno, 2001; Barbosa, 2009a; Szabolcsi, 2009; Livitz, 2011). In this paper, I argue that overt subjects in Romance infinitives rely on an analogical strategy to the overt/covert alternation in finite pro-drop clauses. This reasoning implies that the subject position is of the same general type at the level of External Merge in both configurations. In the vein of Borer (1989), the differences between PRO and pro with respect to Binding Theory are determined by the properties of the T (AGR) head an empty category (ec) is associated with. I argue that the overt/covert alternation in Romance infinitives is the result of an interaction between post-syntactic Spell-Out rules (adopting late insertion as in Distributed Morphology (Halle & Marantz, 1993)) and discourse-sensitive information at the vP phase edge. This way, overt subjects in Romance infinitives are what Alonso-Ovalle & D’Introno (2001) analyze as focused ‘zero pronouns’ (in the sense of Kratzer, 1998). This paper is structured as follows: first, I discuss overt subjects in Spanish, Italian, and Catalan infinitives. Then I present the theoretical background for unifying the two ecs in control and pro-drop. Thereafter, I discuss the overt/covert alternation in finite and non-finite domains. Finally, I focus on mismatches between pronoun and lexical DP licensing.
2. Overt PRO in Romance In this section, I discuss overt subjects in obligatory control (OC) and non-obligatory control (NOC) infinitives and will show that both should receive an analysis in terms of ‘pronounced PRO’.
2.1 Emphatic pronouns are real subjects in OC infinitives As we have seen in (1), emphatic pronouns are licit in Spanish OC structures (see for instance Piera, 1987; Cardinaletti, 1999 for Italian; and Solà, 1992 for Catalan): . The glosses and translations of several examples have been added or modified to make them fit the layout of this paper.
Overt PRO in Romance
(3) En Joani prefereix [de fer-ho elli (mateix)].(Catalan; Solà, 1992: 192) the John prefers to do.inf-it he (self) ‘John prefers to do it himself.’ The addition of a self-element is not required to license anaphoricity of the morphologically pronominal element. Piera (1987) and Pöll (2007) propose an analysis that is compatible with the obligatory emptiness of PRO as required by the PRO-Theorem (Chomky, 1981), or Null Case (Chomsky, 1995): emphatic pronouns are adjoined to VP and, hence, ‘doubles’ of the real thematic subject PRO.2 This analysis draws a parallel to the following finite structure in which the postverbal pronoun doubles a preverbal subject DP: (4) Juliai telefoneó ellai.(Spanish; Piera, 1987: 161) Julia phone.past.3sg she ‘Julia phoned herself.’ However, there is evidence that an adjunction analysis of emphatic pronouns is on the wrong track: first, as Solà (1992) observes for Catalan, semantically-close adverbs have a different distribution than emphatic pronouns in that the latter, but not the former, are incompatible with inverted subjects and objects: (5) Ho ha fet en Joani (* elli / en person). it has done the Joan he in person ‘John has done it (*he/in person).’
(Catalan; Solà, 1992: 59)
This indicates that emphatic pronouns are in complementary distribution with inverted subjects in their base-position. Barbosa (2009a) further notes that European Portuguese applies a separate element (ele próprio) to encode these relations. Emphatic pronouns exclusively relate to preverbal subjects. (6) Falei com o director (* ELE / ele próprio) talk.past-1sg with the director HE himself ‘I talked to the director himself.’ (Portuguese; Barbosa, 2009a: 108) As is well-known, in languages like English and German, emphatic anaphors can refer to (inverted) subjects as well as objects and hence are genuinely adjoined elements: (7) Gestern hat Hansi mit dem Direktorj selbsti/j gesprochen. yesterday has Hans with the director self talked ‘Yesterday Hans talked to the director himself.’ Thus, emphatic pronouns in Romance seem to constitute real instances of ‘pronounced PRO/pro’ in its base-position (see Cardinaletti, 1999; Alonso-Ovalle & D’Introno, 2001;
. See also Sánchez (1993), who analyzes emphatic pronouns as non-arguments.
Peter Herbeck
Mensching, 2000; Barbosa, 2009a; Szabolcsi, 2009). Such an analysis can explain why emphatic pronouns reflect the referential properties of a controlled element (see for instance Mensching, 2000): (8) a. Juliai prometió a Martaj [encargarse ellai/*j del asunto]. Julia promised to Marta take-care.inf she of.the matter ‘Julia promised Marta that she would take care of the matter herself.’ b. Juliai animó a Martaj [a encargarse ella*i/j Julia encouraged to Marta to take-care.inf she del asunto]. of.the matter ‘Julia encouraged Marta to take care of the matter herself.’ (Spanish; Piera, 1987: 160) In the next section, I discuss two potential analyses of the licensing of overt PRO and will show that they both suffer from important drawbacks.
2.1.1 Emphatic pronouns are not licensed by long-distance agree Spanish has been argued to license nominative Case assignment under government, allowing a postverbal subject to remain in situ (see Contreras, 1991; Ordóñez, 1998; Alexiadou & Anagnostopoulou, 1998):
(9) [TP T[NOM]-v-V [vP Subject[NOM] v-V [VP V]]].
Since non-finite control complements are phi- and/or tense deficient (Landau, 2004; Boeckx et al., 2010), they do not constitute strong phases and Case could be argued to be assigned long distance: (10) [TP Julia T[NOM]-quería [XP Xdef [TP T-telefoneari [vP ella v-ti]]]]. Julia
wanted
phone.inf she
Multiple Agree has been proposed for overt subjects in Hungarian OC infinitives by Szabolcsi (2009). Crucial evidence comes from the observation that nominative subjects are blocked from object control structures: (11) *Kényszerítettek (téged) [te is dolgozni]. forced.3pl you.sg.acc you.sg.nom too work-inf (Hungarian; Szabolcsi 2009: 18) In object control configurations, the matrix object blocks nominative Case transmission to the subject position of the embedded infinitive. However, we have already seen in (8) that emphatic nominative pronouns are licit in the complement of object control verbs in Spanish. The same holds for Italian:
Overt PRO in Romance
(12) Maria mii ha chiesto di parlare ioi con Gianni. Maria me has asked to speak.inf I with Gianni ‘Maria asked me that I myself speak with Gianni.’ (Italian; Belletti, 2005: 21) Thus, nominative Case transmission in terms of long distance Agree cannot be responsible for the licensing of overt subjects in Spanish and Italian control infinitives.
2.1.2 Emphatic pronouns are not lower pronounced copies Polinsky & Potsdam (2002) claim that subjects can be fully realized DPs inside control infinitives in a variety of languages if the matrix controller remains empty – the phenomenon of Backward Control (BC). In the Movement Theory of Control (Hornstein, 1999), BC can be explained by means of postulating movement of the subject DP out of the embedded infinitive followed by subsequent deletion of the higher copies (see for example Boeckx et al., 2010). Emphatic pronouns could thus constitute instances of lower default pronounced copies: (13) a. Juani quería [Juani telefonear Juani]. (OC) b. Juani quería [Juani telefonear éli]. (emphatic pronoun)
John wanted
to phone.inf him
However, although Burzio (1986) discusses certain parallels between emphatic pronouns and NP-traces, the former do not exactly reflect the distribution of the latter (Burzio, 1986; Cardinaletti, 1999): (14) Giovannii sembrava [{ti/*luii} conoscere la strada]. Giovanni seemed.3sg he know.inf the way ‘Giovanni seemed (*he) to know the way.’ (Italian; Burzio, 1986: 116; Cardinaletti, 1999: 62) Furthermore, as Szabolcsi (2009) points out, the expected case under a BC analysis would be that full R-expressions and DPs are possible. However, full DPs are blocked in Italian OC infinitives in various contexts (see (15); examples from Belletti, 2005: 20ff.): (15) a. proi penso di parlare ioi di questo problema. (I) think to speak.inf I of this problem ‘I think that I myself will speak about this problem.’ b. *proi pensa di parlare Giannii di questo problema. (he/she) thinks to speak.inf Gianni of this problem In Spanish, full DPs are also blocked in various OC contexts in which emphatic pronouns are licit.3 This is the case with, for example, intervening matrix objects, which . Alexiadou et al. (2010) argue that Backward Control exists in Spanish. However, there are reasons to believe that this phenomenon is only apparent in Spanish and is the result
Peter Herbeck
cause ungrammaticality of BC (see (16) from Ordóñez, 2009: 4) but not of emphatic pronouns (see (8)). (16) *Les prometió a los familiares [darles el jurado CL promised-3sg to the relatives give.inf-them the jury la libertad a los prisioneros]. the freedom to the prisoners Note that even those configurations that allow pronouns as well as R-expressions inside complement infinitives have different semantic properties in Spanish and Italian if the respective subject is associated with a scope-taking particle (Szabolcsi, 2009: 32): (17) No quiere ir sólo Juan a la escuela. (Szabolcsi, 2009: 32) not wants go.inf only Juan to the school Reading1: Only Juan doesn’t want to go to school. *Reading2: Hei doesn’t want it to be the case that only Juani goes to school. (18) No quiere ir sólo él a la escuela. (Szabolcsi, 2009: 32) not wants go.inf only he to the school Reading1: Only he doesn’t want to go to school. Reading2: Hei doesn’t want it to be the case that only hei goes to school. The unavailability of an embedded scope reading with an R-expression in (17), but its availability with a pronoun in (18), indicates that we are dealing with an embedded subject in the latter but a matrix subject in the former case.4
of scrambling and verbal complex formation (see Ordóñez, 2009 and Herbeck, 2014 for discussion). . A reviewer notes that there is another possible analysis within a movement approach to control: adopting Kayne (2002), the matrix antecedent and emphatic pronoun could be initially doubled as in Postal (1969). The antecedent moves into the matrix clause, leaving the emphatic pronoun stranded: (i) En Joani prefereix [de fer-ho [ti ell]] the Joan prefers to do.inf-it he ‘John prefers to do it himself.’ A variant of such an approach has been proposed by Torrego (1996) in parallel to floating quantifiers. According to a reviewer, the impossibility of epithets supports an analysis along the lines of Postal (1969) and Kayne (2002): (ii) *Juan quiere telefonear el tonto al director. John wants phone.inf the stupid to-the director However, Barbosa (2009a) provides appealing evidence against a doubling approach: e.g. it predicts, contrary to fact, that non-referential quantifiers are licit antecedents of emphatic
Overt PRO in Romance
2.2 Overt subjects in NOC infinitives The distribution of overt subjects in NOC infinitives is an intricate issue (see Mensching, 2000; Pérez Vázquez, 2007). In this section, I highlight only two aspects of the distribution of overt subjects which indicate that it is not exclusively Case theory, but also discourse-sensitive information that is crucial for their licensing.
2.2.1 Pronouns versus R-expressions Common to all three of the languages under consideration is the fact that pronouns underlie less-strict licensing conditions than lexical DPs. Thus, Piera (1987) observes for Spanish that substituting a pronoun for an R-expression leads to degradation in some types of infinitives: (19) Salir (tú/ ??Julia) fue un error. go.out.inf you Julia was an error ‘The fact that (you/??Julia) went out was a mistake.’
(Piera 1987: 156)
A similar situation has been observed for Italian by Burzio (1986) and C ardinaletti (1999): (20) [[Andarci (noi/*Giovanni)] sarebbe un errore]. go.inf-there (we/*Giovanni) would-be.3sg a mistake’. ‘For (us /*Giovanni) to go there would be a mistake.’ (Italian; Burzio, 1986; Cardinaletti, 1999) Even though full DPs and R-expressions are possible in certain subject infinitives in Spanish (see Piera, 1987; Mensching, 2000; Pérez Vázquez, 2007), the question arises as to why pronouns should be more freely licensed in certain configurations. In the case of adjunct infinitives introduced by a preposition, on the other hand, lexical DPs are productive in Spanish (see Rigau, 1995; Torrego, 1998): (21) Después de actuar Caballé, cantó Carreras. after of perform.inf Caballé sing.past-3sg Carreras ‘After Caballé performed, Carreras sang.’ (Spanish; Rigau, 1995: 280) However, in preverbal position, Pöll (2007) observes that full R-expressions are degraded in Peninsular Spanish:5
pronouns. Furthermore, Sánchez-López (1996) notes some distributional differences between emphatic pronouns and floating quantifiers. . See Suñer (1986) for a discussion of Caribbean Spanish dialects, where preverbal subjects are more productive.
Peter Herbeck
(22) Antes de (yo/ *Juan) nacer, …6 before of I / Juan be-born.inf ‘Before (I/*Juan) was born’
(Spanish; Pöll, 2007: 107f)
Thus, it seems that pronouns, but not full DPs, can escape the Case Filter (Cardinaletti, 1999; Cardinaletti & Starke, 1999). In the next section, I argue that they are licensed by discourse-sensitive features, rather than structural Case.
2.3 The restriction of overt subjects to the postverbal position It has been frequently observed that overt subjects in infinitives are restricted to the postverbal position (examples from Piera, 1987: 159f.):7 (23) [(*Tú) telefonear (tú) primero] sería un error. you phone.inf you first be.cond-3sg an error ‘That you phone first would be a mistake.’ (24) *Julia quería [ella telefonear]. Julia wanted she phone.inf ‘Julia wanted to phone herself.’ Piera (1987) suggests that the impossibility of preverbal subjects follows from the presence of PRO in Spec,T, which blocks other elements in this position. However, this reasoning still does not explain why emphatic pronouns cannot appear in a higher, C-related position: (25) *Julia quería [CP ella [TP PRO telefonear]]. Julia wanted she phone.inf If emphatic pronouns constitute instances of ‘pronounced PRO’, the unavailability of preverbal subjects follows from the defective left periphery of infinitives. The following examples show that they block Clitic Left Dislocation (CLLD) and Focus Fronting (FF) in Spanish (see also Hernanz, 2011):8
. It should be noted that some Spanish speakers generally consider preverbal subjects in this type of adjunct infinitives marginal, independent of whether the subject is a pronoun or an R-expression: (i) Antes de (?? yo) nacer,… before of I be-born.inf ‘Before I was born…’ . An exception is the preverbal position in adjunct infinitives, as in (22) (see Suñer, 1986; Mensching, 2000; Pöll, 2007). . Rizzi (1997) assumes that CLLD is possible in some infinitives introduced by a preposition in Italian.
(26) ??Luís quiere, los libros, leerlos. Luis wants the books read.inf-them ‘Luis wants the books to read them.’
Overt PRO in Romance
(CLLD; Gallego, 2010: 147)
(27) *Luis quiere CERVEZA beber (y no sidra). Luis wants BEER drink.inf and not cider ‘Luis wants BEER to drink (and not cider).’ (FF; Gallego, 2010: 147) Thus, the possibility of postverbal subjects in Romance infinitives is conditioned by the availability of a low discourse-sensitive position. Belletti (2004, 2005) claims for Italian that inverted subjects are located in a low Focus projection, in between vP and TP: (28) …[TP T [TopP Top [FocP Foc [TopP Top [vP v …]]]]]
(Belletti, 2004: 25)
If low discourse-sensitive positions are not conditioned by the C–T spine but by v, finiteness does not affect their availability. The approach to the overt/covert alternation in infinitives I will outline in the next sections therefore builds on the following hypothesis:9 (29) Overt ‘pronominal’ subjects in infinitives are licensed by [π] and not by Case.10 (Where [π]11 = discourse-sensitive features like [focus], [emphasis], [contrast]) This view is supported by Szabolcsi’s (2009) work on Hungarian, where the overt pronominal subject of control infinitives “occupies the same focus position in the infinitival clause as in the finite one” (Szabolcsi, 2009: 12). Thus, overt subjects in Romance infinitives rely on a parallel strategy to the licensing of focused in situ subjects in finite pro-drop clauses.
3. PRO = pro The pro-drop nature of Romance languages Spanish, Italian and Catalan has been linked to the ‘strong’ character of agreement morphology. Contreras (1991), Barbosa (1995), Alexiadou & Anagnostopoulou (1998), Kato (1999), and many others propose that AGR . Alonso-Ovalle & D’Introno (2001), Belletti (2005), Sundaresan (2010), Livitz (2011), Herbeck (2011), and Landau (2013b) also take the association with [focus] to be crucial. See for example Rigau (1988) and Fernández-Soriano (1989) for discussion of strong pronouns in finite pro-drop. . Miyagawa (2007) assumes that focus in an agreement-less language can fulfill a similar function to agreement in agreement languages. What the data from Spanish, Catalan, and Italian indicate is that even in an agreement language, salient discourse information can have a ‘licensing’ function of morpho-phonological content when AGR does not function as a mere agreement marker but itself acts as a [+referential] element. . I borrow the term from Fortuny (2008), who defines [π]-features as peripheral features.
Peter Herbeck
on T is ‘lexical’ in the sense that it can satisfy EPP-related requirements. In a model of feature valuation and Agree, this reasoning can be implemented by means of equipping T with an interpretable/valued set of phi-features (see Barbosa, 2009b).12 In Barbosa’s theory, the interpretable ϕ-features on T have the consequence of licensing a minimallyspecified nominal in Spec,v. Let us assume that this element is a D head with an unvalued set of phi-features: D[φ:_] (Landau, 2013b) – a ‘reference variable’ (Sigurðsson, 2008) or a ‘minimal pronoun’ (Kratzer, 2009; Landau, 2013b, c). The relation between T and D reflects the traditional identification requirement on pro (see Rizzi, 1986). In English, T has an unvalued set of phi-features and D enters the derivation fully-specified (see (31)): (30) Juan dice [CP que [TP T[iφ: 3p.pl]- durmieroni [vP D[φ:_] v- ti [VP … John says
that
slept.3pl
(31) John says [CP that [TP T[φ:_] [vP they[iφ: 3p.pl] v-slept [VP … Although Kato (1999) assumes that AGR directly bears the relevant theta-role in prodrop configurations, the derivation in (30) assumes that an element with the minimal descriptive content to satisfy a theta-role is merged in Spec,v. On a theory-internal level, this is motivated by Chomsky’s (2008) Dual Semantics, which implies that a nominal expression is externally merged in Spec,v if an external theta-role is to be assigned. On an empirical level, the prediction of approaches that assume direct thetarole satisfaction/absorption by AGR is that overt subjects are necessarily adjoined or dislocated. However, overt subjects in infinitives are thematic subjects (see Section 2). Thus, I take the merger of a nominal element with the minimal content necessary to satisfy a theta-role to be well-motivated:13 (32) EC = θREFα/φβ(Sigurðsson, 2008: 53) In Landau (2000, 2004) and Sigurðsson (2008), the emptiness of PRO in control is dissociated from Case and is linked to the variable nature of reference, correlating with defective phi-features. Unifying these views on control and pro-drop, empty subjects are generally the result of the lack of a phi-specification on D independently of whether we are dealing with a finite or a nonfinite structure:14
. Holmberg (2005) also links the possibility of Romance pro-drop to interpretable features on T. However, the author postulates an interpretable D-feature. In this theory, the empty subject in Spec,v has a fully-specified set of phi-features, in contrast to what is assumed here. . Differently from what was assumed in Herbeck (2011), where T itself satisfies the relevant theta-role. . See Huang (1989) for a PRO/pro unification in Chinese. See also Holmberg (2005) and Landau (2013b) who hint at the possibility of such a unification. Livitz (2011), on the other
Overt PRO in Romance
(33) PRO = pro = D[φ:_] In pro-drop, it is pronominal AGR, implemented as T[iϕ], that identifies D[ϕ:_] in Spec,v via Agree. According to Barbosa (2009a, 2010), T is equipped with interpretable phifeatures also in control infinitives in European Portuguese. This implies that abstract, interpretable AGR can be present on T in Romance infinitives, even though it is defective (i.e. lacking a specification for [number] and/or [person]):15 (34) Juan intentó [TP T[iφdef]-dormir [vP D[φ:_] v-dormir… John tried
sleep.inf
A defective T head cannot identify D[ϕ:_] so that the empty subject is anaphorically linked to a matrix antecedent via Agree (see Landau, 2000, 2004). Given that anaphors are phi-deficient (Reinhart & Reuland, 1993), T in control is the anaphoric element just like T is the pronominal element in pro-drop: T[iϕdef] is Borer’s (1989) and Landau’s (2000) anaphoric AGR. According to this reasoning, the empty subject in control and pro-drop is the same at the level of External Merge, independently of overt agreement markings. Null realization results from the lack of a phi-specification, given that there is no explicit instruction at PF for phonological realization. The only way for D[ϕ:_] to receive phonological content is to receive a phi-value before Spell-Out. This possibility potentially exists in pro-drop as well as control. In the next section, I outline an approach that makes these ideas explicit, building on a model of late insertion and phasal Spell-Out (see also Sundaresan, 2010 for the PRO/self-anaphor alternation).16
4. The overt/covert alternation of subjects in control 4.1 Subjects, late insertion, and the syntax-pragmatics interface Null subjects are nominal elements with variable phi-features (Sigurðsson, 2008; Landau, 2013b). Lack of a phi-value is translated, not only into variable reference at LF, but further into lack of phonological instruction at PF:
hand, argues that pro and PRO are differentiated in terms of valued versus unvalued ϕPs, respectively. . See Rigau (1995) and Torrego (1998) for a discussion of the presence of abstract AGR in some Spanish and Catalan infinitives. See also Gallego (2010) for discussion of Tdef. . In Sundaresan’s (2010) approach, PRO is empty because its features cannot be valued by Cdef in OC complements, but PRO is nevertheless spelled-out. The author further analyses the PRO/anaphor alternation as a 'syntactically conditioned allomorphy', which is different from what I assume here.
Peter Herbeck
(35) D[φ:_] → Ø If morphophonological rules operate on D[φ:_] before valuation by T, null pronunciation arises. Thus, the only way for D[φ:_] to be overtly realized is to receive a phi- specification before Spell-Out – if morphophonological operations apply to D after Agree with T, there will be an explicit instruction for (overt) realization: (36) a. D[1sg] → /ʝo/
b. D[2sg] → /tu/ …
The question is what discriminates between the two options. As claimed in Section 2.3, discourse-sensitive information is crucial to mediate the overt/covert alternation in non-finite domains. If we adopt a phase-based model along the lines of Chomsky (2000, 2001, 2008), there are at least two points in the syntactic derivation at which D interacts with the external interfaces: at the completion of the vP and the CP phases. These two interface points mirror Rizzi’s (1997) high and Belletti’s (2004) low left peripheries (see López, 2009; Gallego, 2010): (37)
CP
interface point TP
C
vP
T v
interface point VP
In (37), D[φ:_] can be associated with discourse-sensitive information at the completion of the vP and the CP phases. López (2002) furthermore assumes that [π]-features are associated with morphosyntactic feature bundles before morphophonological insertion: Morphology → PF (38) Lexicon → CHL → Pragmatics LF (López, 2002: 31) In a model that adopts late insertion as in Distributed Morphology (DM; see Halle & Marantz, 1993; Harley & Noyer, 1999; Embick & Noyer, 2007; Embick & Marantz, 2008; Embick, 2010 and related work), syntax operates on “purely abstract” (Harley & Noyer, 1999: 3) categories and “the assignment of phonological features to morphosyntactic feature bundles takes place after the syntax […]” (Halle & Marantz, 1993: 113). Given the direct communication between the pragmatic and the morphological component in (38), opting between null and overt realization (i.e. (35) or (36)) is mediated by the pragmatic interface in the following way: If D[ϕ:_] is associated with [π], application of (35) will be blocked given that null realization is
Overt PRO in Romance
by definition incompatible with [contrast], [focus] or [emphasis]. Thus, if [π] is assigned to D[φ:_] in Spec,v, the only way for the derivation to converge is for D[ϕ:_] to receive a phi-value: (39) Discourse-sensitive information at the phase edges triggers delay of morphophonological insertion. This statement expresses that phonological insertion into D must be motivated as null pronunciation arises if no explicit trigger is present in the syntactic derivation and, hence, is the unmarked case: (40) Insertion applies if possible.17 Morphophonological insertion into D in Spec,v will apply as soon as the vP phase is completed. At this stage, phi-valuation by T has not taken place, resulting in null insertion. The unmarked status of null subjects has a long tradition in the literature and has been implemented in the form of various principles, such as Chomsky’s (1981) Avoid Pronoun.18 The following derivation depicts the case of a null subject inside a control infinitive:19
. Compare with Pesetsky’s (1989) Earliness Principle. . Further principles are, for example, Bonet’s (1991) Avoid Pronoun Strength and Cardinaletti & Starke’s (1999) Economy of Representation. See also Alonso-Ovalle & D’Introno (2001: 404f). . The derivations I propose apparently go against the ‘standard’ view that it is the complement of phase heads (i.e. VP and TP) that is spelled-out and made inaccessible to further computation and not the full phases CP and vP, as expressed by Chomsky’s Phase Impenetrability Condition:
(i) The domain of H is not accessible to operations outside HP; only H and its edge are accessible to such operations. (Chomsky, 2001: 13)
However, given that phases are defined on the basis of “semantic-phonetic integrity” (Chomsky, 2001: 14) and on the basis of independence at the interfaces by Chomsky (2000, 2001, 2004), we would a priori expect vP and CP to be interface points with respect to semantic and phonological interpretation: At SEM, vP and CP (but not TP) are propositional constructions: vP has full argument structure and CP is the minimal construction that includes tense and event structure and (at the matrix, at least) force. At PHON, these categories are relatively isolable (in clefts, VP-movement, etc.). (Chomsky, 2004: 124) This indicates that the SEM and PHON criteria used for the definition of phases do not correlate with the actual points of Spell-Out (as discussed by Richards, 2011 and references cited therein). Thus, Chomsky (2001: 12) states that “the whole phase is ‘handed over’ to the phonological component”, which indicates that interface interpretation may apply to phases
Peter Herbeck
(41) (i) Construction of vP: [vP D[φ:_] dormir [VP dormir]]
(ii) Morphophonological insertion: D[φ:_] → Ø
(iii) v-to-T head movement and construction of matrix vP: [vP Marta v-prometió [CP Cdef [TP T[φ: def]-dormir [vP D[φ:_] v…
(iv) Agree between matrix DP and D[φ:_] → phi-valuation: [vP Marta v-prometió [CP Cdef [TP T[φ: def]-dormir [vP D[φ:3p.sg.f]…
Martha promised
sleep.inf
In (41), null pronunciation of D arises as a consequence of the absence of an explicit trigger to do otherwise: D[ϕ:_] lacks phi-features by the completion of the embedded vP and, hence, there is no explicit instruction to the morphophonological component to insert a phonological matrix. Given that D is in the phase edge, it is still available to phi-valuation by a matrix antecedent. If, however, vP interfaces with the pragmatic component and, given the right discourse conditions, a [π]-marking ([focus], [contrast], or [emphasis]) is assigned to Spec,v, rule (35) will be blocked. Association of D with [π] has the consequence of delaying morphophonological insertion until the next higher strong phase boundary is construed. Delay is possible because (i) D, being in the phase edge, is available for further computation and (ii) control infinitives are not strong phases and, thus the next higher strong phase is matrix vP, where insertion and valuation takes place: (42) (i) Construction of vP: [vP D[φ:_] dormir [VP dormir]]
(ii) [π]-assignment ([emphasis]) to phase edge (→ blocks null insertion): [vP D[φ:_]/[emphasis] v-dormir [VP …
(iii) v-to-T, construction of matrix vP, phi-valuation: [vP Marta v-prometió [CP Cdef [TP T[φ: def]-dormir [vP D[φ:_]/[emphasis] v … (iv) Insertion at matrix vP level: D[φ:3p.sg.f] → /eʎa/
Crucial for the possibility of overt subjects inside infinitives is therefore the availability of a low vP-related, discourse-sensitive position which is independent of the finiteness of the clause (Belletti, 2004, 2005). If this low position is activated, phonological insertion will be delayed until phi-valuation of D[ϕ:_].
as a whole. In this paper, it is crucial to accept vP and CP as possible interface points for the application of pragmatic rules (see López, 2002, 2009) as well as phonological insertion to account for the overt/covert alternation of the subject position, even though it is VP (and not vP) that is actually made inaccessible to further computation.
Overt PRO in Romance
This approach can explain why the overt subject in OC as well as NOC infinitives is morphologically pronominal: overt realization of D[φ:_] in control infinitives relies on a strategy analogical to finite pro-drop. The difference lies in whether D[φ:_] receives a phi-feature specification via Agree with pronominal AGR (Barbosa, 1995, 2009b) or by a matrix antecedent via anaphoric AGR (Borer, 1989; Landau, 2000). The postsyntactic morphological component, however, does not ‘see’ how reference linking is established – PF only ‘sees’ D with a specified or unspecified phi-value, resulting in overt or null pronunciation.
4.2 Pronouns versus lexical DPs: Why are they different? In the case of OC, the impossibility of lexical DPs could be argued to derive from a violation of Binding Condition C: (43) *Julia quería telefonear Paula. Julia wanted phone.inf Paula
(Piera, 1987: 160)
However, this reasoning cannot account for mismatches between pronouns and full R-expressions in NOC contexts (see (19)–(20)). One way of accounting for the degradedness of lexical DPs would be to assume that NOC involves logophoric control (Landau, 2000, 2013a; Bianchi, 2003; see Pérez Vázquez, 2007 and Sitaridou, 2007 for further discussion in the context of overt subjects in infinitives). Belletti (2005) observes that overt subjects in Italian infinitives are preferably [1st person]: (44) In una situazione del genere, parlare ((?)io/??tu/??lui/ (?)noi/??voi/??loro/*Gianni) sarebbe strano. ‘In a situation of this kind, to speak I/you/he/Gianni would be strange’. (Italian; Belletti, 2005: 23) The degrees of acceptability in (44) indicate that the notions [±speaker] and [±addressee] are vital to explain the licensing of overt subjects. If logophoricity is involved in NOC, this pattern is expected: Bianchi (2003) claims that phi-features can be linked either to an internal or an external logophoric center in the C-domain. Logophoric pronouns refer to an internal protagonist of a narration (Huang, 2000: 172). Given that [1.p] encodes the external speaker, one could speculate that the tendency in (44) is due to shifting from internal to external logophoric anchoring in case the internal logophoric centre is non-recoverable. Another way to reason would be that the preference for [1.p] reflects Spell-Out of the ‘self ’-component involved in logophoric control. Further investigations into the exact pragmatic licensing conditions on overt subjects in infinitives will hopefully shed more light on this issue. What is important at this point is that the pragmatic interface is directly involved in the licensing of overt pronouns in NOC infinitives, an observation that can be captured if we deal with these
Peter Herbeck
elements as [π]-triggered Spell-Out as a result of a direct communication between pragmatics and morphology. A further, complicating factor comes into play if we look at certain NOC configurations in Spanish, where, in contrast to Italian (44), R-expressions are possible: (45) a. [Venir Julia con nosotros] sería una novedad. come.inf Julia with us be.cond-3sg a novelty ‘If Julia came with us, that would be a novelty. (Fernández & Anula, 1994: 479) b. [Presentarse Julia a las elecciones] fue un error. present.inf-SE Julia to the elections was an error ‘It was a mistake that Julia ran for election.’ (Piera, 1987: 164) Thus, while Italian seems to require Aux-to-Comp to license full DP subjects, Spanish allows overt subjects even in several NOC configurations that do not involve an auxiliary (see Mensching, 2000; Pérez Vázquez, 2007). The desirable result would be to derive this pattern from general, independent differences between the two languages. It has been noted in Section 2.1.1 that S panish allows in situ Case assignment to subjects in finite clauses. Cardinaletti (1999), on the other hand, claims that subjects in Italian check structural nominative Case obligatorily in Spec,T.20 This difference correlates with the grammaticality of VSO in Spanish matrix clauses but its ungrammaticality in Italian and Catalan (see Belletti, 2004; López, 2009; Gallego, 2010): (46) a. Todos los días compra Juan el diario. every day buys Juan the newspaper ‘Juan buys the newspaper every day.’ (Spanish; Zubizarreta, 1998: 100) b. ???Avui farà en Joan el dinar. today make.fut the John the lunch ‘Today John will cook the lunch.’ c. *Ha comprato Maria il giornale. has bought Maria the newspaper ‘Mary has bought the newspaper.’
(Catalan; Solà, 1992: 11)
(Italian: Belletti, 2004: 26)
In the grammar of the Spanish speakers that allow full R-expressions in postverbal position of NOC infinitives, a deficient T[iϕ] can assign [nom] Case in situ (see also Mensching, 2000). In Italian, in contrast, [nom] in situ is generally unavailable such that a postverbal subject could only be licensed by [π]-triggered Spell-Out in NOC infinitives.
. López (2009) also argues that subjects move to Spec,T in Catalan.
Overt PRO in Romance
Thus, Case still plays a residual licensing function of overt subjects in non-finite domains, though, as Cardinaletti (1999) argues for Italian, its function is restricted to lexical DPs. Further evidence comes from adjunct infinitives, which, as we have seen in (21), productively license full R-expressions in Spanish and Catalan (Rigau, 1995, examples from Ordóñez, 2009: 1): (47) a. Antes de comprar (Luis) manzanas (Luis),… (Spanish) before of buy.inf Luis apples Luis b. Abans de comprar (?*en Lluis) pomes (en Lluis),… (Catalan) before of buy.inf the Luis apples the Luis ‘Before Luis bought apples,…’ In Rigau’s (1995) analysis, the introducing preposition triggers T-to-C movement. If this mechanism provides an alternative Case-assigning mechanism, similarly to Rizzi’s (1982) Aux-to-Comp, it is expected that full DPs are licit. Furthermore, (47) demonstrates that DP subjects in infinitives obey the language-specific word order restrictions observed in finite clauses, blocking VSO in Catalan. However, also in this case, substituting a lexical DP with a pronoun improves the sentence for some speakers: (48) Abans de comprar (? jo) pomes,… before of buy.inf I apples ‘Before I bought apples,…’ This is expected if DPs in situ need Case and focus marking in the syntax in Catalan (Bonet, 1989) while pronouns can use an alternative strategy of [π]-triggered, postsyntactic insertion.21 The question remains, however, why this should be so, i.e. why do full DPs need Case in postverbal position while pronouns can be exempt from it? Cardinaletti & Starke (1999) suggest that strong pronouns can check ‘default’ Case internally to their own CP-like projection, in contrast to weak pronouns, which lack CP. Cardinaletti (1999) argues that lexical DPs also lack the capacity of DP-internal Case-checking. According to this theory, however, the following puzzle emerges: why do lexical DPs need structural Case like weak pronouns if they should have an equally-rich projection like strong (and not like weak) pronouns? That is, according to their richness of projection, R-expressions should pattern with strong pronouns, however, with respect to Case-licensing, they pattern with weak pronouns. I have followed Halle & Marantz (1993) in assuming that a phonological make-up is provided to morphosyntactic feature bundles post-syntactically. In DM, however,
. As an anonymous reviewer points out, an alternative view analyzes the difference between Spanish and Italian/Catalan in terms of the presence versus absence of an additional position for subjects (see Ordóñez, 2007).
Peter Herbeck
whether the same process of late insertion applies to Roots is not uncontroversial. If we follow Embick & Halle (2005) and Embick & Noyer (2007), morphosyntactic features (i.e. abstract morphemes) like D and phi are subject to Vocabulary Insertion in the post-syntactic component; Roots, on the other hand, “are sequences of complexes of phonological features” (Embick & Noyer, 2007: 295) and “are not subject to insertion” (Embick & Halle, 2005: 41). Let us now have a look at the internal structure of nominal expressions. While empty subjects are ‘reference variables’ (Sigurðsson, 2008) and overt ‘pronominal’ subjects are mere Spell-Out of such elements, R-expressions and lexical DPs cannot be reduced in the same way, given that they contain a Root: (49) a.
DP b.
DP D
ϕP
D
ϕ[_]
ϕP ϕ
√ROOT
I therefore suggest that the more restricted licensing conditions on R-expressions and lexical DPs derive from the following generalization for Spanish:22 (50) A DP must be licensed by structural Case in syntax only if it bears phonological features in syntax. Principle (50) correctly imposes the requirement of structural Case on R-expressions and lexical DPs in Spanish, Italian and Catalan given that they contain a Root and, thus have phonological features in syntax. Strong pronouns, on the other hand, are an assembly of phi- and D-features, which acquire reference in discourse (by means of being linked to an external or internal logophoric centre; Bianchi, 2003). Hence, they lack inherent reference and, arguably, a Root. Given that they lack phonological features in the syntax, they may escape the traditional Case Filter and their realization can be licensed by discourse-sensitive features on a post-syntactic level.
5. Conclusion Spanish, Catalan, and Italian allow instances of [π]-triggered, post-syntactic insertion of morphophonological material into subject positions, which is not dependent on
. This is a modification of the traditional Case Filter:
(i)
*NP if NP has phonetic content and has no Case.
(Chomsky, 1981: 49)
See also Fernández-Salgueiro (2011) for discussion of the Case Filter in pro-drop.
Overt PRO in Romance
structural Case. Thus, the overt/covert alternation of the subject position in infinitives is crucially governed by the availability of discourse-sensitive information at the low vP phase edge (López, 2002; Belletti, 2004) as triggers for delay of morphophonological insertion. The PRO/pro distinction is not an inherent property of nominal elements. Null subjects are uniformly ‘reference variables’ (Sigurðsson, 2008), i.e. minimally specified nominal elements (Barbosa, 2009b; Landau, 2013b), whose phonological emptiness derives from a lack of a phi-specification at PF. Differences with respect to Binding Theory derive from the properties of the C–T spine (Borer, 1989; Landau, 2000, 2013a). Overt realization is effected at PF if a phi-value is provided to D before morphophonological insertion. Thus, the overt/covert alternation of subjects in pro-drop and control depends on the derivational point at which a subject interfaces with the external systems. The ordering of insertion and valuation is crucially mediated by the pragmatic interface.
References Alexiadou, A., & Anagnostopoulou, E. (1998). Parameterizing AGR: Word order,V-movement, and EPP-checking. Natural Language and Linguistic Theory, 16, 491–539. DOI: 10.1023/a:1006090432389 Alexiadou, A., Anagnostopoulou, E, Iordǎchioaia, G., & Marchis, M. (2010). No objections to Backward Control. In N. Hornstein & M. Polinsky (Eds.), Movement theory of control (pp. 89–117). Amsterdam: John Benjamins. DOI: 10.1075/la.154.04ale Alonso-Ovalle, L., & F. D’Introno (2001). Full and Null Pronouns in Spanish – The Zero Pronoun Hypothesis. In H. Campos, E. Herburger, A. Morales Front & T. J. Walsh (Eds.), Hispanic linguistics at the turn of the millenium (pp. 400–414). Somerville, MA: Cascadilla Press. Barbosa, P. (1995). Null subjects. Doctoral Dissertation. MIT. Retrieved from 〈http://www.ai.mit. edu/projects/dm/theses/barbosa95.pdf〉 Barbosa, P. (2009a). A case for an Agree-based theory of control. 11th Seoul International Conference on Generative Grammar Proceedings, 101–123. Barbosa, P. (2009b). Two kinds of subject pro. Studia Linguistica, 63(1), 2–58. DOI: 10.1111/j.1467–9582.2008.01153.x Barbosa, P. (2010). Overt subjects in raising and control complements. LSA Meeting Extended Abstracts. Retrieved from 〈http://elanguage.net/journals/lsameeting/article/view/2010.3/ pdf〉 Belletti, A. (2004). Aspects of the low IP area. In L. Rizzi (Ed.), The structure of CP and IP (pp. 16–51). Oxford: Oxford University Press. DOI: 10.1017/s0022226706304380 Belletti, A. (2005). Extended doubling and the VP periphery. Probus, 17, 1–35. DOI: 10.1515/prbs.2005.17.1.1 Bianchi, V. (2003). On finiteness as logophoric anchoring. In J. Guéron & L. Tasmowski (Eds.), Temps et point de vue / Tense and point of view (pp. 213–246). Paris: Publidix. Boeckx, C., Hornstein, N., & Nunes, J. (2010). Control as movement. Cambridge: Cambridge University Press. DOI: 10.1017/cbo9780511761997 Bonet, E. (1989). Subjects in Catalan. MIT Working Papers in Linguistics, 13, 1–26.
Peter Herbeck Bonet, E. (1991). The Person-Case-Constraint: A morphological approach. MIT Working Papers in Linguistics, 22, 33–52. Borer, H. (1989). Anaphoric AGR. In O. Jaeggli & K. Safir (Eds.), The null subject parameter (pp. 69–109). Dordrecht: Springer. DOI: 10.1007/978-94-009-2540-3_3 Burzio, L. (1986). Italian syntax – A Government-Binding approach. Dordrecht: Reidel. DOI: 10.2307/414791 Cardinaletti, A. (1999). Italian emphatic pronouns are postverbal subjects. University of Venice Working Papers in Linguistics 9(1.2), 59–92. Cardinaletti, A., & Starke, M. (1999). The typology of structural deficiency. In H. van Riemsdijk (Ed.), Clitics in the languages of Europe (pp. 145–233). Berlin: Mouton de Gruyter. DOI: 10.1017/s0022226703282293 Chomsky, N. (1981). Lectures on government and binding – The Pisa lectures. Dordrecht: Foris. DOI: 10.2307/2273965 Chomsky, N. (1995). The Minimalist program. Cambridge MA: The MIT Press. DOI: 10.1017/s0272263197241070 Chomsky, N. (2000). Minimalist inquiries: The framework. In R. Martin, D. Michaels, & J. Uriagereka (Eds.), Step by step: Essays on minimalist syntax in honor of Howard Lasnik (pp. 89–115). Cambridge, MA: The MIT Press. DOI: 10.1017/s0022226702271625 Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A life in language (pp. 1–52). Cambridge, MA: The MIT Press. DOI: 10.1017/s0022226704322747 Chomsky, N. (2004). Beyond explanatory adequacy. In A. Belletti (Ed.), Structures and beyond (pp. 104–131). Oxford: Oxford University Press. Chomsky, N. (2008). On phases. In R. Freidin, C.P. Otero, & M. L. Zubizarreta (Eds.), Foundational issues in linguistic theory. Cambridge, MA: The MIT Press. DOI: 10.7551/mitpress/9780262062787.001.0001 Contreras, H. (1991). On the position of subjects. In S. Rothstein (Ed.), Perspectives on phrase structure: Heads and licensing (Syntax and Semantics 25; pp. 63–79). San Diego, CA: Academic Press. DOI: 10.1017/s002222670000013x Embick, D., & Halle, M. (2005). On the status of stems in morphological theory. In T. Geerts, I. van Ginneken, & H. Jacobs (Eds.), Proceedings of Going Romance 2003 (pp. 37–62). Amsterdam: John Benjamins. DOI: 10.1075/cilt.270.03emb Embick, D., & Noyer, R. (2007). Distributed Morphology and the syntax/morphology interface. In G. Ramchand & C. Reiss (Eds.), The Oxford handbook of linguistic interfaces (pp. 289–324). Oxford: Oxford University Press. DOI: 10.1093/oxfordhb/9780199247455.013.0010 Embick, D., & Marantz, A. (2008). Architecture and blocking. Linguistic Inquiry, 39(1), 1–53. DOI: 10.1162/ling.2008.39.1.1 Embick, D. (2010). Localism versus globalism in morphology and phonology. Cambridge, MA: The MIT Press. DOI: 10.7551/mitpress/9780262014229.001.0001 Fernández-Lagunilla, M. (1987). Los infinitivos con sujetos léxicos en español. In V. Demonte & M. Fernández-Lagunilla (Eds.), Sintaxis de las lenguas románicas (pp. 125–147). Madrid: Ediciones el arquero. DOI: 10.2307/414748 Fernández-Lagunilla, M., & Anula-Rebollo, A. (1994). Procesos de filtrado de rasgos categoriales en la sintaxis. Los infinitivos y la legitimación del caso nominativo. In V. Demonte (Ed.), Gramática del español (pp. 471–530). México: El Colegio de México. Fernández-Salgueiro, G. (2011). Romance null subjects at the sensory-motor interface. Syntax, 14(1): 1–28. DOI: 10.1111/j.1467–9612.2010.00147.x
Overt PRO in Romance
Fernández-Soriano, O. (1989). Strong pronouns in null-subject languages and the Avoid Pronoun Principle. MIT Working Papers in Linguistics, 11, 228–239. Fortuny, J. (2008). The emergence of order in syntax. Amsterdam: John Benjamins. DOI: 10.1075/la.119 Gallego, Á. (2010). Phase theory. Amsterdam: John Benjamins. DOI: 10.1075/la.152 Halle, M., & Marantz, A. (1993), Distributed Morphology and the pieces of inflection. In K. Hale & S. J. Keyser (Eds.), The view from Building 20 (pp. 111–76). Cambridge, MA: The MIT Press. Harley, H., & Noyer, R. (1999). Distributed Morphology. Glot International, 4(4), 3–9. DOI: 10.1515/9783110890952.463 Herbeck, P. (2011). Overt subjects in Spanish control infinitives and the theory of empty categories. Generative Grammar in Geneva, 7, 1–22. Herbeck, P. (2014). (Backward) Control and clitic climbing: On the deficiency of nonfinite domains in Spanish and Catalan. ConSOLE XXI Proceedings 2013, 123–142. Hernanz, M. Ll. (1999). El infinitivo. In I. Bosque & V. Demonte (Eds.), Gramática descriptiva de la lengua española (pp. 2197–2356). Madrid: Espasa. Hernanz, M. Ll. (2011). Sobre la periferia de los infinitivos. In A V. Escandell-Vidal, M. Leonetti, & C. Sánchez López (Eds.), 60 problemas de Gramática dedicados a Ignacio Bosque (pp. 263–270). Madrid: Ediciones Akal. DOI: 10.4312/ah.6.1.171–175 Holmberg, A. (2005). Is there a little pro? Evidence from Finnish. Linguistic Inquiry, 36(4), 533–564. DOI: 10.1162/002438905774464322 Hornstein, N. (1999). Movement and control. Linguistic Inquiry, 30, 69–96. DOI: 10.1162/002438999553968 Huang, C. T. J. (1989). Pro-Drop in Chinese: A generalized control theory. In O. Jaeggli & K. Safir (Eds.), The null subject parameter (pp. 185–214). Dordrecht: Springer. Huang, Y. (2000). Anaphora – A crosslinguistic approach. Oxford: Oxford University Press. DOI: 10.1017/s0022226704273236 Kato, M.A. (1999). Strong and weak pronominals in the null subject parameter. Probus 11: 1–37. DOI: 10.1515/prbs.1999.11.1.1 Kayne, R. (2002). Pronouns and their antecedents. In S. D. Epstein & T. D. Seely (Eds.), Derivation and explanation in the minimalist program (pp. 133–166). Oxford: Blackwell. DOI: 10.1002/9780470755662.ch7 Kratzer, A. (1998). More structural analogies between pronouns and tenses. SALT VIII, MIT. Kratzer, A. (2009). Making a pronoun: Fake indexicals as windows into the properties of pronouns. Linguistic Inquiry, 40(2): 187–237. DOI: 10.1162/ling.2009.40.2.187 Landau, I. (2000). Elements of control. Dordrecht: Kluwer. Landau, I. (2004). The scale of finiteness and the calculus of control. Natural Language and Linguistic Theory, 22, 811–877. DOI: 10.1007/s11049-004-4265-5 Landau, I. (2013a). Control in generative grammar: A research companion. Cambridge: Cambridge University Press. DOI: 10.1017/s0022226713000364 Landau, I. (2013b). A two-tiered theory of control. Ms. Retrieved from 〈http://ling.auf.net/ lingbuzz/001937〉 Landau, I. (2013c). Agreement at PF: An argument from partial control. Ms. Retrieved from 〈http://ling.auf.net/lingbuzz/001813〉 Livitz, A. (2011). Incorporating PRO: A defective goal analysis. NYU Working Papers in Linguistics, 3, 95–119.
Peter Herbeck López, L. (2002). Toward a grammar without TopP or FocP. In S. Mauck & J. Mittelstaedt (Eds.), Proceedings of the Workshop on the Syntax and Semantics of the C-domain – Georgetown Working Papers in Linguistics, 2, 181–219. López, L. (2009). A derivational syntax for information structure. Oxford: Oxford University Press. DOI: 10.1017/s0022226710000150 Mensching, G. (2000). Infinitive constructions with specified subjects. Oxford: Oxford University Press. DOI: 10.1017/s0022226702241638 Miyagawa, S. (2007). Unifying agreement and agreementless languages. In M. Kelepir & B. Öztürk (Eds.), MIT Working Papers in Linguistics, 54, 47–66. Ordóñez, F. (1998). Postverbal asymmetries in Spanish. Natural Language and Linguistic Theory, 16, 313–346. DOI: 10.1023/a:1006051703562 Ordóñez, F. (2007). Observacions sobre la posició dels subjectes postverbals en català i castellà. Caplletra, 42, 251–272. Ordóñez, F. (2009). Backwards or in situ? Infinitival Overt Subjects in Romance (Handout). 19th Colloquium on Generative Grammar. UPV-EHU. Pérez Vázquez, E. (2007). El infinitivo y su sujeto en español. Bologna: Gedit edizioni. Pesetsky, D. (1989). Language-particular processes and the Earliness Principle. Ms. MIT. Retrieved from 〈http://web.mit.edu/linguistics/people/faculty/pesetsky/earliness.pdf〉 Piera, C. (1987). Sobre la estructura de las cláusulas de infinitivo. In V. Demonte & M. Fernández- Lagunilla (Eds.), Sintaxis de las lenguas románicas (pp. 148–166). Madrid: Ediciones el arquero. DOI: 10.2307/414748 Pöll, B. (2007). On the licensing of overt subjects in Spanish infinitival clauses. Probus, 19, 93–120. DOI: 10.1515/probus.2007.004 Polinsky, M., & Potsdam, E. (2002). Backward Control. Linguistic Inquiry, 33, 245–282. DOI: 10.1162/002438902317406713 Postal, P. (1969). On so-called pronouns in English. In D. Reibel & S. Schane (Eds.), Modern studies in English: Readings in transformational grammar (pp. 201–224). Englewood Cliffs, NJ: Prentice-Hall. Reinhart, T., & Reuland, E. (1993). Reflexivity. Linguistic Inquiry, 24, 657–720. Richards, M.D. (2011). Deriving the edge: What’s in a phase? Syntax, 14(1), 74–95. DOI: 10.1111/j.1467–9612.2010.00146.x Rigau, G. (1988). Strong pronouns. Linguistic Inquiry, 19(3), 503–511. Rigau, G. (1995). The properties of the temporal infinitive constructions in Catalan and Spanish. Probus, 7, 279–301. DOI: 10.1515/prbs.1995.7.3.279 Rizzi, L. (1982). Issues in Italian syntax. Dordrecht: Foris. DOI: 10.1515/9783110883718 Rizzi, L. (1986). Null objects in Italian and the theory of pro. Linguistic Inquiry, 17, 501–557. Rizzi, L. (1997). The fine structure of the left periphery. In L. Haegeman (Ed.), Elements of grammar: Handbook of generative syntax (pp. 281–337). Dordrecht: Kluwer. DOI: 10.1007/978-94-011-5420-8_7 Sánchez, L. (1993). On the licensing and interpretation of non-argumental nominals. Ms. University of Southern California. Sánchez-López, C. (1996). Los pronombres enfáticos y la estructura subeventiva. Verba, 23, 147–175. Sigurðsson, H.Á. (2008). The case of PRO. Natural Language and Linguistic Theory, 26, 403–450. DOI: 10.1007/s11049-008-9040-6
Overt PRO in Romance
Sitaridou, I. (2007). Romance infinitives with subjects, subjunctive obviation and control theory. In L. Eguren & O. Fernández Soriano (Eds.), Coreference, modality, and focus (pp. 191–219). Amsterdam: John Benjamins. DOI: 10.1075/la.111.10sit Solà, J. (1992). Agreement and subjects. Doctoral Dissertation. Universitat Autònoma de Barcelona. Retrieved from 〈http://filcat.uab.es/clt/publicacions/tesis/pdf/Sola.pdf〉 Sundaresan, S. (2010). A reductionist treatment of control and anaphora. Ms. University of Tromsø & University of Stuttgart. Suñer, M. (1986). Lexical subjects of infinitives in Caribbean Spanish. In O. Jaeggli & C. SilvaCorvalan (Eds.), Studies in Romance linguistics (pp. 189–203). Dordrecht: Foris. Szabolcsi, A. (2009). Overt nominative subjects in infinitival complements crosslinguistically: Data, diagnostics, and preliminary analyses. Ms. Retrieved from 〈http://ling.auf.net/ lingBuzz/00044〉 Torrego, E. (1996). On quantifier float in control clauses. Linguistic Inquiry, 27, 111–126. Torrego, E. (1998). Nominative subjects and Pro-Drop Infl. Syntax, 1–2, 206–219. DOI: 10.1111/1467–9612.00008 Zubizarreta, M.L. (1998). Prosody, focus, and word order. Cambridge, MA: The MIT Press.
The semantics and pragmatics of andar and venir + gerund* Lorena Sainz-Maza Lecanda & Meagan E. Horn The Ohio State University
This study builds on existing research regarding the characterization of Spanish aspectual periphrases, such as vivir + gerund or ir + gerund, and explores and compares the distribution of the periphrases andar + gerund and venir + gerund in Peninsular Spanish. We propose that a complete analysis requires both semantic and pragmatic considerations. First, although both andar + gerund and venir + gerund behave as frequentative pluractional operators, we show that they differ with respect to the nature of the frequency of event repetitions. Second, we posit that these periphrastic constructions differ pragmatically as a result of two generalized conversational implicatures from which differing meanings involving the agent’s intention towards the accomplishment of a goal and the agent’s level of engagement arise. Keywords: aspectual periphrases; Peninsular Spanish; pluractionality; conversational implicatures
1. Introduction Spanish periphrastic constructions have been studied both from a semantic point of view (Laca, 2002, 2004, 2006) as well as from a pragmatic perspective (Amaral, 2013a, 2013b). Laca (2006) shows that periphrases with motion verbs such as ir and andar + gerund can be analyzed as pluractional operators in verb phrases and further explores the interaction of these operators with nominal arguments such as indefinite, quantified and plural noun phrases. In a similar fashion, Amaral (2013a) proposes that the periphrastic construction of the type vivir + gerund employed in Colombian Spanish conveys a pluractional meaning and posits that the properties of
* We would like to thank Scott A. Schwenter for his valuable feedback on previous versions of this manuscript. We wish to extend our gratitude to the attendees of the poster session at the Hispanic Linguistics Symposium 2013 and to two anonymous reviewers for their insightful comments on this paper. All errors are ours.
doi 10.1075/ihll.4.03sai © 2015 John Benjamins Publishing Company
Lorena Sainz-Maza Lecanda & Meagan E. Horn
vivir + gerund reflect a pragmatic principle associated more generally with expressions denoting high number. In this study, we build on the existing research regarding the characterization of aspectual periphrases carried out by Amaral (2013a, 2013b) and Laca’s pluractionality proposal (2002, 2004, 2006) and we explore and compare the semantic and pragmatic distribution of the periphrases andar + gerund and venir + gerund in Peninsular Spanish, depicted in (1) and (2) respectively: (1) Últimamente ando yendo al psicólogo porque necesito liberar emociones. ‘I have been going to the psychologist lately because I need to let my emotions out.’
(2) Vengo escuchando la misma canción desde hace días. ‘I have been listening to the same song for the past few days.’ We show, following Laca’s (2002, 2004, 2006) description of andar + gerund, that venir + gerund is also a pluractional operator conveying a frequentative meaning. However, we suggest that although both periphrastic constructions have a pluractional meaning, the nature of their event repetitions differs. Additionally, we present a pragmatic analysis of andar and venir followed by a gerund and we suggest that these two periphrastic expressions are best understood as consisting of two generalized conversational implicatures involving the agent’s intention towards the accomplishment of a goal and the agent’s level of engagement. This paper is organized as follows. In Section 2, we show that the periphrastic constructions under investigation here should be crucially understood as monoclausal complex predicates wherein the semantic meaning of motion is lost and that these predicates differ from biclausal predicates in which the verbs of motion keep their semantic meaning and constitute separate predicates from the non-finite gerund form. In Section 3, we investigate the semantic properties of andar and venir followed by a gerund, which reveal substantial differences in their configuration. Section 4 is devoted to the pragmatic analysis of these periphrastic constructions and we posit that two generalized conversational implicatures arise for each aspectual construction. Finally, Section 5 consolidates the main points discussed throughout the paper and offers suggestions to carry out further research intended to investigate Spanish pluractional periphrases.
2. Syntactic criteria In order to demonstrate how the semantic and pragmatic meanings andar + gerund and venir + gerund are permitted to arise, it is important to first understand the potential syntactic arrangements of both periphrastic constructions. To begin, Amaral (2013a)
The semantics and pragmatics of andar and venir + gerund
shows that a similar periphrastic construction in Colombian Spanish containing the verb vivir + gerund can be syntactically analyzed as either a monoclausal or biclausal predicate. When a finite verb functions as a main verb and is modified by a non-finite verb form (i.e. a gerund), the series of verbs is said to be biclausal as both verbs constitute a separate predicate. This is demonstrated in (3), in which vivía functions as a main verb and is being modified by the gerund trabajando and can be paraphrased as (3a).
(3) Vivía en la finca trabajando con sus padres y hermanos en las labores agrícolas. ‘(He) lived on the farm working with his parents and siblings in agriculture’. (Amaral, 2013a: 106) a. Vivía en la finca y trabajaba con sus padres y hermanos en las labores agrícolas. ‘(He) lived on the farm and worked with his parents and siblings in agriculture’. However, in a similar syntactic organization where a finite verb acts as an auxiliary verb and is combined with a non-finite verb form (i.e. a gerund), a monoclausal predicate is formed. In this case, the finite verb functions only as an auxiliary to the gerund as exemplified in (4), in which vivo llamando does not constitute two separate predicates and, hence, it cannot be paraphrased as (4a).
(4) Viven besándose. ‘They kiss (each other) all the time’
(Amaral, 2013a: 107)
a. *Viven y se besan. ‘They live and kiss each other’. Relying on Abeillé & Godard’s (2010) work, Amaral (2013a, 2013b) presents an extensive list of syntactic tests that can be used to identify whether a periphrasis is monoclausal and biclausal in nature. In the present study, although it is not our purpose to develop an exhaustive syntactic analysis of these periphrases, we offer a brief syntactic characterization of andar and venir + gerund both as biclausal and monoclausal complex predicates following Amaral’s syntactic criteria. For the sake of simplicity and the purposes of illustration, we have chosen to discuss two of the five tests offered in her work: constituent preposing and constituent interpolation.1 Based on these two syntactic tests, we posit that when the verbs of motion, andar and venir, function as auxiliary verbs with an aspectual interpretation and are combined with a gerund, they yield a
. Other syntactic tests included in prior work have been ‘clitic climbing’, ‘intervening adverbials’ and ‘coordination of non-finite verb forms and complements’. For further discussion of these tests, see Amaral (2013a).
Lorena Sainz-Maza Lecanda & Meagan E. Horn
monoclausal predicate. When treating them as monoclausal predicates, their meaning of physical motion is lost and subsequently, specific semantic and pragmatic implications arise. Crucially, it is only the monoclausal complex predicate form that allows the emergence of the semantic and pragmatic implications that will be discussed in the following sections. The first syntactic test that demonstrates this point is constituent preposing. In this test, the gerund that follows the finite verb of motion is only able to be preposed to the beginning of the sentence if it is a biclausal predicate as illustrated in (5) and (6). Additionally, these examples show that when functioning as a biclausal predicate, the meaning of motion from the main verb is maintained. (5) Andaba mirando las olas por la orilla del mar. a. Mirando las olas, andaba por la orilla del mar. ‘Looking at the waves, I used to walk along the shoreline.’ (6) Ellos venían leyendo un libro en el autobús. a. Leyendo un libro, ellos venían en el autobús. ‘Reading a book, they would come on the bus.’ However, if the periphrasis is a monoclausal complex predicate and the gerund is preposed, the result will be ungrammatical because the finite verb of motion is functioning only as an auxiliary verb. In example (7) and (8), it is demonstrated that both andar and venir have lost their syntactic role as a main verb and, subsequently, have lost their original semantic meaning of motion. (7) María anda estudiando para los exámenes cuando puede. a. *Estudiando para los exámenes, María anda cuando puede. ‘Studying for her exams, Mary walks when she can.’ (8) Vengo estudiando para los exámenes desde hace un par de meses. a. *Estudiando para los exámenes, vengo desde hace un par de meses. ‘Studying for my exams, I come since a few months ago.’ The second syntactic constituency test, constituent interpolation, allows us to further demonstrate our point. For this particular test, it is assumed that if a constituent is inserted into a biclausal construction between the finite verb and the gerund, it will not render a sentence ungrammatical. Additionally, for our purposes, (9) and (10) show that the meaning of motion supplied by the finite verbs is preserved. (9) Yo andaba mirando las olas por la orilla del mar. a. Yo andaba por la orilla del mar mirando las olas. ‘I used to walk along the shoreline looking at the waves.’ (10) Ellos venían leyendo un libro en el autobús. a. Ellos venían en el autobús leyendo un libro. ‘They would come on the bus reading a book.’
The semantics and pragmatics of andar and venir + gerund
However, as shown in (11) and (12), an intervening constituent in a monoclausal complex predicate results in an ungrammatical sentence. Moreover, for the purpose of our study, it is important to note that in the following examples, as the finite verb acts only as an auxiliary, the meaning of physical motion is absent. (11) María anda estudiando para los exámenes cuando puede. a. * María anda por la biblioteca estudiando para los exámenes cuando puede. * ‘Mary walks through the library studying for her exams when she can.’ (12) Vengo estudiando para los exámenes desde hace un par de meses. a. * Vengo por la biblioteca estudiando para los exámenes desde hace un par de meses. * ‘I come through the library studying for my exams since a few months ago.’ The data presented above indicates that when both andar + gerund and venir + gerund function as monoclausal complex predicates, they behave significantly differently from their biclausal counterparts. Whereas in the latter, the two verb forms are interpreted as independent predicates, such an analysis cannot be adopted for the former. Similar observations, although in a slightly different manner, have been made in Olbertz (1998) and García-Fernández (2006), who suggest that andar and venir + gerund have periphrastic and non-periphrastic uses. These studies, as well as the present one, make the crucial point that when both andar and venir + gerund behave as a monoclausal complex predicate, the semantic meaning conveyed by each verb of motion is lost. Consequently, we posit that these two periphrases have adopted new underlying meanings. The remainder of this article is intended to address and elucidate what these new underlying meanings are when the semantic meaning of motion is lost. In order to do so, we will first analyze the semantic components of these two monoclausal periphrastic constructions and, following this, a pragmatic approach will be adopted in order to explore the distinctions between venir and andar + gerund on the basis of conversational implicatures.
3. Semantic properties of andar and venir + gerund In this section, we analyze both andar and venir + gerund with regard to their semantic properties. First, based on prior work (Laca, 2004, 2006), we thoroughly examine which type of event-denoting verbs venir can combine with based on the Aktionsart of the verb and then we compare it to andar + gerund and ir + gerund, as described by Laca (2004, 2006). In Section 3.2, we show, in accordance with previous semantic
Lorena Sainz-Maza Lecanda & Meagan E. Horn
characterizations of andar + gerund, that venir + gerund is also a frequentative pluractional operator. Finally, in Section 3.3, we propose a difference between the pluractionals andar and venir + gerund concerning the degree of frequency and the nature of the event repetitions entailed by each periphrastic construction.
3.1 Eventuality type restrictions Periphrastic constructions of the kind analyzed in the present study, where the verb of motion has lost its semantic meaning giving rise to a monoclausal complex predicate, have been moderately investigated on semantic grounds. Laca (2004, 2006), also discussed in Squartini (1998), analyzed the aspectual periphrases andar and ir + gerund. In her study, she suggests that andar and ir + gerund are subject to specific selectional restrictions. In other words, they cannot appear with every type of eventuality. She argues that ir and andar + gerund both generally reject statives and achievements. (13) ??Iba/??andaba siendo víctima de una alucinación. [stative] (Laca 2006: 3) ‘She was suffering a hallucination.’ (14) ??Vas/??andas cometiendo el peor error de tu vida. [achievement] (Laca 2006: 3) ‘You are making the worst mistake of your life.’ Whereas it is clear to us that ir and andar + gerund constructions cannot combine with verbs denoting achievements, the infelicity of andar with regard to stative verbs is certainly questionable. We disagree with Laca (2004, 2006) in that andar disallows co-occurrence with stative verbs. Our judgments appear to be shared by Squartini (1998), who provides felicitous examples of andar co-occurring with stative verbs such as querer ‘want’ in Luis todavía anda queriendo comprar el coche ‘Luis still wants to buy the car’ (1998: 263). In addition, according to Laca (2004, 2006), ir + gerund does not easily combine with verbs denoting activities or processes, whereas andar + gerund does. Moreover, while ir + gerund combines with degree achievements, andar + gerund does not.2 (15) María #iba/andaba trabajando. [activity/process] ‘Mary was working.’
(Laca, 2006: 3)
(16) El río iba/#andaba creciendo. [degree achievement] ‘The river was rising.’
(Laca, 2006: 3)
. Squartini suggests that ir + gerund can co-occur with statives and activities or processes under special circumstances and meanings. The reader is referred to Squartini (1998) for an elaborate discussion of this periphrastic construction.
The semantics and pragmatics of andar and venir + gerund
As for accomplishments, Laca (2006) points out that both ir and andar + gerund can co-occur with what traditionally have been described as verbs denoting accomplishments: (17) María fue/anda leyendo La Guerra y la Paz. [accomplishment] (Laca 2006: 5) ‘Mary read/Mary is reading War and Peace.’ From these observations, among others, Laca (2004, 2006) proposes that both aspectual periphrases convey pluractional meanings, that is to say both monoclausal predicates indicate a multiplicity of events. More specifically, she puts forth that ir + gerund is an incremental pluractional, “in as far as it expresses a monotonic mapping between successive events and successive position on some scale” (2004: 4). On the other hand, andar + gerund is specified as a frequentative pluractional, “which can involve the serial repetition of a complete event (as in ‘to tell a joke over and over again’) or subevents associated to temporal gaps in the development of a single event (as in ‘to tell a story in installments, on and off ’)” (2004: 4). The present study adds a third periphrastic construction to the picture developed by Laca (2004, 2006). We take as point of departure the behavior of andar + gerund and compare it to that of venir + gerund. In terms of its selectional restrictions based on the Aktionsart of the verb, Squartini (1998), and also Fernández-García (2006) have pointed out that venir + gerund seems to allow verbs denoting states and activities. Our data comparing the behavior of venir with that of andar in (18) and (19) suggest that this is in fact the case: (18) Ana anda/viene siendo víctima de un soborno desde hace días.3 [stative] ‘Ana has been being blackmailed for days.’ (19) Juan anda/viene riéndose de Pedro desde hace horas. [activity/process] ‘John has been laughing at Pedro for hours.’ As far as accomplishment denoting verbs, Fernández-García (2006) notes that this construction can appear with those predicates, though there is uncertainty with regards to the accomplishment of the event. We believe that this intuition is accurate as is illustrated in (20), wherein it is unresolved whether the mansion has been finished.
. Venir + gerund requires a past referent which should be understood and made available in the discourse context or in the sentence itself in order to be felicitous; andar + gerund, on the other hand, does not. However, for the sake of comparison, we add a past referent to sentences with andar as well given that its use is shown not to be infelicitous. For an extensive discussion of the Reference Time which venir + gerund is subject to, see Fernández de Castro (1990), Squartini (1998) and García-Fernández (2006).
Lorena Sainz-Maza Lecanda & Meagan E. Horn
(20) Juan anda/viene construyendo una mansión desde hace unos años. [accomplishment] ‘John has been building a mansion for a few years.’ New to the study of this aspectual periphrasis, we believe, is the observation that venir + gerund cannot be accompanied by verbs denoting achievements as shown in (21): (21) Ana #anda/#viene encontrando la cartera desde hace un par de horas. [achievement] ‘Ana has been looking for her purse for a couple of hours.’ These observations, together with Laca’s (2004, 2006), are summarized in Table 1. Table 1. Summary of the selectional restrictions of ir, andar and venir + gerund Ir + gerund
Andar + gerund
Venir + gerund
Laca (2004, 2006)
Laca (2004, 2006)
Present study
Present study
NO
NO
YES
YES
Activity
NO
YES
YES
YES
Accomplishment
YES
YES
YES
YES
Achievement
NO
NO
NO
NO
Stative
On the surface, from these distributional properties it seems that andar and venir + gerund pattern identically. As shown in Table 1, we differ from Laca (2004, 2006) with respect to the behavior of the andar + gerund construction and its compatibility with stative verbs. Although there seems to be a one-to-one correspondence between andar and venir + gerund regarding their selectional restrictions, it will be shown in the next sections that these aspectual periphrases differ in the degree of frequency with respect to which each event is repeated as well as the implicatures that each periphrasis is related to.
3.2 Andar and venir + gerund as pluractionals Event plurality has been well-established as an intrinsic characteristic of periphrastic constructions involving verbs of motion followed by a gerund (see Section 3.1). As for venir + gerund, we take cases like (22) as evidence to conclude that venir + gerund is also a pluractional and, therefore, the event is repeatable during a certain time interval. (22) Vengo escuchando la misma canción en la radio desde hace días. ‘I have been listening to the same song on the radio for the past few days.’ In addition, we further posit that, in accordance with the semantic description of andar + gerund proposed by Laca (2004, 2006), venir + gerund is also a frequentative pluractional as it can involve a serial repetition of a complete event.
The semantics and pragmatics of andar and venir + gerund
(23) Ando yendo al psicólogo de vez en cuando. ‘I have been going to the psychiatrist occasionally.’ (24) Vengo escuchando la misma canción en la radio una y otra vez desde hace días. ‘I have been listening to the same song on the radio over and over again for the past few days.’ In (23), the event of going to the psychologist expressed by means of andar + gerund is not a one-time event, but a repeated event and, similarly, the sentence in (24) with venir + gerund also conveys that the event of listening to the same song on the radio has not happened just once, but on multiple occasions. This multiplicity of events can be further shown by the infelicitous use of each periphrastic construction with the temporal modifier una vez ‘once’ as described in (25) and (26): (25) # Ando yendo al psicólogo. He ido una vez. # ‘I have been going to the psychiatrist. I have gone once.’ (26) # Vengo escuchando la misma canción en la radio desde hace días. La he escuchado una vez. # ‘I have been listening to the same song on the radio for the past few days. I have listened to it once.’ Thus, it is clear that both andar and venir + gerund convey a frequentative pluractional meaning. However, in the following section we explore in more detail the pluractional meaning conveyed by each periphrastic construction and show that there are certain differences in the degree of event frequency entailed by each monoclausal complex predicate.
3.3 Degree of frequency entailed by venir and andar + gerund When looking more closely at (23) and (24), here repeated as (27) and (28), we observe that the temporal expressions de vez en cuando ‘once in a while’ and una y otra vez ‘over and over again’ under the scope of andar + gerund and venir + gerund respectively do not express the same degree of frequency: (27) Ando yendo al psicólogo de vez en cuando. ‘I have been going to the psychiatrist occasionally.’ (28) Vengo escuchando la misma canción una y otra vez desde hace días. ‘I have been listening to the same song over and over again for the past few days.’ The co-occurrence of andar + gerund with temporal expressions denoting low frequency suggests that the amount of time between each repetition of the event is longer with andar + gerund, whereas the opposite pattern is observed with venir + gerund as shown by Figure 1.
Lorena Sainz-Maza Lecanda & Meagan E. Horn
ANDAR + GERUND: “Ando yendo al psicólogo de vez en cuando.” Event = Going to the psychologist
ev1
ev2
ev3
ev4
ev5
VENIR + GERUND: “Vengo escuchando la misma canción una y otra vez desde hace días.” Event = Listening to a song
ev1 ev2 ev3 ev4 ev5 ev6 ev7 ev8 ev9 ev10
Figure 1. Schematic representation of the periods of time between each repetition of an event
This suggests that andar + gerund entails a lower frequency of the event in question than venir + gerund, which entails high frequency of the event under discussion. Therefore, we posit that each periphrasis differs in the entailment relationship that arises from their pluractional meaning. More specifically, we suggest that andar + gerund entails fewer repetitions of an event with longer periods of time between each repetition, hence a low frequency of occurrence of the event; whereas more repetitions with shorter periods of time between each repetition is entailed by venir + gerund, resulting in a high frequency of occurrence of the event. These entailment relationships are shown in (29) and (30) respectively. (29) Ando llamando a mi hermano desde las 9 de la mañana. ‘I have been calling my brother since 9 o’clock this morning.’ ⊨ He llamado a mi hermano unas cuantas veces esta mañana. ‘I have called my brother a few times this morning.’ (30) Vengo llamando a mi hermano desde las 9 de la mañana. ‘I have been calling my brother since 9 o’clock this morning.’ ⊨ He llamado a mi hermano muchísimas veces esta mañana. ‘I have called my brother many, many times this morning.’ An utterance of the sentence in (29) entails that the event of calling has been repeated infrequently, whereas (30) entails that the event of calling has been repeated many times. The specific entailments of each periphrastic construction are schematically illustrated in Figure 2. Unlike Figure 1, which exemplifies the degree of frequency of each periphrastic construction accompanied by temporal expressions, Figure 2 illustrates the actual difference between the two constructions arising exclusively from their specific entailments. Further evidence for the claim that andar + gerund entails lower frequency of repetition comes from the fact that andar + gerund is compatible with adverbials
The semantics and pragmatics of andar and venir + gerund
ANDAR + GERUND: “Ando llamando a mi hermano desde las 9 de la mañana.” Event = Calling my brother
ev1
ev2
ev3
ev4
ev5
VENIR + GERUND: “Vengo llamando a mi hermano desde las 9 de la mañana.” Event = Calling my brother
ev1 ev2 ev3 ev4 ev5 ev6
ev7 ev8 ev9 ev10
Figure 2. Schematic representation of the periods of time between each repetition by construction
denoting low cardinality of occasions as mentioned in the introduction of Section 3.2. In (31a), andar + gerund is compatible with de vez en cuando ‘once in a while’, whereas the same adverbial co-occurring with venir + gerund is not (31b). (31) Pedro has been feeling pretty depressed and has been seeing a psychologist lately. One day, he runs into a close friend and this friend asks him, ¿qué tal te encuentras? ‘how are you doing?’, to which he responds: a. Algo mejor, ando yendo al psicólogo de vez en cuando. ‘A little better, I have been going to the psychiatrist occasionally.’ b. # Algo mejor, vengo yendo al psicólogo de vez en cuando desde hace un par de semanas. # ‘A little better, I have been going to the psychiatrist occasionally for the past few months.’ Furthermore, the fact that venir + gerund entails higher frequency of repetition of the event is supported by its felicitous use in a context that demands high frequency of occurrence of the event (32b), where andar + gerund is not compatible (32a). (32) I can’t believe this! I just found out that my boyfriend cheated on me! I’m trying to reach him on the phone, but he’s not answering. When my roommate sees me upset, she asks, ¿qué haces? ‘what are you doing?’, to which I respond: a. # Ando llamando a mi novio porque me ha puesto los cuernos. # ‘I have been calling my boyfriend because he cheated on me.’ b. Vengo llamando a mi novio desde hace un par de horas porque me ha puesto los cuernos. ‘I have been calling my boyfriend for the past couple of hours because he cheated on me.’
Lorena Sainz-Maza Lecanda & Meagan E. Horn
Assuming that the agent cares for her boyfriend, the context in (32) demands high repetition of event and, therefore, only venir + gerund is felicitous. On a final note, the degree of frequency should be understood as being relative to the maximal potential of the repetition of the event in question. For example, in a context like (33) or (34), where the agent of the action has a standing appointment with a psychiatrist once a week, the maximal repetition of the event would be weekly. However, the context in (33) requires the use of andar + gerund as the agent does not maximize the potential repetitions, resulting in lower frequency of repetition. Relating back to Fernández de Castro (1990), the agent in this case only attends a subset of the possible meetings, in an erratic and disorganized fashion. On the other hand, in (34), venir + gerund is required as the agent repeats the event in question with maximal potential by attending every meeting without skipping even one, resulting in a higher degree of frequency. (33) I have a standing appointment with my psychiatrist on Mondays at 3pm. She doesn’t charge a cancellation fee, so some weeks I go and other weeks I skip the appointment. a. Desde hace unos meses, ando yendo al psicólogo de vez en cuando. ‘For the past couple months, I have been going to the psychiatrist occasionally.’ b. # Desde hace unos meses, vengo yendo al psicólogo de vez en cuando. # ‘For the past couple of months, I have been going to the psychiatrist occasionally.’ (34) I have a standing appointment with my psychiatrist on Mondays at 3pm. She charges a high cancellation fee, so I make sure that I make every appointment. a. # Desde hace unos meses, ando yendo al psicólogo sin faltar a ninguna cita. # ‘For the past few months, I have been going to the psychiatrist without missing a single appointment.’ b. Desde hace unos meses, vengo yendo al psicólogo sin faltar a ninguna cita. ‘For the past few months, I have been going to the psychiatrist without missing a single appointment.’ This frequency relationship for each periphrasis is depicted in Figure 3, where the vertical dashes represent potential repetitions on a time continuum while the check marks represent the agent’s actual realization of the potential repetition. To summarize, it has been shown that there are selectional restrictions on the types of verbs that may occur in the gerund determined by the Aktionsart of the verb. While it has been shown that both andar and venir + gerund respond to the same eventuality type restrictions and that venir + gerund, similar to andar + gerund, acts as frequentative pluractional, it has been proposed that these frequentative
The semantics and pragmatics of andar and venir + gerund
ANDAR + GERUND: “Desde hace unos meses, ando yendo al psicólogo de vez en cuando.” Event = Going to weekly appointments at the psychiatrist
✓ ✓ ✓ ✓ ✓ VENIR + GERUND: “Desde hace unos meses, vengo yendo al psicólogo sin faltar a ninguna cita.” Event = Going to weekly appointments at the psychiatrist
✓
✓
✓
✓
✓
✓
✓
✓
Figure 3. Schematic representation of the realization of potential repetitions by construction
luractionals do not entail the same degree of frequency: first, andar + gerund entails p fewer repetitions of an event with longer periods of time between, thus, low frequency for a certain time interval, whereas venir + gerund entails more repetitions of an event with shorter time periods between and, therefore, a higher degree of frequency; and second, when a clear maximum quantity of repetition realizations is present, the use of andar + gerund entails fewer repetitions than the maximal amount possible, whereas venir + gerund entails repetition of the event to its maximal potential.4
4. Pragmatic properties of andar and venir + gerund As a result of both the monoclausal nature and frequentative pluractionality of andar and venir + gerund, we propose that the pragmatic properties of each allow for . An anonymous reviewer brought to our attention the following examples where andar appears to be felicitous with a high frequency event:
(a)
Ando teniendo pesadillas muy seguido desde hace un par de días. ‘I have been having nightmares very often for the past couple of days.’
(b)
Vengo teniendo pesadillas muy seguido desde hace un par de días. ‘I have been having nightmares very often for the past couple of days’.
We think that these two examples do not fall under the same scope as the current analysis in that the subject of these two sentences are experiencers, instead of agents. We agree with the reviewer that both andar and venir are felicitous in the above case, despite the similarity in frequency. Nevertheless, we suggest that the difference in these cases lies in the impact of the action on the experiencer. That is, in both the above sentences, the speaker is experiencing frequent nightmares. However, the speaker’s choice to use andar in (a) entails a less intense effect of the event of having nightmares. On the other hand, the use of venir in (b) entails a significantly more impactful effect on the speaker. Be this as it may, no conclusions will be drawn in this respect until further investigation is conducted.
Lorena Sainz-Maza Lecanda & Meagan E. Horn
a clear differentiation of their felicitous uses. Fernández de Castro (1990) provides preliminary insight into each periphrasis analyzed here by stating that venir + gerund functions as a precise mechanism for quantifying temporal relations while andar + gerund embodies a multiplicity of erratic and disorganized approximations. In this section, using these emerging insights by Fernández de Castro (1990) as an impetus for a pragmatic investigation, we further develop them and subsequently posit two related, yet contrastive, generalized conversational implicatures that rise from each periphrases, following the Gricean notion of implicature (Grice, 1975): first, the agent’s intention towards the accomplishment of the goal; and second, the agent’s level of engagement in the action. Additionally, since cancellability is an inherent property of conversational implicatures (Huang, 2007; Birner, 2013, inter alia), it will be shown that each implicature can be cancelled for each periphrasis. The felicitous cancellation of each implicature will provide compelling evidence suggestive of the fact that the agent’s intention towards the accomplishment of the goal and the level of engagement should be understood exclusively as generalized conversational implicatures.
4.1 Agent’s intention towards the accomplishment of a goal In this section, we investigate the generalized conversational implicatures generated by andar and venir + gerund concerning the agent’s intention towards the accomplishment of a particular goal. First, we provide data suggestive of the fact that the use of andar + gerund implicates the agent’s indifference about the accomplishment of a goal; subsequent to this, we show that a sentence containing venir + gerund implicates an agent’s definitive commitment to accomplishing said goal. (35) Juan, a laid-back, billionaire architect, won the lottery ten years ago. He decided that he might as well build a house in order to invest money in a new project. Yesterday, he traveled to visit a friend that he has not seen in years. As he gets off his private jet, his friend asked him, ¿a qué te has dedicado durante los últimos años? ‘what have you been working on in the last few years?’, to which he responded: a. Nada nuevo. Ando construyendo una mansión nueva desde hace unos años. ‘Nothing new. I have been building a new mansion for the past few years.’ b. # Nada nuevo. Vengo construyendo una mansión nueva desde hace unos años. # ‘Nothing new. I have been building a new mansion for the past few years.’ In a contextual situation like the one in (35), only the utterance of a sentence like (35a) is felicitous. The situation describes a relaxed man who is not in a hurry to finish building the house anytime soon. Since the implicature generated from an utterance of
The semantics and pragmatics of andar and venir + gerund
(35a) is that of indifference towards the accomplishment of the goal, the sentence with andar + gerund is felicitous in this particular context. The agent in this sentence has no necessity of building the house right away and, therefore, he is not eager to accomplish said goal. On the other hand, (35b) contains venir + gerund and thus is infelicitous because the conversational implicature attached to it is not of indifference; instead, an utterance involving venir + gerund is only felicitous in a context demanding a clear intention to accomplish a goal, as is demonstrated in (36). (36) Pedro lost everything in a terrible storm two months ago. His house is gone and his children and wife have nowhere to live. They moved into Pedro’s parents’ house after the tragedy, but he has been working desperately to build a new home for his family. Last week, Pedro appeared on a TV show to tell the world his story in hopes that someone would donate money. The interviewer asked him, ¿cómo estás haciendo frente a esta situación tan difícil? ‘how are you doing in light of such a difficult situation?’, to which he responded: a. # ¡Imagínate! Estoy desolado. Ando construyendo la nueva casa desde hace unos meses. # ‘Imagine it! I am devastated. I have been building the new house for the past few months.’ b. ¡Imagínate! Estoy desolado. Vengo construyendo la nueva casa desde hace unos meses. ‘Imagine it! I am devastated. I have been building the new house for the past few months.’ The context in (36) describes the situation of a man who is desperately trying finish building his new home in order to move out of his parents’ house. An utterance of the sentence containing venir + gerund, as in (36b), is felicitous under this context, which shows that venir + gerund does in fact implicate the agent’s definite intention to accomplish the goal. Accordingly, since andar + gerund conversationally implicates the agent’s indifference towards the accomplishment of the goal, (36a) is not felicitous. Further evidence supporting the contrastive conversational implicatures that arise from andar and venir + gerund is shown in (37) and (38). (37) Andar + gerund: Indifference towards the accomplishment of a goal a. Nada nuevo. Ando construyendo una mansión desde hace unos meses con mucha calma y no me preocupa cuándo acabarla. ‘Nothing new. I have been building a mansion for the past few months very calmly and I’m not worried about when I finish it.’ b. # Nada nuevo. Ando construyendo una mansión desde hace unos meses para acabarla cuanto antes. # ‘Nothing new. I have been building a mansion for the past few months in order to finish it as soon as possible.’
Lorena Sainz-Maza Lecanda & Meagan E. Horn
(38) Venir + gerund: Definite intention towards the accomplishment of a goal a. ¡Imagínate! Estoy desolado. Vengo construyendo la nueva casa desde hace unos meses sin parar para acabarla cuanto antes. ‘Imagine it! I am devastated. I have been building the new house non-stop for the past few months in order to finish it as soon as possible.’ b. # ¡Imagínate! Estoy desolado. Vengo construyendo la nueva casa desde hace unos meses con mucha calma. # ‘Imagine it! I am devastated. I have been building the new house for the past few months very calmly.’ The felicitous use of andar + gerund in (37a) and infelicitous use in (37b) suggests that the accomplishment of a goal is not of paramount importance to the agent since the agent is not concerned about how long it will take him to finish the house. With regard to venir + gerund, on the other hand, the felicitous utterance of a sentence like (38a) and the infelicitous one in (38b) indicates that it is the agent’s intent to finish building the house and, thus, he is highly committed to the accomplishment of that particular goal. As mentioned in the introduction of Section 4, it is a well-established fact that generalized conversational implicatures can be cancelled within the same situational context. As illustrated in (39), the implicature associated with andar + gerund can be cancelled, wherein the accomplishment of the goal is of paramount importance to the agent. Similarly, (40) shows the cancellation of the definite intention of the agent to finish building the house. (39) Ando construyendo una mansión nueva desde hace unos años, pero durante este tiempo siempre he tenido claro que voy a acabarla. ‘I have been building a new mansion for the past few years, but during that time it has always been clear that I am going to finish it.’ (40) Vengo construyendo la nueva casa desde hace unos años, pero nunca he tenido muy claro que vaya a terminarla. ‘I have been building the new house for the past few years, but it has never been very clear to me that I am going to finish it.’ From this set of data, we draw the conclusion that andar and venir + gerund differ from each other in the kind of implicature that they generate. Andar + gerund implicates that the agent is indeterminate towards the accomplishment of a goal, potentially only approximating it, while venir + gerund conversationally implicates the agent’s determined intention to accomplish a particular goal. Overall, though both andar and venir + gerund share a frequentative pluractional meaning, they are different with respect to their pragmatic distribution. In the remainder of this chapter, we look at an additional conversational implicature that we believe is closely related to the agent’s intention towards the completion of a particular goal.
The semantics and pragmatics of andar and venir + gerund
4.2 The level of agent engagement The second pragmatic property that we posit arises when andar and venir + gerund form a monoclausal predicate is the level of agent engagement. We suggest that andar + gerund conversationally implicates low agent engagement, whereas venir + gerund conversationally implicates high engagement on the part of the agent. This pragmatic distinction is illustrated in (41) and (42) by the felicitous and infelicitous uses of each periphrasis in a given context: (41) My father is obsessed with the Betis soccer team and he is an unconditional supporter, whereas I’m not. Sometimes, though, I like to spend time with him so I join him in going to a game. a. Ando asistiendo a los partidos del Betis desde hace años. ‘I have been attending the Betis games for years.’ b. # Vengo asistiendo a los partidos del Betis desde hace años. # ‘I have been attending the Betis games for years.’ The above context in (41) describes a situation wherein the agent is not an unconditional supporter of the Betis soccer team and only sometimes attends games when he wants to spend time with his father. Thus, in this particular scenario, the context suggests a relatively low engagement on the part of the agent in regards to the attendance of the games. As a result, the use of the andar + gerund periphrasis in (41a) is the only felicitous response since (41b) does not contain such an implicature. In contrast, the context in (42) explains that the agent is an unconditional supporter of the Betis soccer team and that under no circumstances would he miss any of the team’s games. From this context, it follows that only (42b) would be felicitous since venir + gerund implicates a high level of agent engagement. Consequently, (42a) is not felicitous because the context requires that the agent is actively engaged in the performance of the action. (42) I’m a member of the Betis soccer team and I’m an unconditional supporter of the team. Every season, soccer and Betis are the only two things I can think about and I never miss a game. a. # Ando asistiendo a los partidos del Betis desde hace años. # ‘I have been attending the Betis games for years.’ b. Vengo asistiendo a los partidos del Betis desde hace años. ‘I have been attending the Betis games for years.’ Further evidence of this distinction between both periphrastic expressions can be provided through the addition of linguistic expressions demanding high engagement to the utterance context itself. For instance, without providing any situational context to the utterance in (43), the inclusion of an expression such as sin perderme ni uno ‘without missing even one’ is only felicitous in (44), which supports our hypothesis
Lorena Sainz-Maza Lecanda & Meagan E. Horn
that venir + gerund generates the implicature that the agent is highly engaged in a particular situational action, in this case, that of going to every soccer game. (43) # Ando asistiendo a los partidos del Betis sin perderme ni uno desde hace años. # ‘I have been attending the Betis games without missing even one for years.’ (44) Vengo asistiendo a los partidos del Betis sin perderme ni uno desde hace años. ‘I have been attending the Betis games without missing even one for years.’ Similarly, the same effect can be achieved for andar + gerund with an expression that indicates apathy or indifference, such as algo desganado ‘without much enthusiasm’. As shown in (45) and (46), the insertion of this adjective phrase denoting indifference is only felicitous in (45) which strengthens the hypothesis that andar + gerund implicates passive engagement on the part of the agent. (45) Ando asistiendo a los partidos del Betis algo desganado desde hace años. ‘I have been attending the Betis games without much enthusiasm for years.’ (46) # Vengo asistiendo a los partidos del Betis algo desganado desde hace años. ‘I have been attending the Betis games without much enthusiasm for years.’ As discussed previously, a prominent feature of conversational implicatures is that they are cancellable. Thus, the implicature of low agent engagement arising from andar + gerund can effectively be cancelled by supplying additional information about the agent. This is shown in example (47), where the fact that the agent acknowledges that he is an actively engaged fan of the team cancels the implicature of low engagement. (47) Ando asistiendo a los partidos del Betis algo desganado desde hace años, pero esto no quiere decir que no sea forofo del Betis. ‘I have been attending the Betis games without much enthusiasm for years, but that doesn’t mean that I am not a Betis fan.’ In contrast, in the sentence (48) below, the implicature of high agent engagement arising from venir + gerund is cancelled by the additional information that the agent is not always enthused, thus not highly engaged, while attending the soccer games. (48) Vengo asistiendo a los partidos del Betis sin perderme ni uno desde hace años, pero tampoco es que me entusiasme tanto ir. ‘I have been attending the Betis games without missing even one for years, but I am not particularly excited about going.’ Hence, to conclude, through an examination of the contexts in (41) and (42), it is evident that the conversational implicatures arising from each periphrasis are different and that their felicitous use is licensed based on differing pragmatic implicatures. Specifically, we propose that andar + gerund is used to conversationally implicate a low level of agent engagement while venir + gerund implicates a high level of engagement.
The semantics and pragmatics of andar and venir + gerund
5. Conclusion Throughout the course of this paper, the aspectual periphrases andar and venir + gerund have been analyzed syntactically, semantically and pragmatically in order to provide a clear understanding of their linguistic similarities and differences. First, two syntactic tests were provided to demonstrate that both periphrases can be either monoclausal or biclausal in nature. It has been then exemplified that when both andar and venir + gerund act as a monoclausal complex predicates, the semantic meanings of motion are lost and instead both act as frequentative pluractionals, which denotes a multiplicity of events. However, it has been demonstrated that although both periphrases are frequentative pluractionals, they do not entail the same degree of event frequency. Specifically, andar + gerund entails a lower degree of event frequency whereas venir + gerund entails a higher degree of event frequency. Finally, a pragmatic analysis revealed that the two conversational implicatures that arise from both andar and venir + gerund provide a differentiation between the two periphrases and that their felicitous use is licensed based on differing pragmatic conditions. Specifically, we have proposed that andar + gerund implicates an indifference towards the accomplishment of a particular goal and a low level of agent engagement while venir + gerund implicates a determination towards the completion of a goal and a high level of engagement. Whether these semantic and pragmatic features are shared by other varieties of Spanish, in addition to Peninsular Spanish, is left open for future research. Moreover, it should be further explored how the characteristics of these monoclausal complex predicates are to be understood when the subject of an utterance takes a different semantic role and, for instance, behaves as the experiencer of a given action rather than the agent. All in all, we contend that although andar and venir + gerund may function similarly in a syntactic sense, a combination of both their semantic and pragmatic features is crucial in order to fully understand their meanings and usage in discourse.
References Abeillé, A., & Danièle, G. (2010). Complex predicates in the Romance languages. In A. Abeillé & G. Danièle (Eds.), Fundamental issues in the Romance languages (pp. 107–170). Stanford, CA: CSLI. Amaral, P. (2013a). The pragmatics of number: The evaluative properties of vivir + V[Gerund]. Journal of Pragmatics, 51, 105–121. DOI: 10.1016/j.pragma.2013.03.004 Amaral, P. (2013b). On how to live and keep dying. In C. Howe, S. Blackwell, & M. Quesada (Eds.), Selected proceedings of the 15th Hispanic Linguistics Symposium, (pp. 269–280). Somerville, MA: Cascadilla Proceedings Project. Birner, B.J. (2013). Introduction to pragmatics. Chichester: Wiley-Blackwell. DOI: 10.1111/modl.12168
Lorena Sainz-Maza Lecanda & Meagan E. Horn Fernández de Castro, F. (1990). Las perífrasis verbales en español: Comportamiento sintáctico e historia de su caracterización. Oviedo: Departamento de Filología española. García Fernández, L. (2006). Diccionario de perífrasis verbales. Madrid: Gredos. DOI: 10.1515/zrp-2012-0016 Grice, H.P. (1975). The logic of conversation. In P. Cole & J.L. Morgan (Eds.), Syntax and semantics 3: Speech acts (pp. 41–58). New York, NY: Academic Press. DOI: 10.1017/s0022226700005296 Huang, Y. (2007). Pragmatics. Oxford: Oxford University Press. DOI: 10.1108/09504121311320152 Laca, B. (2002). Romance “aspectual” periphrases: Eventuality modification versus “syntactic” aspect. In J. Guéron & J. Lecarme (Eds.), The syntax of time (pp. 425–440). Cambridge, MA: The MIT Press. Laca, B. (2004). Progressive, pluractionals and the domains of aspect. In O. Crouzet, H. Demidache, & S. Wauquier (Eds.), Domain(e)s, Proceedings of the Journée de Linguistique de Nantes (pp. 87–92). Laca, B. (2006). Indefinites, quantifiers, and pluractionals. What scope effects tell us about event pluralities. In S. Vogeleer & L. Tasmowski (Eds.), Non-definiteness and Plurality (pp. 191–217). Amsterdam: John Benjamins. DOI: 10.1075/la.95.10lac Olbertz, H. (1998). Verbal periphrases in a functional grammar of Spanish. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110820881 Squartini, M. (1998). Verbal periphrases in Romance: Aspect, actionality, and grammaticalization. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110805291
Sequence of tenses in complementation structures Lexical restrictions and effects on language acquisition* Rui Marques1, Purificação Silvano2, Anabela Gonçalves1 & Ana Lúcia Santos1 1Universidade
de Lisboa / 2Universidade do Porto
In this paper we discuss the combinations of tenses in main and complement clauses of European Portuguese, focusing on the issue that restrictions on the tenses allowed in complement clauses are observed with some predicates but not with others. We show that these lexical restrictions are independent of the mood occurring in the complement clause, though an integrated analysis of mood and tense may be achieved. The proposal is made that the observed lexical restrictions on embedded tenses have a semantic basis and follow from the fact that Portuguese is an SOT-language, that is, a language where embedded tenses have semantic import. A preliminary investigation is conducted on the sequences of tenses produced at early stages of language acquisition. Keywords: sequence of tenses; temporal dependence; complement clauses; language acquisition
1. Introduction Sequences of tenses in the matrix and the embedded clause have been studied within the frameworks of Formal Semantics and Generative Syntax. In the former, analyses on the subject have been focused on deriving the correct interpretation of tenses in embedded clauses. We do not focus on this issue here and, following Gennari (2003), we assume
* This work was developed at the Centro de Linguística da Universidade de Lisboa (CLUL; PEst-OE/LIN/UI0214/2013), within the project Complement Clauses in the Acquisition of Portuguese (CLAP) (PTDC/CLE-LIN/120897/2010) sponsored by Fundação para a Ciência e a Tecnologia (FCT).
doi 10.1075/ihll.4.04mar © 2015 John Benjamins Publishing Company
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
that in SOT languages such as Portuguese the interpretation of embedded tenses does not involve particular rules that apply to embedded, but not to non-embedded, tenses. In particular, the present tense has utterance time (t0) as its temporal perspective point (TPpt)1 whether it occurs in embedded on non-embedded clauses. We will instead consider a point that has been observed in several works of Generative Syntax: the fact that (in Portuguese and other languages) some verbs impose restrictions on the tenses allowed in their complement clauses, contrary to other verbs. This is shown by examples (1)–(3), from European Portuguese (henceforth EP), the language that we focus on. We observe that with verbs like querer (‘to want’), if the main clause is in the present tense, the complement clause needs to have the present as well, as shown in (1), while if the main verb is inflected in the past, only a past tense is allowed in the complement, as shown by (2). By contrast, if the main verb is a declarative verb, like dizer (‘to say’), any tense can occur in the complement clause, regardless of the tense of the main clause, as shown by (3): (1) Ele quer que a Ana {esteja / *estivesse} he want-PRES that the Ana {be-PRES-SUBJ / *be-PAST-IMPERF-SUBJ } em casa. at home
‘He wants Ana to be home.’ (2) Ele quis que a Ana {*esteja / estivesse} he want-PAST that the Ana {*be-PRES-SUBJ / be-PAST-IMPERF-SUBJ} em casa. at home
‘He wanted Ana to be home.’ (3) Ele {diz/disse} que a Ana {está / estava / he {say-PRES/say-PAST} that the Ana {be-PRES-IND / be-PAST-IMPERF-IND / esteve / estará} em casa. be-PAST-PERF-IND / be-FUT-IND} at home
‘He {says/said} that Ana {is/was/will be} home. In this paper, we will take into account such contrasts as the ones in (1)–(3) aiming at (i) contributing to a better understanding of the facts that rule sequences of tenses in sentences with finite complementation, and (ii) showing that the evaluation of child productions involving the combination of tenses suggests avoidance of temporal dependence in the first stages of initial production of finite complement clauses.
. Temporal Perspective Point (Kamp & Reyle, 1993) is the time interval from which the situation is viewed or perspectivized.
Sequence of tenses in complementation structures
2. L exical restrictions on sequences of tenses and the hypothesis of temporal dependence In the framework of Generative Syntax, contrasts involving the combination of tenses have been mainly explained by a lexical property of verbs that select for a complement clause, which we can roughly call temporal dependence (see, among others, Picallo, 1984, for Spanish; Raposo, 1987; Ambar, 1992, for EP). Approaches that rely on this property generally claim that some verbs require the presence of a dependent tense in their complement, the infinitive and the subjunctive being dependent tenses, contrary to the indicative, which would be specified for tense features. According to this, in indicative clauses, C is specified for T-features and CP is a temporal domain independent from the matrix; this would be the case of complement clauses of epistemic, (some) doxastic and declarative verbs (e.g. the equivalents of know, think and say), that rule the indicative. On the contrary, in subjunctive clauses, C is unspecified for T-features, and CP is a temporal domain dependent from the matrix tense. This would be the case of volition verbs (e.g. the equivalent of want), that rule the subjunctive. Such approaches lead to the prediction that restrictions on sequences of tenses are observed only in structures with subjunctive complements (as in (2)), indicative complements being free (as in (3)). Moreover, they predict obligatory disjoint reference of the subjects in subjunctive complement structures, a phenomenon known as obviation (as in (4)), and optional disjoint reference of the subjects in indicative complement structures (as in (5)): (4) Elei quer que [ ]*i/j chegue a tempo. he wants that [ ]*i/j arrive-SUBJ at time ‘He wants that she/he arrives on schedule.’ (5) Elei pensa que [ ]i/j chega a tempo. he thinks that [ ]i/j arrive-IND at time ‘He thinks that he arrives on schedule.’ According to the syntactic analyses mentioned above, if subjunctive complements are linked to the tense of the main clause, the subjects cannot be co-referent because this would violate principle B of Binding Theory. However, there are some problems with these classical analyses. First, the concept of tense dependence needs a more accurate definition. In fact, it seems reasonable to assume that there is temporal dependence when the temporal perspective point (TPpt) of the embedded tense is the time interval occupied by the situation described by the main clause. Thus, if the embedded tense has utterance time (t0) as its TPpt, main and embedded clauses form two temporal domains and there is no temporal dependence (Silvano, 2002; Gonçalves, Cunha, & Silvano, 2010). But, given this, one might observe that temporal dependence is not lexically conditioned, contrary to what
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
is proposed by Raposo (1987), Ambar (1992), and others. In fact, the same main verb can enter in constructions with temporal dependence (see (6a) and (7a)), as well as in constructions with temporal independence (see (6b) and (7b)), a fact also remarked by Kempchinsky (1990) for Spanish: (6) a. Ele decidiu que ia ao cinema com he decide-PAST that go-PAST-IMPERF-IND to-the cinema with a Rita. the Rita
‘He decided that he would go to the cinema with Rita.’
b. Ele decidiu que vai/irá ao cinema com he decide-PAST that go-PRES-IND/go-FUT-IND to-the cinema with a Rita. the Rita
‘He decided that he will go to the cinema with Rita.’
(7) a. Ele lamentou que estivéssemos atrasados. he regret-PAST that [we] be-PAST-SUBJ late
‘He regretted that we were late.’
b. Ele lamentou que estejamos atrasados. he regret-PAST that [we] be-PRES-SUBJ late
‘He regretted that we are late.’
In fact, temporal dependence exists or not depending on whether the embedded tense is deictic (i.e. takes t0 as its TPpt) or anaphoric, in which case its TPpt may be given by the main clause and temporal dependence obtains. Another problem for the considered proposals is that they correlate temporal (in)dependence with mood selection: indicative complements would be temporally independent while subjunctive complements would be temporally dependent. However, this correlation does not apply, since some verbs that select for the indicative impose restrictions on the tenses allowed in the complement clause (8), and some verbs that select for the subjunctive allow for any tense in the complement clause (9). (8) Ele decidiu que {*tinha ido / *foi / he decide-PAST that [he] {*go-PAST-PERF-IND / *go-SIMPLE_PAST-IND ia / vai / irá} ao cinema com go-PAST-IMPERF-IND / go-PRES-IND / go-FUT-IND} to-the cinema with a Rita. the Rita
‘He decided that he {*had gone/*went/would go/will go} to the cinema with Rita.’
Sequence of tenses in complementation structures
(9) O João pediu que a irmã {fosse / vá} the João ask-PAST that the sister {go-PAST-SUBJ / go-PRES-SUBJ} ao cinema. to-the cinema
‘João asked his sister to go to the cinema.’ More generally, restrictions on combinations of tenses in verb complement structures of EP are independent of mood selection: some verbs that select the indicative (e.g. decidir ‘to decide’) impose restrictions on the tense acceptable in their complement (as in (8)), contrary to other indicative rulers (e.g. perceber ‘to realize’; see (10)), and the same is observed with verbs that select the subjunctive, that is, subjunctive rulers like querer (‘to want’) impose restrictions on the tense allowed in the complement clause (see (11)) while other subjunctive rulers, like pedir (‘to ask’), do not (as in (9)). (10) O João percebeu que a irmã {tinha ido / the João realize-PAST that the sister {go-PAST-PERF-IND /
foi / ia / vai / irá} go-SIMPLE-PAST-IND / go-PAST-IMPERF-IND / go-PRES-IND / go-FUT-IND} ao cinema com a Rita. to-the cinema with the Rita.
Lit.: John realized that his sister {had gone/went/would go/goes/will go} to the cinema with Rita.’ (11) O João quis-PAST que a irmã {*tivesse ido / the João want-PAST that the sister {*go-PAST-PERF-SUBJ /
fosse / *vá} ao cinema com a Rita. go-PAST-IMP-SUBJ / *go-PRES-SUBJ} to-the cinema with the Rita Lit.: John wanted that his sister {*had gone/would go/*goes to the cinema with Rita.
3. Temporal orientation In work on the distribution of inflected and uninflected infinitive in EP complement clauses, Duarte, Gonçalves & Santos (2012) question the relevance of temporal dependence as the mechanism that rules that distribution. The authors show, contrary to Raposo (1987) and others, that not all temporally independent domains allow for inflected infinitives and they propose instead that the relevant lexical property is temporal orientation (i.e. whether the main verb specifies the temporal location of the embedded situation as being in a relation of anteriority, overlapping or posteriority with the situation described by the main clause).
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
Assuming temporal orientation as a mechanism implied in sequences of tenses allows us to account for interesting contrasts. Verbs such as querer (‘to want’) and decidir (‘to decide’) are prospective: the temporal location of the situation described by their complement clause follows the attitude time. It seems clear that this temporal orientation specified by the main verb blocks some combinations of tenses, such as the sequence pres + past in (1) – *Ele quer-PRES que a Ana estivesse-PAST-IMPERF-SUBJ em casa; lit.: ‘he wants Ana to have been home’ – or past + past perfect in (8) – *Ele decidiu-PAST que {tinha ido-PAST-PERF-IND/foi-SIMPLE_PAST-IND} ao cinema com a Rita: lit: ‘he decided to have gone to the cinema with Rita’. It makes no sense to desire or to decide now to do something in the past. On the contrary, declarative verbs (dizer ‘to say’) and doxastic verbs (pensar ‘to think’) do not impose restrictions on the temporal location of the situation described by their complement clause and the sequence pres + past is allowed (see (3)). However, the temporal orientation specified by the matrix verb is not sufficient to explain all the lexical restrictions, given that the sequence past + pres is blocked in structures with some doxastic verbs ((12)–(13)), which do not specify any temporal orientation (we may believe that something was the case, is the case or will be the case): (12) Supus que ela {*está / estava em casa. [I] suppose-PAST that she {*be-PRES-IND / be-PAST-IND} at home ‘I supposed that she {*is/was} at home.’ (13) Duvidei que ela {*esteja-PRES-SUBJ / estivesse-PAST-SUBJ} em casa. [I] doubt-PAST that she {*be-PRES-SUBJ / be-PAST-SUBJ} at home ‘I doubted that she {*is/was} at home.’
4. T owards an explanation of lexical restrictions on the sequence PAST + PRES In this section, we focus on the sequence of tenses past + pres in EP. As seen above, this sequence of tenses is blocked or allowed depending on the verb of the main clause. We suggest that an investigation on the semantics of the matrix verbs that takes into account the idea of reference to the real world will shed some light on the question of why this sequence of tenses is allowed in some cases but not in others. But first, it is important to consider the interpretation of the present tense. We know that present under past (past + pres) yields different readings in so-called SOT languages (English, French, Portuguese) and non-SOT languages (Japanese) (see Ogihara, 1999). Specifically, in SOT languages present under past has a Double Access reading, that is, the embedded tense must denote an interval overlapping two reference times – the matrix situation time and the utterance time – while in non-SOT languages the present tense will be semantically vacuous and a simultaneous
Sequence of tenses in complementation structures
reading obtains. This means that in SOT languages the interpretation of the present tense forces the consideration of t0 both when it occurs in complement clauses and when it occurs in other kinds of sentences. At least concerning EP, this is a valid observation for the present indicative as well as for the present subjunctive. Moreover, the present subjunctive cannot occur in counterfactual sentences, which shows that this tense points not only to t0 but also to the reference world (which, by default, is the real world, w0). In sum, in SOT languages the present tense is deictic. Its interpretation requires the consideration of the context of assertion, given that it points to utterance time and the reference world. In other words, in Portuguese (as, presumably, in other SOT languages), the present tense points to how the world is or might be, as seen from the context of assertion. Given this, let us now consider the lexical restrictions on the sequence past + pres. Examples (14)–(18) show that this sequence is possible if the verb of the main clause is one of the following classes: a. b. c. d. e.
factive verbs (e.g. surpreender ‘to surprise’, saber ‘to know’) implicative verbs2 (e.g. conseguir ‘to manage’, preocupar-se ‘to bother’) declarative verbs (e.g. dizer ‘to say’, garantir ‘to assure’) commissive verbs (e.g. prometer ‘to promise’, jurar ‘to swear’) deontic verbs (e.g. ordenar ‘to order’, permitir ‘to allow’)
(14) Nunca te surpreendeu que haja tantas estrelas? never you surprise-PAST that be-PRES-SUBJ so-many stars? ‘Were you never surprised that there are so many stars?’ (15) «A Lotus conseguiu que, além da facilidade de utilização, the Lotus manage-PAST that, apart from-the ease of use, os seus potenciais utilizadores não se sintam frustrados the its potential users not them feel-PRES-SUBJ frustrated por limitações na sua nova estrela.» by limitations in-the its new star
(CetemPúblico,3 par=ext178447) ‘Lotus managed that, apart from user-friendliness, their potential customers do not feel frustrated by limitations of its new star.’
. Both the terms ‘implicative verbs’ and ‘factive verbs’ are taken from Karttunen (1971). Factive verbs allow the inference that its complement clause is true, regardless of the truth value of the main clause; implicative verbs allow the inference that its complement clause is true when the main clause is affirmative, but do not allow such an inference when the main clause is negated. . 〈http://www.linguateca.pt/〉
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
(16) Ele disse que está em casa. he say-PAST that [he] be-PRES-IND at home ‘He said that he is home.’ (17) Ele prometeu que me avisa. he promise-PAST that me warn-PRES-IND ‘He promised that he warns me.’ (18) O general ordenou às tropas que cerquem the general order-PAST to-the troops that surround-PRES-SUBJ o aeroporto. the airport
‘The general ordered the troops to surround the airport.’ The examples (19)–(21) show that the sequence past + pres is blocked if the main verb is one of the following classes: a. volition verbs (e.g. querer ‘to want’, esperar ‘to hope’) b. doxastic verbs (e.g. supor ‘to suppose’, duvidar ‘to doubt’) c. negative verbs (e.g. impedir ‘to prevent’, evitar ‘to avoid’) (19) *Ele quis que a Ana lhe telefone. he want-PAST that the Ana him phone-PRES-SUBJ Lit.: ‘*He wanted that Ana will phone him.’ (20) *Supus que ele está em casa. [I] suppose-PAST that he be-PRES-IND at home ‘*I supposed that he is at home.’ (21) *O nevoeiro impediu que o avião aterre (agora). the fog prevent-PAST that the plane land-PRES (now) ‘*The fog didn’t allow the plane to land (now).’
4.1 Lexical restrictions and possible worlds semantics The observed division between the verbs that allow the sequence past + pres and those that block it is not equivalent to the division between the verbs that select the indicative and those that select the subjunctive. In fact, both the group of verbs that allow the sequence past + pres and the group of verbs that block it include verbs that select the indicative and verbs that select the subjunctive (e.g. the equivalents of promise, an indicative ruler, and order, a subjunctive ruler, for the first group of verbs, and the equivalents of suppose, indicative ruler, and doubt, subjunctive ruler, for the second group of verbs). However, given that in Portuguese (as in other Romance languages) verbal inflection does not involve independent morphemes for mood and tense, it makes sense to have a common analysis of mood and tense in complement clauses. As
Sequence of tenses in complementation structures
we will endeavor to show, a semantic analysis within possible worlds semantics, which has proved to be useful in the semantic analysis of modality and mood, may shed some light on the lexical restrictions on sequences of tenses. In possible worlds semantics, the meaning of a proposition is a set of worlds: the ones where the state of affairs described by the proposition is verified. Concerning mood, the observation has been made (Marques, 2009; Giannakidou, 2013; among others) that the indicative occurs in contexts where meaning involves the consideration of only p-worlds (i.e. worlds where the proposition p is verified) and the subjunctive occurs when at least one non-p world is taken into consideration. This is different from the claim of traditional grammars, which state that the opposition between indicative and subjunctive mirrors the realis/irrealis opposition. In other words, it is not the case that the indicative is selected if the proposition describes reality. The verb to think, for instance, selects the indicative even if its complement clause is taken to be false in the real world. The reason this verb selects the indicative is that the epistemic (or doxastic) model the verb points to includes only p-worlds. For instance, the sentence John thinks that Sydney is the capital of Australia means that in every world compatible with John’s beliefs, Sydney is the capital of Australia. If John is not certain that the proposition is true, then his doxastic model includes at least one non-p world (that is, he admits the possibility that the sentence is false), and in this case the subjunctive has to be selected. In sum, the selection of mood is not dependent on whether the sentence is actually true in the real world: in the considered example, the complement clause of to think is false in the real world (the capital of Australia is Camberra), but the indicative would be selected. Instead, what is relevant for the selection of mood is whether the meaning of the construction involves only p-worlds or at least one non-p world. Given this, let us now return to the lexical restrictions on the past + pres sequence of tenses. It was observed that the sequence past + pres is ruled out when the main verb is a volition verb, a doxastic verb or a negative verb. If we consider the sets of possible worlds denoted by the complement clause of these verbs it is clear that in the case of negative verbs the real world does not belong to this set. That is, these verbs are antiveridical (Giannakidou, 1999): they allow the inference that their complement clause is false. As for doxastic verbs, they do not allow the inference that their complement clause is true in the real world nor that it is false. In other words, the set of worlds denoted by the complement clause of these verbs might or might not include the real world. Hence, negative verbs and doxastic verbs have in common the fact that their complement proposition does not have to include the real world. As for volition verbs, we might think that their complement clause denotes a set of worlds where the desire is fulfilled. Obviously, this set of worlds does not have to include the real world (to want p does not guarantee that p will actually happen). If this is so, the observation follows that the meaning of the three considered kinds of verbs (negative, doxastic and volition verbs) does not impose that the real world belongs to the set of worlds denoted
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
by the complement clause. In other words, the meaning of these verbs does not force the consideration of reality. By contrast, the meaning of the verbs with which the sequence past + pres is allowed seems to impose the consideration of the real world. This does not mean that the complement clause of all these verbs is taken to be true in reality. It means that the denotation of the complement clause of these verbs is a set of worlds that necessarily includes the real world. This is clearly the case of factive verbs, whose complement clause describes a fact of the real world, and also of implicative verbs, which, in affirmative sentences, allow the inference that the complement clause is true in the real world. As for the other kinds of verbs that allow the sequence past + pres − declarative, commissive, and deontic verbs − though they do not convey the truth of their complement clause in the real world, we argue that their meaning also involves the consideration of reality. Concerning declarative verbs, a sentence like John said that Bill is a spy reports an assertion, and, as it is known from Speech Act Theory, by asserting a sentence, the speaker expresses his belief that the sentence is true (in the real world) and that he has adequate evidence that it is true. Likewise, concerning commissive verbs, a sentence like John promised to read the book expresses the information that John committed to reading the book and thus turning the complement clause true in the real world. Hence, both declarative and commissive verbs express someone’s commitment to the truth of the complement proposition in the real world. This means that these verbs are veridical, in the sense of Giannakidou (1999). Doxastic verbs, like to think, are also veridical verbs: they allow the inference that the complement clause is true in the relevant model. However, doxastic verbs do not commit the attitude holder to the truth of the proposition in the real world in the same way as declarative and commissive verbs. If someone says that p is true and it turns out that in fact p is false, or if someone promises to do something and does not do it, he may be accused of being dishonest. But if someone believes that p is true and it turns out that p is false, he is not accused of being deceptive. Hence, the conclusion follows that the set of worlds denoted by the complement clause of declarative and commissive verbs necessarily includes the real world (these verbs express a commitment of someone to the truth of their complement in the real world), contrary to the set of worlds denoted by the complement clause of doxastic verbs. This issue seems to be relevant for the lexical restrictions on sequences of tenses, given that, as seen above, EP doxastic verbs do not allow the sequence past + pres, contrary to declarative and commissive verbs. Finally, deontic verbs also allow such a sequence of tenses. These verbs are identical to commissive verbs in the sense that both kinds of verbs point to the attainment of the action described by their complement. The main difference is that in the case of commissive verbs the responsibility for the fulfillment of the action belongs to the agent of the promise and, in the case of deontic verbs, the agent of the action is someone other than the one who gives the order or
Sequence of tenses in complementation structures
makes the request. This difference aside, both commissive and deontic verbs point to the performance of an action in the real world. Hence, also in the case of deontic verbs, reality has to be considered in the denotation of the complement clause. If this is so, the following generalization holds: in EP, the sequence past + pres is possible only if the set of worlds denoted by the complement clause necessarily includes the real world. The explanation for this comes easily if one considers that the present tense is deictic, pointing to the utterance time and the real world: if the set of worlds denoted by the complement clause does not include the real world or may not include it, the present tense would have no reference or could have no reference. Hence, the anomaly of past + pres with the observed kinds of verbs would be a case of reference failure of the present tense. However, there are three issues in this analysis, which will be considered in the next subsection: (i) the assumption that the real world does not necessarily belong to the denotation of the complement clause of volition verbs is questionable; (ii) at least for some speakers of Portuguese, the sequence past + pres is acceptable when the main verb is sonhar ‘to dream’,4 as shown by (22); (iii) if the sequence past + pres is blocked by some verbs because the present tense would have no reference, how can the fact that the sequence pres + pres or pres + fut is acceptable with the same verbs be explained? (22) Na noite passada, sonhei que o meu nome está in-the night passed, dream-PAST that the my name be-PRES no quadro dos vencedores. at-the board of-the winners
‘Last night I dreamt that my name is on the winners’ board.’
4.2 Aktionsart and accessible possibilities In the previous subsection, we proposed that the denotation of volition verbs does not have to include the real world (it is possible that desires are never fulfilled), contrary to the denotation of the complement of deontic verbs. However, this may be problematic if one assumes the analysis of volition verbs proposed by Heim (1992). Briefly, she observes that these verbs express an ordering of the available possibilities. A sentence like John wants to teach on Mondays means that, given the possibilities for John to teach on Mondays or another day of the week, he prefers the former. It does not (necessarily) mean that in all worlds that conform to John’s desires he teaches on Mondays (for him, the best world may be one where he does not teach at all). In other words, the
. We thank Inês Duarte (p.c.) for pointing this out.
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
meaning of volition verbs involves a comparison between the p-worlds and the non-p worlds that are doxastically accessible to the attitude holder. One may add that the same kind of comparison of available possibilities seems to be involved in the meaning of deontic verbs. To ask or to order someone to do something means that there is the possibility of doing that or not doing it and the request is made to choose the first possibility. This being so, it seems difficult to maintain that volition verbs are different from deontic verbs concerning the set of worlds considered in the complement clause. Thus the question remains why the sequence past + pres is impossible with volition verbs and acceptable with deontic verbs. Following Marques (2014), we propose that, apart from the set of worlds denoted by the complement clause, the Aktionsart value of the main predicate also plays a role in the restrictions on SOT.5 Roughly, the idea is that a relevant factor for the lexical restrictions on sequences of tenses is whether the kind of attitude the main predicate refers to is temporally bounded or not. For instance, the attitudes of knowing, believing or desiring are temporally unbounded (these are stative predicates), but the attitudes of ordering, saying, or becoming aware, for instance, do not extend indefinitely in time. Now, when the reported attitudes are temporally unbounded, by inflecting the attitude verb in the past the speaker conveys the information that the attitude is no longer verified at the utterance time. However, such information is not conveyed if the reported attitude is temporally bounded, even if the verb is inflected in the past. For instance, the sentence John wanted Mary to leave does not allow the information that he still wants her to leave, but the sentence John asked Mary to leave may convey the information that John’s request maintains. In other words, the inflection in the past of a verb that refers to an unbounded attitude is a strategy to impose a boundary on the attitude and convey the information that the attitude is not verified at the report time. Given this, the construction *Ela quis-PAST que lhe telefones-PRES (lit: ‘she wanted that you phone her’) would be contradictory: on one hand, the present tense in the complement clause points to the utterance time (the report time); on the other hand, the inflection of the main verb in the past indicates that the attitude is not verified at the utterance time. On the contrary, the construction ela pediu-PAST que lhe telefonesPRES (‘lit.: ‘she asked that you phone her’) is natural because the reported attitude is temporally bounded and the inflection of the main verb in the past does not imply that the request is no longer verified at the report time. Let us now consider the second problem identified above: the acceptability of the sequence past + pres with the verb sonhar ‘to dream’. This verb indicates that
. For the influence of Aktionsart (or inner aspect) in the account of the restrictions of tense sequences, see also Tenny (1987), Smith (1991) or MacDonald (2008), among others.
Sequence of tenses in complementation structures
its complement proposition is true in the dream world (in the dream, the dreamer believes that the proposition is true). The complement proposition may or may not also be true in the real world. In this respect, this verb is identical to verbs like to think or to believe, which block the sequence past + pres. However, contrary to these verbs, to dream (in its literal sense) refers an attitude temporally bounded (the attitude ends at least when one wakes up). Thus, the inflection of the verb in the past does not give rise to the inference that the dreamer no longer believes that the complement proposition is true. Hence, the prediction arises that the sequence past + pres is acceptable with the verb to dream. It indicates that the dreamer still affirms the truth of the complement proposition. Finally, let us consider the third problem identified above: if some verbs block the access to t0 by the embedded tense, why is the sequence pres + pres possible with these verbs? One possibility is to consider that in these cases the embedded tense is semantically vacuous, assuming a mechanism like Ogihara’s (1996) deletion under identity (or an equivalent mechanism available in the literature). However, this approach would have problems with the sequence pres + fut, and an analysis that assumes that embedded tenses always have semantic import, in line with Gennari (2003), would be preferable. Moreover, following Portner (1997), it may be pertinent to replace the notion of possible world with the notion of possible situation, as defined by Kratzer (1989).6 Given this, a sentence with the present tense would denote a set of possible situations that include t0, and, as seen above, the real world must be considered in the denotation of the proposition. Hence, a sentence with the present tense points to a set of possible situations (“slices” of possible worlds) and this set has to include a part of the real world that contains t0. Likewise, a sentence with the future tense would denote a set of possible situations that temporally follow t0 (a set of possible continuations of the real world in the future). In other words, the present tense points to what is or might be the case in the context of utterance and the future tense points to possible continuations of reality. The possibilities (how the world is, might be or will be) are a set of possible situations. In main and adverbial clauses, this set is part of the context set (the set of possible worlds or possible situations compatible with what is known or assumed to be the case). In the case of complement clauses, the possibilities are a set expressed by the main verb that is available to the holder of the attitude (it is the set of possible worlds or possible situations compatible with what the attitude holder knows or assumes to be the case). Now, given that the present tense points to t0, a complement clause with
. Roughly, possible situations are parts of possible worlds; a possible world being a maximal possible situation.
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
this tense denotes a set of situations that must be available to the attitude holder both at the time of the attitude and at utterance time. This is what happens in the cases of pres + pres or pres + fut. However, in the case of past + pres, the situations will be available to the attitude holder both at the time of the attitude and in the context of utterance only if the attitude extends from the time of the attitude until utterance time. Given this, consider the examples in (23)–(24): (23) A Sara acha que está grávida. the Sara think-PRES that [she] be-PRES-IND pregnant ‘Sara thinks that she is pregnant.’ (24) *A Sara achou que está grávida. the Sara think-PAST that be-PRES-IND pregnant ‘Sara thought that she is pregnant.’ In both constructions the complement proposition denotes a set of possible situations where Sara is pregnant at utterance time. The attitude expressed in (23) relates the attitude holder (Sara) and the same set of situations. That is, Sara being pregnant at t0 is a possibility available to Sara at the time of the attitude and thus the set denoted by the complement clause is available to Sara at utterance time and at the attitude time. But (24) does not allow the inference that Sara still thinks that she is pregnant. Therefore, the set denoted by the complement clause − a set of possible situations where she is pregnant at t0 − is not available to the attitude holder at utterance time and at the attitude time.
5. A cquisition of complementation, first combinations of tenses and SOT: Some data Given the state of affairs described in the last section, we can say that the interpretation of tenses in the main and the embedded clause results from general properties of grammar (subordination) and the value of particular tenses. However, the combination of tenses in the main and in the embedded clause is also constrained by a parameter (the SOT parameter; briefly, in non-SOT languages present under past is semantically vacuous, while in SOT languages it has semantic import) and by lexical semantics, in the sense that it is constrained by the semantics of the matrix predicate. According to Hollebrandse (2000), acquisition of SOT, as any relation of temporal dependence between matrix and embedded clauses, is part of the acquisition of complementation. In the first stages, children may not interpret a dependent tense, always taking utterance time as the TPpt of the embedded tense.
Sequence of tenses in complementation structures
When children acquire the possibility of a dependent tense, they will need to determine whether their language is an SOT or a non-SOT language, something that can be thought of as parameter setting (Demirdache & Lungu, 2009). This will condition the interpretation of present under past structures. As described in Section 4, positively setting the SOT parameter for Portuguese will also make the lexical semantic restrictions on past+pres available. In this section, we examine child performance in the beginning stages of the acquisition of complementation, as a first step in a larger research project on the development of syntax and semantic aspects of complementation. We report here the results of the analysis of spontaneous complement clause production by European Portuguese monolingual speakers, on the assumption that choice of tenses in the first complement clauses may contribute to the discussion of the acquisition of SOT. We therefore posed two questions with respect to early child productions in language acquisition: (i) Do children produce structures of verbal complementation where main and embedded clauses form a single temporal domain? (ii) How do children deal with lexical restrictions on SOT? We examined the spontaneous production corpus of Santos (2006), extended version. The corpus includes spontaneous data from three children, treated according to the CHAT format (MacWhinney, 2000) (details are given in Table 1). Each file corresponds to 45–50 minutes of adult-child interaction in a naturalistic setting (children were recorded at their homes while interacting with their family and the researcher). Table 1. Corpus (Santos, 2006, enlarged) Child
Age
MLUw
Number of files
INI
1;6.6–3;11.12
MLUw 1.5–3.8
21
TOM
1;6.18–3;0.22
MLUw 1.2–3.3
21
INM
1;5.9–2;9.3
MLUw 1.3–1.95
16
The corpus was manually coded for all complement clauses and subsequently annotated for all combinations of tenses. We identified 1358 cases of VP or clausal complementation of verbs (including auxiliaries, modals and verbs selecting for complement clauses and including elliptical complements);7 within this set, only a subset of 37 cases are finite (indicative or subjunctive) complements – we will concentrate
. EP licenses VP ellipsis with either a stranded auxiliary or a stranded main verb (see Matos, 1992). Some of these cases are cases with a stranded verb and an elliptical VP or clausal complement.
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
on this subset. In all the cases, the researchers used their own intuition to interpret the combination of tenses in the matrix and the embedded clause, also looking at the discourse context when relevant. The first occurrences of the complementizer que ‘that’, which introduces a finite complement clause, are found at 2;1.7 (TOM), 2;1.10 (INI) and 2;6.19 (INM) (see Santos, 2006). However, the first occurrences are cases such as (25), where no embedded tense is found. (25) TOM: xxx [//] # acho que sim. think that yes ‘I think that it is true./I agree.’
TOM (2;1.7)
The first finite subordinate clauses whose combination of tenses could be analyzed occur at 2;5. Of the total 37 cases of finite complement clauses, in 21 cases (57%) the embedded tense has t0 as its TPpt; in 16 cases the combination is pres + pres (see (26)); and in 5 cases the matrix tense is present and the embedded tense is a different tense, t0 being maintained as the TPpt (see (27)). (26) não # eu ach(o) qu(e) é um menino e no I think-PRES that is a boy and uma menina. a girl
INI (3;0.15)
‘No # I think that it is a boy and a girl.’ (27) eu ach(o) qu(e) ela vai chorar. I think- PRES that she will cry ‘I think that she is going to cry.’
INI (3;0.15)
In 2 cases, the TPpt of the embedded tense is not t0. These cases occur after 3;0, and show that the child may define a TPpt different from t0; however, they do not exhibit temporal dependence, since in this case the TPpt is not provided by the main clause but rather is given by discourse (see (28) and (29)). (28) não acham # qu(e) ele ia assim com neg think-PRES that he go-PAST-IMPERF like-that with (u)ma cara? a expression
INI (3;4.6)
‘Don’t you think that he was going with such an expression?’ (29) eu ach(o) qu(e) el(e) ia era cai(r) dali I think-PRES that he go-PAST-IMPERF was fall from de cima. up there
‘I think that he was going to fall from upstairs.’
INI (3;4.6)
Sequence of tenses in complementation structures
Only in 3 cases, produced by the same child and always after 3 years of age, do the embedded and the complement clause clearly share the same temporal domain, the TPpt for the embedded tense being provided by the matrix. In these cases, we have a past (perfect “Pretérito Perfeito” or imperfect “Pretérito Imperfeito”) in the matrix clause and an anaphoric tense, a past (“Pretérito Imperfeito”), in the embedded clause.8 (30) mas o rei qu(e)ria qu(e) a filha casasse com but the king want-PAST that the daughter marry-PAST-SUBJ with um príncip(e). a prince
INI (3;11.12)
‘But the king wanted the daughter to marry a prince’ (31) ele disse que não podia dar. he said-PAST that neg could-PAST-IMPERF give ‘He said that he couldn’t give.’
INI (3;4.6)
(32) papá disse qu(e) er(a) um te(r)mómetro. INI (3;0.15) daddy say-PAST that be-PAST-IMPERF a thermometer ‘Daddy said that it was a thermometer.’ The data presented thus far suggest that, at the first stages of production of complement clauses, a clear preference is observed for tenses in the complement clause that take t0 as its TPpt. The greater frequency of deictic tenses, which take t0 as the TPpt, may be due to the situations in which the data was collected, mostly centered in the here and now. But it may also be due to the inability to compute temporal dependence at these early stages. This is compatible with the view that not all semantic aspects of subordination may be available at the beginning stages of production of clausal complementation. Though there is some evidence of temporal dependence in the observed data, there is no such evidence at the earliest stages and neither was this observed in the speech of all the children. On the other hand, there were no instances of a lexically restricted combination of tenses (namely pres+past, past+past (perf) or past+pres) found in the child speech analyzed in this paper. It is to be expected that lexical restrictions are acquired later than the general possibilities of the grammar.9 In the case of past+pres, we have seen that such a lexical restriction may be derived from the Double Access interpretation of
. Other combinations of tenses also present in the data are imperative + present or infinitive + present, as well as cases of an omitted auxiliary. . As noted by one anonymous reviewer, this is generally true in language acquisition, as it is also true that innate components of language constrain child grammar generally before language-specific components (see e.g. the contrast between acquisition of wh- movement and auxiliary inversion in English – revision and comments are found in Guasti, 2002: 201–2).
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos
a present under past in a SOT language (therefore a language-specific aspect of grammar). Thus, acquisition of past+pres may be conditioned by positively setting the SOT parameter in the acquisition of Portuguese. The absence of such a combination of tenses at these beginning stages is therefore also expected in light of previous work suggesting a fluctuation between SOT and non-SOT values of the parameter in 5 to 7 year-olds (see Demirdache & Lungu, 2009).10
6. Conclusion In this work, we developed an approach to sequences of tenses in matrix and finite complement clauses and investigated child data in order to shed some light on the acquisition of complementation. We claimed that, although temporal dependence and temporal orientation are two mechanisms relevant to account for the sequences of tenses in finite complements, some lexical restrictions (independent of the mood occurring in the complement clause) are also required, namely to explain the impossibility of past + pres with volition, doxastic and negative verbs. We then proposed that this sequence is only possible when the set of worlds denoted by the complement clause necessarily includes the real world. Furthermore, we showed that all verbs allow pres + pres and pres + fut by assuming that present tense points to t0 and, therefore, a complement clause with this tense denotes a set of possible situations that must be available to the attitude holder both at the time of the attitude and at utterance time. As for the acquisition issue, the preliminary data point to an initial stage where a strong preference for t0 as the TPpt is observed. Only later do children spontaneously produce structures where temporal dependence may be identified.
References Ambar, M. (1992). Temps et structure de la phrase en portugais. In H. Obenhauer & A. Zribi-Hertz (Eds.), Structure de la phrase et théorie du liage (pp. 29–49). Saint-Denis: Presses Universitaires de Vincennes. DOI: 10.1075/li.18.1.13mul
. In this paper, and given space limitations, we will not discuss the acquisition of tense and tense morphology (which could lead us to discuss the “Aspect first hypothesis” – see Wagner, 2001); instead, here we are interested in the constraints on the combination of tenses in the matrix and the embedded clause. However, on the assumption that the acquisition of matrix tense must be prior to the acquisition of sequence of tense, and given that we are looking at child spontaneous production from very early stages it might be that certain aspects of tense marking itself are not yet fully acquired by these children.
Sequence of tenses in complementation structures
Demirdache, H., & Lungu, O. (2009). Sequence of tense in (French) child language. Linguistic Variation Yearbook, 8, 101–130. DOI: 10.1075/livy.8.04dem Duarte, I., Gonçalves, A., & Santos, A.L. (2012). Infinitivo flexionado, independência temporal e controlo. In A. Costa, C. Flores, & N. Alexandre (Eds.), Textos selecionados do XXVII Encontro Nacional da Associação Portuguesa de Linguística (pp. 217–234). Lisboa: APL. Gennari, S.P. (2003). Tense meanings and temporal interpretation. Journal of Semantics, 20, 35–71. DOI: 10.1093/jos/20.1.35 Giannakidou, A. (1999). Affective dependencies. Linguistics and Philosophy, 22, 367–421. DOI: 10.1023/a:1005492130684 Giannakidou, A. (2013). The epistemic basis of mood: selection, nonveridical weakening, and emotive verbs. In Blaszack, J. et al. (Eds.), Tense, mood, and modality: New perspectives on old questions. Chicago, IL: University of Chicago Press. Gonçalves, A., Cunha, L.F., & Silvano, P. (2010). Interpretação temporal dos domínios infinitivos na construção de reestruturação do Português europeu. In A. Brito, J. Veloso, & A. Fiéis (Eds.), Textos seleccionados do XXV Encontro Nacional da Associação Portuguesa de Linguística (pp. 435–447). Porto: APL. Guasti, M.T. (2002). Language acquisition. The growth of grammar. Cambridge, MA: The MIT Press. DOI: 10.1017/s0142716402233081 Heim, I. (1992). Presupposition projection and the semantics of attitude verbs. Journal of Semantics, 9, 183–221. DOI: 10.1093/jos/9.3.183 Hollebrandse, B. (2000). The acquisition of sequence of tense. Unpublished Doctoral Dissertation. University of Massachusetts, Amherst. Kamp, H. & Reyle, U. (1993). From discourse to logic: Introduction to modeltheoretic semantics of natural language, formal logic and discourse representation. Dordrecht: Kluwer. DOI: 10.2307/415752 Karttunen, L. (1971). Implicative verbs. Language, 47, 340–358. DOI: 10.2307/412084 Kempchinsky, P. (1990). Más sobre el efecto de referencia disjunta del Subjuntivo. In I. Bosque (ed.), Indicativo y subjuntivo (pp. 234–258). Madrid: Ediciones Taurus. Kratzer, A. (1989). An investigation of the lumps of thought. Linguistics and Philosophy, 12, 607–653. DOI: 10.1007/bf00627775 Marques, R. (2009). On the selection of mood in complement clauses. In L. Hogeweg, H. de Hoop, & A. Malchukov (Eds.), Crosslinguistic semantics of tense, aspect, and modality (pp. 179–204). Amsterdam: John Benjamins. DOI: 10.1075/la.148.08mar Marques, R. (2014). Référence au réel et concordance de temps en portugais européen. Travaux linguistiques du Cerlico, 27. (http://www.pur-editions.fr/detail.php?idOuv=3659). Matos G. (1992). Construções de elipse do predicado em Português. SV Nulo e Despojamento. Unpublished Doctoral Dissertation. Universidade de Lisboa. MacDonald, J. (2008). The syntactic nature of inner aspect. A Minimalist perspective. Amsterdam: John Benjamins. DOI: 10.1075/la.133 MacWhinney, B. (2000). The CHILDES Project: Tools for Analyzing Talk. 3rd Edition. Mahwah, NJ: Lawrence Erlbaum Associates. Ogihara, T. (1999). Double-access sentences generalized. In T. Matthews & D. Strelovitch (Eds.), Proceeding of SALT IX (pp. 224–236). Ithaca, NY: CLC Publications, Cornell University. Ogihara, T (1996), Tense, Attitudes, and Scope. Dordrecht: Kluwer. DOI: 10.1007/978-94-015-8609-2_6 Picallo, C. (1984). The Infl node and the null subject parameter. Linguistic Inquiry, 15, 75–102.
Rui Marques, Purificação Silvano, Anabela Gonçalves & Ana Lúcia Santos Portner, P. (1997). The semantics of mood, complementation and conversational force. Natural Language Semantics, 5, 167–212. DOI: 10.1023/a:1008280630142 Raposo, E. (1987). Case theory and Infl-to-Comp: The inflected infinitive in European Portuguese, Linguistic Inquiry, 18, 85–109. Santos, A.L. (2006). Minimal answers. Ellipsis, syntax and discourse in the acquisition of EP. Unpublished Doctoral Dissertation, Universidade de Lisboa. Silvano, P. (2002). Sobre a semântica da sequência de tempos em Português Europeu. Análise das relações temporais em frases complexas com completivas. Unpublished MA Dissertation, Universidade do Minho. Smith, C. (1991). The Parameter of Aspect. Dordrecht: Kluwer. Tenny, C.L. (1987). Grammaticalizing Aspect and Affectedness. Unpublished Doctoral Dissertation, MIT. Wagner, L. (2001). Aspectual influences on early tense interpretation. Journal of Child Language, 28(3), 661–681. DOI: 10.1017/s0305000901004792
Fue muerto Suppletion in Spanish analytic passives Grant Armstrong1 & Jason Doroga2 1University
of Wisconsin-Madison / 2Centre College
While allomorphy is a widespread phenomenon in the functional vocabulary of the world’s languages (i.e. English past tense -d, -t and -Ø), its status in the domain of roots is still an open question. Does root allomorphy, or suppletion, exist? In this paper we review the recent controversy attached to this question within the framework of Distributed Morphology (DM) and then provide an argument in favor of the existence of suppletion from the distribution of the vocabulary items mat- (kill) and mor- (die). Using a model for analyzing suppletion outlined in Harley (2014), we propose that these two vocabulary items are members of the same set of lists that are accessed by a single syntactically-active root node. Mat- is specified to appear in agentive verbal environments while mor- is an elsewhere morpheme. Keywords: suppletion; passives; participles; Distributed Morphology
1. Introduction Allomorphy refers to variation in the surface form of a particular morpheme and constitutes one of the most important topics in morphology. The most widely studied cases of allomorphy involve surface variation in the functional domain that is triggered by a local environment. For example, the Spanish plural morpheme has three allomorphs, -s (gato-s, ‘cats’), -es (flor-es, ‘flowers’) and -Ø (análisis-Ø, ‘analysis’), which are conditioned by the local phonological environment in which they appear. The regular English past tense morpheme -d exhibits phonologically-conditioned allomorphy in cases like play-ed, walk-ed and start-ed; however, there are additional allomorphs that are conditioned by a particular set of roots such as hit-Ø, broke-Ø and ben-t. Given that cases such as these are widespread in the world’s languages, two major generalizations about allomorphy have emerged from the detailed treatment it has received in the literature: (i) it is (largely) restricted to the functional (as opposed to the “contentful”) vocabulary of different languages and (ii) it is subject to strict locality conditions (see Bobaljik, 2000, 2012; Embick, 2010 for overviews). The statement in (ii) appears to be doi 10.1075/ihll.4.05arm © 2015 John Benjamins Publishing Company
Grant Armstrong & Jason Doroga
without exception and speaks to the need for cyclic operations in grammar (Embick, 2010) while the statement in (i) is more controversial and leads to a number of important questions. Does root allomorphy, or suppletion, constitute an empirical reality in human language? If so, what is the most appropriate way to analyze it? In this paper we tackle both of these questions by taking a detailed look at, as far as we know, a heretofore unanalyzed case of suppletion in the analytic passive of the transitive verb matar (‘kill’). Our point of departure is data such as (1), where ‘was killed’ is rendered fue muerto (lit. ‘was died/dead’) instead of the expected fue matado (lit. ‘was killed’). (1) Un fugitivo fue muerto a tiros en la puerta de su casa por la policía A fugitive was died by shots at the door of his house by the police ‘A fugitive was shot to death at the door of his house by the police’ (Source: La Nación, 12/3/12) At first blush, a plausible way of explaining this datum is to claim that when the root associated with the meaning kill appears in the environment specified as [passive], a form of the semantically-related verb die is spelled out in its place, much like certain phonological features that trigger the appearance of a particular allomorph in the plural examples given in the first paragraph. This is, in a nutshell, the analysis we propose. However, before reaching that point it is imperative to situate the data in the context of the controversies surrounding suppletion that have developed over the last two decades and to motivate why the vocabulary items in question should be considered allomorphs of a single √root terminal node. Section 2 provides an overview of the theoretical controversies associated with suppletion. Section 3 motivates the treatment of the vocabulary items mat- (root for kill) and mor- (root for die) as forms competing for insertion at the same √root terminal node. Section 4 sketches an analysis of (1) based on a recent model of grammar outlined in Harley (2014) that is specifically oriented to handle suppletion. Finally, Section 5 offers a brief conclusion.
2. Allomorphy and the status of roots in Distributed Morphology (DM) While allomorphy is an important part of any theory of morphology, the status of allomorphy within the domain of “contentful” vocabulary (or roots) has been of particular importance to development of the theory of Distributed Morphology (DM). For this reason our discussion of suppletion will be cast within this framework. DM is a theory of grammar developed from the original proposals of Halle & Marantz (1993, 1994) and Marantz (1995, 1997), motivated largely by certain shortcomings of lexicalist theories of morphology highlighted in those works. Instead of applying morphological and lexical rules to structured lexical entries which in turn
Fue muerto
feed syntactic derivations/representations, in DM it is syntax that feeds morphology. Syntax manipulates abstract terminal nodes that have grammatical features, not words and morphemes. Those structures are then (i) interpreted by an encyclopedic/ semantic component of the grammar and (ii) the grammatical features of terminal nodes are manipulated in constrained ways by morphological rules before being discharged to vocabulary items prior to PF. Sets of vocabulary items compete for insertion at terminal nodes based on shared sets of features. A vocabulary item that is maximally specified for a subset of the features present at a particular node is the winning candidate for insertion (the subset principle, Halle, 1997). A visual representation of the architecture of grammar in DM is shown in (2) and (3) (adapted from Embick & Noyer, 2007; Halle & Marantz, 1993; Harley & Noyer, 1999; Harley, 2013; Siddiqi, 2009). (2)
{Numeration: subset of list 1} Syntactic operations (MERGE, AGREE, …)
Morphological operations (fission, fusion, …) Vocabulary Insertion (list 2)
Spell-out Interpretation (list 3)
LF
PF
(3) Lists a. List 1: syntactic terminal nodes with feature bundles b. List 2: phonological exponents of terminal nodes c. List 3: instructions for interpreting terminal nodes in context The controversy surrounding the status of roots in DM and whether suppletion exists begins with a set of claims in Marantz (1995). In this paper it is hypothesized that terminal nodes corresponding to “contentful” information, which are labeled as √root in DM, are subject to late insertion just like more “abstract” terminals such as T[past] or Num[plural]. Such a view is compatible with including only features that are relevant for the narrow syntax as part of list 1. The phonological and conceptual content of the vocabulary items dog and cat are not relevant for syntactic operations. Thus, √root terminals are not individuated phonologically or semantically but may have syntactically- relevant features such as [±count] or [±animate]. The key caveat introduced in the paper is that late insertion of vocabulary items at √root terminals is not subject to competition, unlike other terminals. While the vocabulary items -en and -s compete for
Grant Armstrong & Jason Doroga
insertion in the context of Num[plural] with the more highly specified form -en winning in certain contexts (i.e. children, oxen), Marantz (1995) suggests that dog and cat exhibit no such competition. Instead, a speaker is free to choose either vocabulary item, and the choice will have consequences for pronunciation and interpretation. That is, the PF and LF components must be in sync with respect to how cat is pronounced and how it is interpreted so as to avoid pronouncing /kæt/ (list 2) and interpreting dog (list 3). An important consequence of this particular version of late root insertion is that there is no such thing as root allomorphy, or what we have been calling suppletion. A speaker will choose an appropriate vocabulary item to be inserted at a √root terminal after which that particular item may be adjusted in the phonology in certain contexts, accounting for pairs such as break-broke. Marantz (1995, 1997) claims that such a situation is desirable given that from the standpoint of acquisition, the learner is likely to assume that novel “contentful” vocabulary items map to different meanings (see Harley, 2014 for detailed discussion of this point). In addition, cases of radically different forms that are part of the same meaning such as go-went and bad-worse are arguably rare and limited to light, categorizing heads such as v, n and a. Thus, these cases do not constitute examples of distinct vocabulary items competing for insertion under a single √root terminal but rather functional morphemes competing for insertion in v or a in certain local grammatical contexts defined by higher functional nodes such as T[past] and cmpr (comparative). Subsequent work in DM, initiated by Embick (2000), abandons the idea that vocabulary items are inserted late at √root terminals. Instead, roots are treated as individuated entities in list 1 and must be part of the derivation from the very beginning. An example from Embick (2010: 47) serves to illustrate this point. The syntax generates a series of terminals that must be interpreted by the LF and PF branches of the grammar as in (4). (4) Past tense allomorphy a. You hit me T v √HIT
b. I played soccer T
T[past] v
v √PLAY
T[past] v
Note here that √root terminals are not devoid of syntactically irrelevant content; they contain information regarding how √hit and √play should be interpreted, which is accessed in list 3, and they are also endowed with what we might think of as a basic phonological index. This index has repercussions regarding which vocabulary items of list 2 are accessed. (5) a. T[past] ↔ -Ø/__{√hit, √sing, √break, …} b. T[past] ↔ -d/〈elsewhere〉
Fue muerto
Note that the conditions on allomorphy in T make crucial reference to specific roots that are present prior to vocabulary insertion at the functional terminals. The more specific forms such as -Ø (5a) are sensitive to particular roots while the underspecified form -d is an “elsewhere” morpheme.1 In addition to establishing the contexts for insertion of a particular allomorph, it is clear from this discussion that roots have some kind of basic phonological index associated with them. That particular index may be the input to a readjustment rule in the phonology for some roots that take one of the more specific forms of the past tense morphemes such as sing-sang and break-broke. Since there is no late insertion of roots on this particular view, there is no suppletion, only phonological readjustment in a conditioning, local context. Embick (2010: 84–87) adopts the Marantz (1995) view that the most cited cases of suppletion such as go-went and be-was involve insertion of vocabulary items at light verb terminals, vgo and vbe, respectively, which is conditioned by the local grammatical context, mainly tense/aspect and person/number agreement. An alternative to the view described above is outlined in work by Bobaljik & Harley (2014), Haugen & Siddiqi (2013), Siddiqi (2009) and Tubino Blanco (2011), among others. These authors argue that insertion at √root terminals is both late and subject to competition. This claim is motivated in two primary ways. First, suppletion, upon deeper examination, appears to be an empirical reality that extends beyond light verbs, adjectives and nouns (see Veselinova, 2006 for a survey of suppletion in the world’s languages). Second, like allomorphy at functional terminals, suppletion is conditioned by local domains, which indicates that the same set of principles that trigger insertion of one candidate over another at functional terminals are also operative at √root terminals. One striking example of this comes from a suppletive pattern in the Uto-Aztecan languages. In this language family, suppletive verb forms are triggered by number specification of a verbal argument: certain intransitive verbs exhibit suppletion that is conditioned by the number of the subject (6) while in transitive verbs, the same pattern is triggered by the number of the object (7) (Harley & Bobaljik, 2013; Harley, 2013; Haugen & Siddiqi, 2013 and works cited therein).
(6) Hiaki intransitive suppletive verbs: weye-kate (‘walk, go’), vuite-tenne (‘run’), muuke-koko (‘die’), kikte-hapte (‘stand up’), … (Bobaljik & Harley, 2013)
a. Aapo weye b. Vempo kate(Bobaljik & Harley, 2013: 4) 3sg walk.sg 3pl walk.pl ‘S/he is walking’ ‘They are walking.’
. This specific application of the ‘elsewhere principle’ to the insertion of functional morphemes at terminal syntactic nodes has been part of the Distributed Morphology literature since the foundational work of Halle & Marantz (1993). This particular example comes from Embick (2010).
Grant Armstrong & Jason Doroga
(7) Hiaki transitive suppletive verbs: yecha-hoa (‘put down/place’), kivachi-kiima (‘bring in’), me’a-sua (‘kill’), … (Bobaljik & Harley, 2013; Harley, 2013)
a. Aapo/Vempo uka koowi-ta me’a–k(Bobaljik & Harley, 2013: 5) 3sg/3pl the.sg pig-acc.sg kill.sg-prf ‘He/They killed the pig.’ b. Aapo/Vempo ume kowi-m sua-k 3sg/3pl the.pl pig-pl kill.pl-prf ‘He/They killed the pigs.’ In these works it is argued that intransitive suppletive verbs, even ones such as vuitetenne (‘run’), pass unaccusativity diagnostics in the language and thus form a natural class with objects of transitive verbs from a structural perspective. It is further argued that the local relation between this argument and the root is what conditions suppletion. Since the internal argument merges directly with the root it defines a local domain in which the number of the DP has access to the √root terminal node. The reason external arguments do not condition suppletion in Hiaki verbs is because they are introduced by a separate, cyclic head (Voice, or v*), which would place the external argument and the root in separate domains. The local configuration between √root and internal argument delimits the domain that triggers suppletion: the number specification of the DP in the shaded portion of structure may condition a particular form of the root, but any DP outside of that domain may not.
(8) a. [√P [DP …] √walk]
b. [VoiceP [DP …] Voice [√P [DP … ] √kill]
(9) a. √walk ↔ weye /{[√P [DP …][-pl] ___ ] } b. √walk ↔ kate 〈elsewhere〉 (10) a. √kill ↔ me’a /{[√P [DP …][-pl] ___ ] } b. √kill ↔ sua/〈elsewhere〉 As shown in (8)–(10), a plausible analysis of suppletive verbs in Hiaki and other UtoAztecan languages makes crucial use of the idea that vocabulary items compete for insertion at the same √root terminal. This particular version of the analysis is based on the idea that what is accessed as part of list 1 is a root that is a meaningful sign, or a root that is semantically individuated, but lacks phonological content. While such a hypothesis provides a plausible way of accounting for data like (9) and (10), the idea that roots are meaningful signs has been called into question by authors such as Acquaviva & Panigiotis (2012), Borer (2013) and Harley (2014). Two main problems with this idea have been noted. The first is that there are roots in some languages that are individuated phonologically yet have no identifiable meaning outside a particular context. These include the roots of verbs descended from Latin that require prefixes for interpretations such as -duc(ir) (aducir, ‘offer as proof ’; conducir,
Fue muerto
‘drive’; deducir, ‘deduce’; inducir, ‘infer’; reducir, ‘reduce’) and -sum(ir) (asumir, ‘assume’; consumir, ‘consume’; presumir, ‘presume’; resumir, ‘summarize’) among others. If all roots do not have an identifiable meaning outside a particular grammatical context, what is it, precisely, that the items labeled √walk and √kill in (8) are identifying? A way around this problem is to adopt the idea that roots only have a meaning in a particular grammatical context (Arad, 2003). A second problem is that once the move toward individuating roots as meaningful signs is made, a problem presents itself with respect to how many vocabulary items could conceivably compete for insertion under a specific node. For example, if √change is a meaningful sign that can be spelled out by different vocabulary items in different contexts, then should we say that in English, change, transform and morph all compete for insertion at the same √root terminal? (see Borer, 2014 for an explicit argument against such an idea based on data from Modern Hebrew). Borer (2013), following Arad (2003), has proposed that roots do not constitute meaningful signs and should be individuated phonologically, without radical readjustment rules. The meaning of a given root is determined by the grammatical context in which it appears. While this avoids the problems cited above, Harley (2014) notes that it creates a new set of problems that are perhaps even more concerning. In the case of go-went, we must assume that √went is a distinct root than √go, individuated phonologically, and that it has a meaning in a precisely complementary set of contexts than √go. Thus, the fact that they share a meaning and are in complementary distribution is left largely to coincidence. Harley (2014), following work by Acquaviva (2009) and Pfau (2009), offers an alternative, claiming that roots are not individuated in the syntax through phonological indices or meaning. Instead √root terminals are represented as numeric indices that provide a series of instructions at the PF and LF interfaces. The case of suppletion in the Hiaki transitive verb kill is handled as in (11) and the different contextual meanings associated with the vocabulary item throw are shown in (12). (11) Interface instructions for √567 in Hiaki PF Instructions (list 2) a. √567 ↔ me’a /{[√P [DP …][-pl] ___ ] } b. √567 ↔ sua/〈elsewhere〉
LF Instructions (list 3) √567 ↔ “kill”
(12) Interface instructions for √77 in English PF Instructions (list 2) √77 ↔ /θrow/
LF Instructions (list 3) a. √77 ↔ “vomit”/{[v [√ __ ] [up]P ]vP] b. √77 ↔ “blanket”/{[n [√ __ ]] c. √77 ↔ “throw”/〈elsewhere〉
Harley’s (2014) proposal represents an important synthesis of previous work regarding the nature of roots and the status of suppletion. By conceiving syntactic roots as
Grant Armstrong & Jason Doroga
indices that contain instructions to the interfaces, the issues associated with radical readjustment in cases like go-went as well as the controversies associated with how to individuate roots in the syntax can be dealt with in a plausible way. It also presents us with a new set of challenges. Perhaps the most important question in this model is the following: given some syntactically-active numeric index in some language L, say √56 in Spanish, how do we go about determining the nature of the lists (2 and 3 above) accessed by that index? The question appears to be easier to answer in a case like (12), where a single set of phonological instructions has multiple encyclopedic instructions depending on the context. In a case like (11), where two sets of phonological instructions correspond to a single set of encyclopedic instructions, how do we go about determining that those items in particular are the items that are accessed as part of list 2? For cases like (11), there are two guiding principles that serve to motivate treating two or more vocabulary items as members of a list accessed by a single numeric index (Harley, 2014): (i) co-variation of contextually-determined meanings and (ii) blocking. In clear-cut cases of suppletion such as go-went, the contextually-determined meanings involving the root do not change according to whether the vocabulary item go or went is present as in (13). With respect to point (ii) above, the form *goed is blocked since went is specified to occur in the context of T[past]. (13) a. John goes/went/*goed to town b. The milk goes/went/*goed bad c. The teacher went/goes/*goed bananas These guiding principles can be used to avoid using “broad synonymy” as a potential way of generating an instantiation of list 2 accessed by a particular numeric index. Now that the relevant portions of the recent controversies regarding suppletion have been presented, along with some guiding principles regarding how to identify suppletion in Harley’s (2014) model, we turn to a case of heretofore unanalyzed suppletive forms in Spanish.
3. Killing and dying in Spanish The relation between words that correspond to the concepts kill and die has a special place in the history of Generative Grammar. In Generative Semantics, it was explicitly proposed that the English words kill and die are derived from the same underlying syntactic structure (McCawley, 1968; though see Fodor, 1970 for argumentation that ‘kill’ does not correspond to the literal string ‘cause to die’), which was part of a more general analysis applied to causative-inchoative pairs such as breakTR-breakINTR. The fact that some languages contain a single vocabulary item meaning die, which is augmented by
Fue muerto
a causative morpheme to form the transitive verb meaning kill, can be taken as preliminary evidence that some languages exhibit kill-suppletion whereas others do not. (14) kill-suppletion languages: Spanish, English a. Juan mató (a) la mosca b. Murió la mosca Juan killed a the fly Died the fly ‘John killed the fly’ ‘The fly died’ (15) No kill-suppletion languages: Yucatec Maya2 (shown below), Quechua a. Táan u kíim-il le miis-o’ b. Táan in kíin-s-ik le miis-o’ prog 3s die-imp the cat-suf prog 1s die-cause-imp the cat-suf ‘The cat is dying’ ‘I am killing the cat’ Now let us consider one way of analyzing kill-suppletion. This analysis is based on two observations: (i) (14a) entails (14b), so there is a shared core of meaning that could correspond to the list accessed in the encyclopedia and (ii) there is a potential blocking mechanism whereby a causative light verb, vcause, will block insertion of the vocabulary item corresponding to the unaccusative verb. Given Harley’s (2014) model, this would look something like (16), where the two vocabulary items mat- and mor- compete for insertion at a hypothetical root/numeric index called √m16. (16) Root √m16: PF Instructions (list 2) LF Instructions (list 3) a. √m16 ↔ mat-/{[√ ___ [vcause ]]} √m16 ↔ “kill”/{[√ ___ [vcause ]]} b. √ m16 ↔ mor-/〈elsewhere〉 √m16 ↔ “dead”/〈elsewhere〉 While certainly plausible, the inclusion of causative-non-causative pairs of phonologically distinct verbs within the overall typology of suppletion has been questioned for a number of reasons (for recent criticisms see Borer, 2014). For one, many pairs such as feed-eat do not actually exhibit the correct entailment pattern: it is possible to feed someone without them actually eating anything. This militates against an approach that derives these two roots from the same grammatical structures and treats them as independent formatives in the grammar. In Harley’s (2014) approach this would mean that those two vocabulary items and their respective meanings are accessed by distinct syntactically-active numeric indices. How do we know that the same should not apply in the case of mat- versus mor-? Furthermore, how do we know that other vocabulary items that share similar meanings such as fallec(er) (‘die’), or even the idiomatic phrase estirar la pata (literally: ‘stretch the leg’, meaning: ‘die’) are not part of the list accessed by the hypothetical root √m16? In our view, it boils down to the fact that in cases
. Note that there is phonologically-conditioned allomorphy here as the nasal consonant /m/ assimilates in (15b) and is realized as [n].
Grant Armstrong & Jason Doroga
such as kill-die, the context that triggers insertion of one form over the other is also the context that is instrumental in determining aspects of the construction typically associated with what we might intuitively label “basic conceptual meaning” (agency, causation, etc.). In other cases, the triggering context is outside of this basic domain (Tense/Aspect; number agreement, etc.). Therefore, stronger cases for suppletion in (14) should involve some kind of independent reason to think that the vocabulary items mat- and mor- are accessed by the same numeric index in the language that goes beyond the causative-inchoative alternation. We claim that there are three areas that provide more solid evidence for treating mat- and mor- as competing members of a list accessed by the same numeric index. The first is that past participles of mat- (matado, ‘killed’) and mor- (muerto, ‘dead’), in the contexts of active perfects and eventive passives, have been in overlapping distribution in the history of the language and, for many modern speakers, demonstrate blocking effects typical of suppletive forms. In Old Spanish, the past participle of matar in both active (17a) and passive (17b) contexts surfaced as muerto (Corominas & Pascual, 1980–91). (17) a. Capitulo.xxxj. que muestra commo auariçia ha muertos & condepnados grandes omnes (Libro de las donas, 1448) ‘Chapter 31 shows how avarice has killed and condemned great men’ b. En el obispado de Nicaragua fueron prelados: primero, D. Fr…dominico, que fue muerto por los dos hermanos Contreras (Historia eclesiástica indiana, 1596) ‘In the bishopric of Nicaragua there were prelates: first, D. Fr. Dominico, who was killed by the Contreras brothers’ In modern Spanish, muerto does not productively appear in active perfects (18a), but continues to be, for the vast majority of speakers, the only way of passivizing matar (18b). (18) a. He muerto una liebre3 I have died a hare ‘I have killed a hare’
(Archaic entry from the DRAE)
. In both Catalan and Portuguese, the past participle mort/morto may also appear in active perfects meaning to have killed, but is more productive than in modern Spanish. We thank Eulàlia Bonet (Catalan) and Patrícia Amaral (Portuguese) for informing us of this. María Cristina Cuervo (p.c.) relates this to the idea that the two roots mat- and mor- are of different ontological types: mat- is a manner root while mor- is a result root. Perfective aspect would condition a result root (where available) and since mat- and mor- are arguably competing for insertion in the same syntactic structure, it might be a question of ontological compatibility of one (mor-) in a particular grammatical context. These interesting questions will have to await further investigation.
Fue muerto
b. Un ciclista fue {muerto/%matado4} en la calle A cyclist was died/killed in the street ‘A cyclist was killed in the street’ This claim is corroborated by the results of a comparison of active perfects and passives of matar in a corpus of modern newspaper Spanish. This is a corpus we compiled of modern journalistic prose from the past five years that included the Spanish newspaper El País, the Mexican newspaper El Universal and the Argentine newspaper La Nación. A summary of the data is provided in Table 1. The participle muerto is virtually excluded from active perfect constructions meaning ‘have killed’, yet it is overwhelmingly preferred in analytic passive constructions ‘be killed.’ Table 1. Active perfects and passives of ‘matar’ Participial construction type
matado (%)
muerto (%)
Active perfects (auxiliary = haber)
122 (99.1%)
1 (0.9%)
Passives (auxiliary = ser)
16 (7.6%)
194 (92.4%)
In order to include other genres besides journalistic prose, we consulted El Corpus de Referencia del Español Actual (CREA), which includes a wide variety of both written and oral genres. Data from the CREA corpus further support the idea that muerto is the most widely used form in analytic passives meaning ‘be killed.’ (19) a. Example of ‘fue muerto’ search (CREA) – 113 hits del asesinato del mahatma [Gandhi, que fue muerto por un fanático] que lo acusaba de haber permitido la dimisión del territorio ‘Gandhi, who was killed by a fanatic’ (ORAL – Informe semanal 3/11/84 – TVE-1, CREA search 4/20/2013)
. We use % to indicate variable acceptability. Some speakers reject matado completely while others find it marginally acceptable. Others may accept both muerto and matado with no intuitive difference in meaning. Additionally, some speakers have communicated that there is a difference in meaning between the two but have trouble saying exactly what that is. We feel that in this latter case there might simply be a sense in which they mean different things because they are different forms and speakers sometimes attempt to link different meanings to different forms even if they don’t exist. There is much to investigate here, but we will limit ourselves to a description and analysis of the grammars that use muerto and not matado in passives as a first step in a more complete characterization of the phenomenon.
Grant Armstrong & Jason Doroga
b. Example of ‘fue matado’ search (CREA) – 8 hits Gaudí, cuyo estilo sinuoso, sensual y revolucionario puede ser visto por toda Barcelona (él la convirtió en su ciudad) fue matado por un tranvía … ‘Gaudí … was killed by a streetcar’ (Carlos Fuentes, El espejo enterrado, CREA search 4/20/2013) The data in (18) – (19) and Table 1 indicate that the connection between mat- and morextends beyond the causative alternation and constitutes an empirical reality in the realm of participles from both a diachronic and a synchronic perspective. Diachronically, it appears that the syntactic environment of past participles (active perfects and eventive passives) commonly blocked the presence of the vocabulary item mat- while synchronically it is the combination of past participle and unaccusative syntax that blocks mat-. The generalization that unaccusative environments block the insertion of matcan be further corroborated by the range of se constructions in which matar and morir may appear. As discussed in detail in Mendikoetxea (1999) and Schäfer (2008), transitive verbs that describe caused changes of state may be partitioned according to whether they admit anticausatives (or inchoatives) marked by the reflexive clitic se or not. Following the well-known observation made by Levin & Rappaport Hovav (1995) for English causative verbs, Mendikoetxea (1999) demonstrates that there is a strong correlation between the semantic range of external arguments a transitive verb may take and its ability to yield a marked anticausative. Verbs such as romper (= break), which may appear with agent, instrument or causer external arguments, may yield a marked anticausative form whereas verbs such as cortar (‘cut’), which only appear with agents, do not yield marked anticausative forms. This is shown in (20) and (21). (20) a. b.
Juan/este martillo/el viento rompió la ventana Juan/this hammer/the wind broke the window. ‘Juan/this hammer/the wind broke the window.’ La ventana se rompió The window se.3s broke ‘The window broke’
(21) a. El peluquero/#la maquinita cortó el pelo de Juan The barber/the clippers cut the hair of Juan b. *El pelo de Juan se cortó The hair of Juan se.3s cut Intended: ‘Juan’s hair cut’ The case of matar is interesting in this respect because it admits a range of external arguments that extends beyond agents, like romper (‘break’) (22a), but does not have a marked anticausative (22b).
Fue muerto
(22) a. Pedro/ una bala perdida/ el alcohol mató a José Pedro/ a bullet lost/ the alcohol killed acc José ‘Pedro/a stray bullet/alcohol killed José’ b. *José se mató José se.3s killed Intended: ‘José killed’ (possible meaning: ‘He killed himself ’)5 This fact can be accounted for if mat- is blocked from appearing in unaccusative environments. What is even more striking is that morir may appear with or without se as in (23). (23) a. Murió mi tío Died my uncle ‘My uncle died’ b. Se murió mi tío se.3s died my uncle ‘My uncle died’ Though there is no consensus on precisely what the difference in meaning between (23a) and (23b) is, it has been noted that (23b) shares certain grammatical properties with anticausatives such as disallowing bare nouns in the post-verbal position (see Cuervo, 2014; De Miguel & Fernández Lagunilla, 2000 for discussion). Other nondirected motion unaccusative verbs6 such as crecer (= grow) and fallecer (= die/perish) do not appear in constructions like (23b). If we posit that (23b) is essentially an anticausative and mat- is blocked from unaccusative syntactic constructions generally, then the unacceptable (22b) and the acceptable (23b) can be accounted for in the same way. This explanation, however, would only be possible if mat- and mor- compete for insertion at the same √root node. A third piece of evidence in favor of the idea that the vocabulary items mat- and mor- are accessed as part of the same list can be found by looking at contextually determined idiomatic interpretations.7 As mentioned above in the case of go-went, the strongest cases of suppletion involve strict co-variance of idiomatic readings (see (13) above). Certain difficulties present themselves in the case of mat- and mor- because there appear to be idiomatic readings that depend crucially on transitive versus
. Some dialects, including Mexican Spanish, have a specialized use for matar-se that is not reflexive. It means ‘get killed in an accident.’ . Some directed motion verbs do permit se: ir-se (‘go’), salir-se (‘leave’), subir-se (‘go up’), bajar-se (‘go down’). See Cuervo (2014) for an analysis of these. . We thank an anonymous reviewer for pointing out the importance of such facts for the analysis and urging us to look into them.
Grant Armstrong & Jason Doroga
unaccusative syntax (see Ramchand, 2008: 84–85 for similar observations regarding unselected arguments in transitive versus anti-causative/inchoative verbs). Some idiomatic expressions are only possible with matar (24) or morir (25), respectively. (24) a. matar una esquina/#morir una esquina kill a corner/die a corner ‘to bevel/to cut a corner at a slant’ (terminology used in construction/ architecture) b. matar el tiempo/#morir el tiempo kill the time/die the time ‘to pass time idly’ (25) a. #matar a alguien de sed/morir alguien de sed kill a someone of thirst/die someone of thirst ‘to experience thirst intensely’ b. #matar a alguien vestido/morir alguien vestido kill a someone dressed/die someone dressed ‘to die violently’ Importantly, there are other cases in which special meanings are maintained with both mat- and mor- as in (26). Note that substituting a “synonym” such as fallec(er) in such contexts breaks up the idiomatic interpretation, adding further credence to the idea that mat- and mor- are accessed by the same numeric index to the exclusion of other vocabulary items. As we mentioned in Section 2, divorcing lists from meaningful signs in lieu of numeric indices allows us to avoid this problem.8 (26) a. estar matado/muerto to be killed/dead ‘to be extremely tired’ b. matar/morir el fuego kill/die the fire ‘to extinguish/put out the fire’ c. matar-se por algo/morir-se por algo kill-se for something/die-se for something ‘to try hard to get something/to desire something’ Given the data in (24)–(26), one might question whether there truly is co-variance in meaning which is held across both forms. However, as we mentioned above, in cases like go-went, co-variance in idiomatic interpretations is absolute since the argument
. It should be noted that the data in (26) would constitute weak evidence in favor of treating mat- and mor- as competitors for insertion at the same root node since it involves a special interpretation of the root rather that than root+object. We thank Heidi Harley (p.c.) for informing us of this.
Fue muerto
structure properties of go have already been determined by the time the suppletion triggering information, T[past], is merged. Following Marantz (1997), in DM, idiomatic interpretations are essentially parts of list 3 that are accessed by syntactic chunks that are larger than √roots. From this perspective, it is possible that certain chunks of structure built with √m16 that have unaccusative syntax, and thus trigger insertion of mor-, have special meanings that are independent of other chunks of transitive syntax built on top of √m16 that trigger the insertion of mat-. The fact that a subset of these idiomatic interpretations is maintained across the unaccusative-transitive syntactic boundary would be surprising if the two vocabulary items were inserted at independent numeric indices. To sum up, we have shown that beyond the controversial arguments associated with the causative alternation, there are many reasons to believe that mat- and mor- truly are vocabulary items that are linked to the same syntacticallyactive numeric index. Let us now propose an analysis that captures the distribution of these two vocabulary items.
4. Analysis If, as we argued in Section 3, the vocabulary items mat- and mor- compete for insertion at the same syntactically-active root node (=√m16), then the next step is discerning what the instructions from lists 2 and 3 are for spelling out and interpreting that node in different contexts. Our main hypothesis is that mat- is a more highly specified vocabulary item than mor- in that it is conditioned to appear as sister to an agentive verbalizing head, which we call v*, following Chomsky (2001). In all other contexts mor- appears, making it the elsewhere morpheme. This is spelled out in (27). (27) Root √m16: PF Instructions (list 2) LF Instructions (list 3) a. √m16 ↔ mat-/{[√ ___ [v*]]} √m16 ↔ “kill”/ {[√ ___ [v*]]} b. √ m16 ↔ mor-/ 〈elsewhere〉9 √m16 ↔“dead”/ 〈elsewhere〉
. We are abstracting away from some important details for ease of presentation. First, as argued in Oltra-Massuet (2013), and literature discussed therein, spell out of v (and any other functional head) in Romance triggers an additional node in which the theme vowel is inserted for language-internal well-formedness reasons. We have not represented the position [Th] here. Second, the actual forms that are inserted for mor- include the allomorphs mur- (23b) and muer- (25), which are conditioned by phonological and grammatical factors, and arguably muert- and mort-, which appear in nominal environments as illustrated in (31)–(32). A more detailed version of (27) would include: mat-, mor-, which may surface as muer- when the root vowel is stressed and mur- when the root vowel is stressed in the presence of T[past] along with 3rd person agreement, and finally, when √M16 is directly nominalized, there is a distinct allomorph in list 2, mort-, which may surface as muert- when the root vowel is stressed. Going
Grant Armstrong & Jason Doroga
The simple cases involving transitive and unaccusative syntactic structures built on top of √M16 are shown in (28). (28) Transitive versus active unaccusative a. Transitive b. Unaccusative El policía mató al fugitivo Murió el fugitivo ‘The police killed the fugitive’ ‘The fugitive died’ vP
v*P
v*΄
DP v*
v √P
√M16
√P
√M16
DP
DP
After head movement generates a head adjunction structure that can be interpreted by the instructions in (27), it is only in (28a) that √M16 is sister to v*, and thus only in this context will mat- be inserted. Evidence in favor of this proposal comes from an examination of a wider range of environments in which mat- appears. In (29), we show three different nominalizations. The prediction is that mat- should never appear if √M16 is directly nominalized. Any noun containing this particular numeric index spelled out as mat- must have come from a structure in which √M16 is sister to v*. (29) mat-a-nza mat-a-dor mat-a-dero mat-v*[Th]-nom mat-v*[Th]-nom mat-v*[Th]-nom ‘slaughter’ ‘killer/matador’ ‘slaughter house’ This prediction obtains in two ways. First, the morphological structure of mat-nouns always contains a theme vowel that is identical to the one that appears with the verb. Adopting Oltra-Massuet’s (2013) proposal that theme vowels are overt realizations of functional nodes in Romance, this can be taken as evidence of the presence of v*. Second, the interpretation of mat-nouns involves agentive causation of death (“killing”), which is precisely what we expect if list 3 accesses this particular interpretation when √M16 is a sister to v*. Now, consider the range of possibilities for mor-. Our prediction is that there should be a much wider range of insertion possibilities for mor- since it is essentially an elsewhere morpheme for this particular numeric index. The noun muerte describes the “act of dying” and is derived from an unaccusative verbal source, inheriting the event and argument structure associated with vP. into the minute details of this is beyond the scope of the paper and would lead us astray of the main idea that mat- and mor- (+ other allomorphs) are competing for insertion at the same √root terminal node.
Fue muerto
(30) La muerte del toro a las 3 de la tarde The death of.the bull at 3 in the afternoon There are other nouns, however, which are not eventive in any useful sense and thus are most likely derived by merging a particular type of n directly with √M16. These include the nouns mortal (‘a mortal’) and mortaja (‘shroud’). Given that there is no evidence of any event structure in these nouns, we propose that they are derived by directly merging n with √M16 as follows. (31) Direct nominalizations of √M16 a. mortal b. mortaja ‘(a) mortal’ ‘shroud’ nP nP n -al
√M16
n -aj
√M16
Since √M16 is sister to n in (31), we predict that mor- will be inserted since this is a context that falls outside the one for which mat- is specified for insertion. In order to explain the presence of mort- (rather than mor-), we must additionally claim that there is an allomorph mort- that is specified to appear when √M16 is sister to n. There are a number of adjectives and verbs that are further derivations of this particular nominal stem as shown in (32). (32) [mort]n al]a; [mort]n ífero]a; [mort]n ecino]a ; [mort]n ificar]v ‘deadly’ ‘deadly’ ‘dim/dull/dying’ ‘mortify’ The adjective mortal (‘mortal/deadly’) contains an adjectivizing morpheme -al that only combines with nouns, supporting the idea that mort- is the root inserted when √M16 is directly nominalized. In addition, a verb like mortificar (‘mortify’) is a causative verb but it obviously does not mean ‘to kill/cause to die’ but rather ‘cause a sensation related to death/to horrify.’ If the verb is derived from an already formed noun, which restricts the interpretation of √M16 to something like ‘death’ or ‘death-related sensation’, then we can account for why the causative morphology present in the verb cannot alter this meaning in any way. Though we have abstracted away from the origin of the different vocabulary items mor-, mur-, muer- and mort-, the main point is that the treatment of mor- as an elsewhere allomorph can account for its distribution in a wider range of syntactic environments than mat-, which is specified to appear in only one. With the basics of the hypothesis in place, we now turn to the original puzzle that was presented at the start of the paper: why does muerto appear in the passive? It is not obvious based on (27) why this should be the case. This is because the primary difference between v and v* is that the latter introduces a predicate such as agent that can be saturated by the external argument while the former does not. In the case of passives, we want to say that the argument position of the agent predicate is still there, but it is existentially quantified and thus introduces semantically relevant information
Grant Armstrong & Jason Doroga
into the syntactic structure even though no argument position (specifier) is projected. This means that passives must have something in common with v*, even though they are syntactically unaccusatives. We might distinguish between these two different types of agentive light verbs, both v* according to the criterion above, by using the binary feature [± pass]. One way of accounting for muerto in the passive is to make the rule in list 2 more specific with something such as √m16 ↔ mat-/{[√ ___ [v*[-pass]]]}. This rule makes some incorrect predictions. First, we have certain nominalizations such as matadero (‘slaughterhouse’) in which v* in no obvious way projects a specifier for an agent argument. Second, passive interpretations are available in se constructions as in (33). (33) En mi pueblo se matan cochinos para la fiesta del In my town se.3s kill-3pl pigs for the party of.the año nuevo new year ‘In my town, pigs are killed/slaughtered for New Year’s’ If se suppresses the external argument in (33), it would provide a counterexample to the rule proposed above. These two observations indicate that it is not the lack of a syntactic external argument that is the only conditioning factor involved in muerto passives. As discussed in Section 3, diachronic evidence shows that it is past participial morphology that is crucial for accounting for why muerto appears in synchronic passives. Following Embick (2003, 2004), we treat past participles as aspectual projections with the feature [perf]. The difference between active perfects and eventive passives is shown in (34). Active perfects project a full argument structure and condition the auxiliary haber (‘have’) while eventive passives do not project a full argument structure and condition the auxiliary ser (‘to be’). (34) Syntactic constructions involving past participles a. Active perfect vPaux
vaux HABER
AspP v*P
Asp [perf]
v*'
DP v* [-pass]
√P √M16
DP
Fue muerto
b. Eventive passive vPaux
vaux SER
AspP v*P
Asp [perf] v* [+pass]
√M16
√P DP
The main question now concerns how such structures are spelled out and interpreted. Obviously, the instructions at LF will not eliminate or change any of the functional nodes present in (34); however, morphological operations such as fusion, impoverishment and local dislocation may change the nature of the syntactic structure to be interpreted at PF (see the diagram in (2) and the discussion in Embick & Noyer, 2007). In the case of past participles, such alterations to the syntactic input are common. For example, in Spanish many past participles ending in -to/ta lack the presence of a theme vowel (i.e. romper-roto (‘break-broken’); abrir-abierto (‘open-opened’)) while regular participles ending in -do/da maintain the theme vowel (i.e. quebrar-quebrado (‘break-broken’); comer-comido (‘eat-eaten’)). While such “irregular” participles have their origins in Latin, a plausible synchronic explanation for the form they take is that Asp and v* fuse into a single terminal node for the purposes of vocabulary insertion. The reason why roto (‘broken’) lacks a theme vowel is because v* and Asp[perf] fuse into a single terminal node, Asp, prior to vocabulary insertion. The -to/ta10 morpheme is a realization of this node. In cases of “regular” participles, v* is realized as the appropriate theme vowel and Asp is realized as -do/da. What this means in cases like roto (‘broken’) and abierto (‘open’) is that v* is no longer visible to vocabulary insertion and the set of instructions in list 2, though it is certainly part of the interpretation that is assigned to the structure in list 3. Taking these observations into consideration, let us return to muerto passives, shown in (35). (35) El fugitivo fue muerto (por el policía) The fugitive was killed (by the policeman)
. We are abstracting away from the AGR node that is generated in the case of the passives but not in the active perfects. In the passives, Asp would be realized as -t or -d, respectively, and a separate AGR node is realized with the person and number features of the argument that agrees with T.
Grant Armstrong & Jason Doroga
All we have to say to account for these sentences is the following: Asp and v*may fuse to form a single node post-syntactically, Asp[perf], prior to vocabulary insertion. The LF branch of the sentences is not altered as this structure contains v*and thus is interpreted as “kill.” However, the PF branch contains a structure in which the √M16 is not adjacent to v*, triggering the elsewhere form mor-. This is represented in (36). (36)
Asp v* √M16
v* [+pass]
Asp [perf]
Fusion: Asp √M16
Asp [perf]
PF Instructions (list 2)
Root √M16:
√M16 ↔ mat-/{[√ ___ [v*]]} √ M16 ↔ mor-/
LF Instructions (list 3) √M16 ↔”kill”/{[√ __[v*]]} √M16 ↔ “dead”/
Let us briefly touch on the diachronic data mentioned in Section 3 as well as in Footnote 1. In modern Spanish, fusion is restricted to v*[+pass] with Asp, but it may also happen with v*[-pass] and Asp, which explains the diachronic situation as well as the acceptability of the equivalent of muerto in active perfects in both Catalan and Portuguese. The explanation in those cases would be identical. Fusion of v and Asp is common in past participles in Romance and it leads to the spell out of the participle in the absence of a theme vowel. In the case of √M16, this operation has consequences beyond simply eliminating the theme vowel. It changes how the structure is interpreted at PF since the fusion operation eliminates a sisterhood relation between √M16 and v*, which would condition the insertion of mat-, and replaces it with one in which √M16 is the sister to Asp, which conditions the insertion of mor-.
5. Conclusion In this paper we have tackled the controversial issue of whether suppletion, or root allomorphy, exists in human language. We have shown based on a
Fue muerto
g rammatically-conditioned alternation between the words typically associated with the meanings kill (matar) and die (morir) that suppletion is an empirical reality in Spanish and we provided an analysis of it using new insights by Harley (2014). This particular analysis, if on the right track, adds support to a broad view of grammar in which words are not atomic units that are opaque to syntactic operations but rather are products of syntactic operations that must be interpreted at the interfaces, which is encompassed by many frameworks such as Borer’s (2013) exo-skeletal model, DM, and Nanosyntax (Caha, 2009; Fábregas, 2014; Starke, 2009). Beyond this, the details discussed in Section 3 add credence to the idea that roots are not individuated phonologically or semantically in the syntax, which is a point of debate among practitioners of these various models.
Acknowledgements We would like to thank the audience members at HLS 2013 (University of Ottawa) for helpful comments. We would also like to thank Heidi Harley and two anonymous reviewers for insightful commentary/criticisms that have improved this paper. All errors are our responsibility.
References Acquaviva, P. (2009). Roots and lexicality in distributed morphology. In A. Galani, D. Redinger, & N. Yeo (Eds.), York-Essex Morphology Meeting 2, 1–21. York: University of York. Acquaviva, P., & Panagiatis, P. (2012). Lexical decomposition meets conceptual atomism. Available at lingbuzz/001558 Arad, M. (2003). Locality constraints on the interpretation of roots: The case of Hebrew denominal verbs. Natural Language and Linguistic Theory, 21, 737–778. Bobaljik, J. (2000). The ins and outs of contextual allomorphy. In K. Grohmann & C. Struijke (Eds.), Proceedings of the Maryland Mayfest on Morphology 1999 (University of Maryland Working Papers in Linguistics 10; pp. 35–71). College Park, MD: University of Maryland, Department of Linguistics. Bobaljik, J. (2012). Universals in comparative morphology. Cambridge, MA: The MIT Press. Bobaljik, J., & Harley, H. (2013). Suppletion is local: Evidence from Hiaki. Available at lingbuzz/001982 Borer, H. (2013). Structuring sense, volume III: Taking form. Oxford: Oxford University Press. DOI: 10.1093/acprof:oso/9780199263936.001.0001 Borer, H. (2014). Derived nominals and the domain of content. Lingua, 141, 71–96. DOI: 10.1016/j.lingua.2013.10.007 Caha, P. (2009). The nanosyntax of case. Unpublished Doctoral Dissertation. University of Tromso, Norway. Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A life in language, 1–52. Cambridge, MA: The MIT Press. DOI: 10.1017/s0022226704322747
Grant Armstrong & Jason Doroga Corominas, J., & Pascual, J. (1980–91). Diccionario crítico etimológico castellano e hispánico. Madrid: Gredos. Cuervo, M.C. (2014). Alternating unaccusatives and the distribution of roots. Lingua, 141, 48–70. DOI: 10.1016/j.lingua.2013.12.001 De Miguel, E., & Fernández Lagunilla, M. (2000). El operador aspectual se. Revista Española de Lingüística, 30, 13–41. Embick, D. (2000). Features, syntax and categories in the Latin perfect. Linguistic Inquiry, 31, 185–230. DOI: 10.1162/002438900554343 Embick, D. (2003). Locality, listedness, and morphological identity. Studia Linguistica, 57, 143–169. DOI: 10.1111/j.0039–3193.2003.00102.x Embick, D. (2004). On the structure of resultative participles in English. Linguistic Inquiry, 35, 355–392. DOI: 10.1162/0024389041402634 Embick, D. (2010). Localism versus globalism in morphology and phonology. Cambridge, MA: The MIT Press. DOI: 10.7551/mitpress/9780262014229.001.0001 Embick, D., & Noyer, R. (2007). Distributed Morphology and the syntax/morphology interface. In G. Ramchand & C. Reiss (Eds.), The Oxford handbook of linguistic interfaces, 289–325, Oxford: Oxford University Press. DOI: 10.1093/oxfordhb/9780199247455.013.0010 Fábregas, A. (2014). Argument structure and morphologically underived nouns in Spanish and English. Lingua, 141, 97–120. DOI: 10.1016/j.lingua.2013.02.011 Fodor, J. (1970). Three reasons for not deriving ‘kill’ from ‘cause to die’. Linguistic Inquiry, 1, 429–438. Halle, M. (1997). Distributed Morphology: Impoverishment and fission. In B. Bruening, Y. Kang, & M. McGinnis (Eds.), MITWPL 30: Papers at the Interface (pp. 425–449). Cambridge, MA: MIT. Halle, M., & Marantz, A. (1993). Distributed Morphology and the pieces of inflection. In K. Hale & S.J. Keyser (Eds.), The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger (pp. 111–176). Cambridge, MA: The MIT Press. Halle, M., & Marantz, A. (1994). Some key features of Distributed Morphology. In A. Carnie, H. Harley, & T. Bures (Eds.), Papers on phonology and morphology (MIT Working Papers in Linguistics 21; pp. 275–288). Cambridge, MA: MIT. Harley, H. (2014). On the identity of roots. Theoretical Linguistics, 40, 225–276. Harley, H., & Noyer, R. (1999). State of the article: Distributed Morphology. GLOT, 4, 3–9. DOI: 10.1515/9783110890952.463 Haugen, J., & Siddiqi, D. (2013). Roots and the derivation. Linguistic Inquiry, 44, 493–517. DOI: 10.1162/ling_a_00136 Levin, B., & Rappaport-Hovav, M. (1995). Unaccusativity: At the syntax lexical semantics interface. Cambridge, MA: The MIT Press. DOI: 10.1017/s0022226796276571 Marantz, A. (1995). A late note on late insertion. In Y–S. Kim et al. (Eds.), Explorations in generative grammar (pp. 357–368). Seoul: Hankuk. Marantz, A. (1997). No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In A. Dimitriadis, L. Siegal, C. Surek-Clark, & A. Williams (Eds.), Proceedings of the 21st Penn Linguistics Colloquium (pp. 201–225). Philadelphia, PA: University of Pennsylvania, UPenn Working Papers in Linguistics. McCawley, J. (1968). Lexical insertion in a transformational grammar without Deep Structure. In B. Darden et. al. (Eds.), Fourth regional meeting of the Chicago Linguistics Society (pp. 71–80). Chicago, IL: BLS.
Fue muerto
Mendikoetxea, A. (1999). Construcciones inacusativas y pasivas. In I. Bosque & V. Demonte, Gramática descriptiva de la lengua española, Volumen 2, 1575–1629. Madrid: Espasa-Calpe. Oltra-Massuet, I. (2013). Deverbal adjectives at the interface: A crosslinguistic study into the morphology, syntax and semantics of -ble. Berlin: Mouton de Gruyter. DOI: 10.1515/9781614510659 Pfau, R. (2009). Grammar as processor: A Distributed Morphology account of spontaneous speech errors. Amsterdam: John Benjamins. DOI: 10.1075/la.137 Ramchand, G. (2008). Verb meaning and the lexicon: A first phase syntax. Cambridge: Cambridge University Press. DOI: 10.1017/s0022226709990363 Real Academia Española. (2009). Nueva gramática de la lengua española. Madrid: Espasa-Calpe. DOI: 10.1353/hpn.2013.0037 Real Academia Española: Banco de datos (CREA) [en línea]. Corpus de referencia del español actual. 〈http://www.rae.es〉 (Consulted 20 April 2013, 13 October 2013). Schäfer, F. (2008). The syntax of (anti-)causatives. Amsterdam: John Benjamins. DOI: 10.1075/la.126 Siddiqi, D. (2009). Syntax within the word: Economy, allomorphy and argument selection in Distributed Morphology. Amsterdam: John Benjamins. DOI: 10.1075/la.138 Starke, M. (2009). Nanosyntax: A short primer to a new approach to language. In P. Svenonius et al. (Eds.), Nordlyn 36(1), Special issue on nanosyntax. Septentrio: Tromso. Tubino Blanco, M. (2011). Causatives in minimalism. Amsterdam: John Benjamins. DOI: 10.1075/la.179 Veselinova, L. (2006). Suppletion in verb paradigms: Bits and pieces of the puzzle. Amsterdam: John Benjamins. DOI: 10.1075/tsl.67
Temporal and spectral dependencies in the processing of Spanish and English stop consonant voicing Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz Purdue University
Previous findings show that listeners do not rely on onset f0 when processing stop consonant voicing with lead VOT (Llanos et al., 2013). According to a temporal-based account, onset f0 has no effect because the voicing decision is biased by the presence of prevoicing that precedes the occurrence of onset f0 in the signal. According to a spectral-based account, onset f0 has no effect because its contribution to low frequency energy is insignificant compared to that of VOT in voicing lead. Specific predictions of the two hypotheses for the whole VOT continuum were evaluated in English and Spanish voicing decisions. Results support the spectral-based account, demonstrating that the peak effect of onset f0 occurred in both languages at the VOT values where onset f0’s contribution to the low frequency energy would be most appreciable. Keywords: VOT; onset f0; stop consonant voicing; auditory enhancement; spectral processing; logistic deviance
1. Introduction Speech sounds are complex acoustic entities composed of multiple acoustic correlates whose realization in the speech signal may vary systematically across different phonetic categories (McMurray & Jongman, 2011; Repp, 1982). These correlates may occur at different times within the signal even when they cue the same phonological property. For example, in voicing of stop consonants, the acoustic properties of voice onset time (VOT, the time between the burst release and the onset of voicing) and the duration of the following vowel are temporally asynchronous – VOT is necessarily complete before the vowel duration can be ascertained. Such asynchrony may have consequences for perception. For example, Toscano & McMurray (2012) showed that, in the perception of English stop consonant voicing (/b/ versus /p/), the perceptual contribution of VOT and the duration of the vowel that follows the stop consonant are
doi 10.1075/ihll.4.06lla © 2015 John Benjamins Publishing Company
Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz
processed in the order in which they appear in time (VOT then vowel duration) even though both contribute to the same voicing decision. That is, listeners do not wait to hear the vowel duration cue before making a decision based primarily on the earlieroccurring VOT cue. In the present study, we examine another case of temporal asynchrony in cues to voicing: that between the acoustic correlates of VOT and onset f0 (the fundamental frequency at the onset of the vowel following the consonant release; Kingston et al., (2008)). In this case, the cues are once again temporally asynchronous, with the acoustic correlates of VOT (i.e. prevoicing and aspiration) necessarily completed before those of onset f0 begin, after the consonant release. Results of a study of voicing perception in Spanish and English (Llanos et al., 2013) suggest that listeners are not using onset f0 information to judge voicing in tokens in which voicing begins before the burst release (so-called voicing lead or prevoiced tokens). This pattern is consistent with a temporally dependent processing of the two cues in that the presence of voicing before the consonant burst release may be sufficient to cue the perception of voicing – listeners do not need to wait until they hear the later-occurring cue. On the other hand, it has been argued that VOT and onset f0 constitute integrated components of a higher-level perceptual cue. According to Kingston & Diehl (1994) and Kingston et al. (2008), the encoding of stop consonant voicing depends primarily on the presence or absence of low frequency energy in proximity to the burst release of the consonant. In voiced stops, low frequency energy is typically enhanced by the presence of voicing in close proximity to the burst (before and/or soon after) as well as by lowering onset f0. In contrast, in voiceless stops the absence of low frequency energy near the consonant release is enhanced by prolonging VOT and by raising onset f0. Thus, although the acoustic components of VOT and onset f0 are temporally asynchronous, their joint contribution to a single, temporally constrained, spectral property may encourage listeners to postpone voicing decisions until they have information about the contribution of both cues, meaning that they are processed essentially in parallel. In this case, the absence of an effect of onset f0 in tokens with voicing lead must be explained in terms of the differential strength of the two cues as components of the low frequency property. That is, the perceptual effect of onset f0 may be insignificant in voicing lead tokens because it simply contributes too little to the larger spectral property of energy at low frequencies when compared to the much greater amount of energy provided by the presence of voicing immediately preceding the burst release. In this study, we evaluate these competing temporal and spectral accounts of the relative contribution of onset f0 and VOT in the perception of initial stop consonant voicing in Spanish and English by extending the predictions of the two mechanisms
Temporal and spectral dependencies in the processing of Spanish and English
to other regions along the VOT continuum. The goal is to determine which of the two accounts is more predictive of listeners’ behavior in each language. Results will contribute to a better understanding of the mechanisms of speech perception more generally, and, specifically, to a better understanding of the processing of English and Spanish stop consonant voicing.
2. VOT and onset f0 in Spanish and English stop consonant voicing Crosslinguistically, the production of VOT spans three linguistic phonetic categories: voicing lead, in which voicing starts between 40 – 200 ms before the consonant release; short lag, in which the onset of voicing is coincident with or slightly behind the consonant release by approximately 0 – 20 ms; and long lag, in which the onset of voicing occurs between 40 – 150 ms after the consonant release (Cho & Ladefoged, 1999; Holt, Lotto & Diehl, 2004; Keating, 1984). Spanish and English both contrast voicing in stop consonants, but the VOT categories of Spanish are voicing lead and short lag while the two main VOT categories of initial stops in English are short lag and long lag (Lisker & Abramson, 1964). In addition, English stop productions typically include a small but significant percentage of voicing lead tokens (Shultz, Francis, & Llanos, 2012). As a consequence, the main difference between Spanish and English voicing is not entirely based on production but rather is defined by the perceptual VOT boundary used to encode phonological distinctions: while English relies on a 20 ms VOT boundary to distinguish voiced stops (VOT 〈 20 ms) from voiceless stops (VOT 〉 20 ms), Spanish relies on a 0 ms boundary for the same phonological distinction (Dmitrieva et al., 2015; Llanos et al., 2013). In both Spanish and English, onset f0 correlates with VOT and serves as a secondary cue to stop consonant voicing. In Spanish, significantly lower onset f0 values correlate with voicing lead VOT in voiced stops, while higher onset f0 values correlate with short lag VOT in voiceless stops. In English, both short lag VOT and voicing lead VOT are realizations of the voiced category and are characterized by lower onset f0 values in comparison to long lag VOT of voiceless stops, characterized by higher onset f0 values (Dmitrieva et al., 2015; Hombert, 1978; Löfqvist et al., 1989; House & Fairbanks, 1953; Petersen, 1983) (Figure 1). Previous research shows that both English and Spanish listeners rely on VOT as well as onset f0 when making voicing decisions (Idemaru & Holt, 2011; Whalen et al., 1993; Lisker, 1978). However, the degree to which onset f0 contributes to the voicing categorization is not independent of the VOT (Llanos, et al., 2013). Specifically, onset f0 appears to play no role in voicing decisions when combined with lead voicing VOT.
Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz
500
4000
400
3000
300
2000
200
1000
100
0
0
100
300 200 Time (ms)
400
m (Hz)
Frequency (Hz)
English Beat 5000
0 500
500
4000
400
3000
300
2000
200
1000
100
0
0
100
300 200 Time (ms)
400
m (Hz)
Frequency (Hz)
Spanish visa 5000
0 500
500
4000
400
3000
300
2000
200
1000
100
0
0
100
300 200 Time (ms)
400
m (Hz)
Frequency (Hz)
English pete 5000
0 500
500
4000
400
3000
300
2000
200
1000
100
0
0
100
300 200 Time (ms)
400
m (Hz)
Frequency (Hz)
Spanish pisa 5000
0 500
Figure 1. Spectrograms showing VOT and onset f0 in the Spanish and English production of voiced and voiceless stops. From top to bottom: English word beat (short lag: [p]eat); English word pete (long lag: [ph]ete); Spanish word visa (lead voicing: [b]isa); Spanish word pisa (short lag: [p]isa). VOT intervals are indicated by white rectangles. Onset f0 values correspond to the onsets of pitch contour lines. Higher onset f0 values, characteristic of voiceless stops, result in the falling pitch contour at the beginning of pete and pisa
Temporal and spectral dependencies in the processing of Spanish and English
For Spanish, this means an overall diminished role of onset f0 in voicing categorization, since lead voicing VOT encompasses a whole voiced stops category. Furthermore, as production data show (Dmitrieva et al., 2015), the lack of onset f0 effect in Spanish voiced stops is not caused by a lower degree of correlation between VOT and onset f0 across Spanish voicing categories as compared to English; this correlation is equally strong in both languages.
3. Perceptual interaction of VOT and onset f0 in stop consonant voicing The findings reported by Llanos, et al. (2013) and Dmitrieva et al. (2015) can be interpreted within the auditory model of cue integration proposed by Kingston et al. (2008). According to this model, VOT and onset f0 are perceptually integral because both cues combine to enhance the perception of the same auditory property, namely the presence (or absence) of low frequency energy in the vicinity of the consonant release. In voiced stops, the presence of such a property is enhanced by anticipating the onset of voicing (e.g. by shortening VOT) and lowering f0 (introducing more energy into the critical spectro-temporal region), while in voiceless stops, the absence of the same property is enhanced by lengthening VOT and raising f0 (pulling low frequency energy away from the critical spectro-temporal region). Thus, the perceptual integration of VOT and onset f0 in the auditory model results directly from the fact that the two cues affect roughly the same spectro-temporal property of the speech signal. Therefore, this model predicts that perceptual contributions of the two cues are mutually dependent, such that onset f0’s contribution to the perception of stop consonant voicing is inversely proportional to that of VOT. That is, onset f0 is expected to serve as a strong cue to voicing only in circumstances in which VOT is a relatively weak cue, and vice versa. When VOT provides a strong contribution to the perception of the presence (or absence) of low frequency energy in proximity to the burst, the small perturbations in low frequency energy introduced by onset f0 differences will have little effect. However, when VOT’s contribution to the perception of the presence (or absence) of low frequency energy in proximity to the burst is smaller, onset f0-related differences can have an appreciative effect on the perception of the low frequency property, and therefore on the phonetic voicing decision (Kingston et al., 2008; Llanos et al., 2013). As a consequence, the auditory model accounts for the lack of onset f0 effect in the processing of voicing lead VOT by arguing that the spectral contribution of onset f0 in these tokens is made perceptually irrelevant by the large amount of energy that is already provided by VOT. Given that all Spanish voiced stops are produced with voicing lead VOT, this model explains why there is no onset f0 reliance in the encoding of Spanish voiced stops.
Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz
4. VOT and onset f0: Spectral or temporal dependency? Although the perceptual interaction of voicing cues in voicing lead VOT can be accounted for by arguing for a spectral dependency between VOT and onset f0, an alternative temporal-based explanation cannot be ruled out. In voicing lead tokens, onset f0 becomes available to listeners after a period of 40 to 200 ms of prevoicing (the period of time from the onset of closure voicing to the onset of vowel voicing). Since voicing lead VOT is unequivocally mapped onto voiced stops, such period of time could be informative enough to bias listeners’ decisions towards the voiced category before onset f0 is completely processed. For this reason, the lower perceptual role played by onset f0 in the processing of voicing lead VOT could be motivated by the temporal dependency between onset f0 and the acoustic components of VOT. This temporal-based analysis can also be extended to the processing of long lag VOT, where onset f0 is provided after a long period of aspiration. Since no voicing is available during the aspiration portion of the long lag stops, the prolonged absence of the low frequency energy may bias the listeners towards the ‘voiceless’ decision before onset f0 information is processed. When extended to the whole VOT continuum, the spectral and temporal accounts make different predictions concerning the relative strength of onset f0 as a cue to voicing at different points of the VOT range. Specifically, they predict different VOT locations for the peak contribution of the onset f0 factor. A temporal account, which assumes that VOT and onset f0 are processed independently, in cascade, at two consecutive time-points, predicts that the peak perceptual contribution of onset f0 will be located around 0 ms of VOT, where the voicing decision cannot be biased by a long period of prevoicing or aspiration. The spectral account, which assumes that VOT and onset f0 cues are processed in parallel, due to the enhancement of the same spectrotemporal property, predicts that the peak perceptual contribution of onset f0 will be located at a VOT value significantly higher than 0 ms, where the spectral contribution of onset f0 is not overpowered by the contribution of VOT. Figure 2 shows a schematic representation of the two predictions.
5. Methods In order to estimate the location of the peak perceptual contribution of onset f0 in Spanish and English, we analyzed a database of voicing decisions made by Spanish and English native listeners (Section 5.1). Following this analysis, peak locations were compared across languages to determine whether the mechanism of perceptual interaction of VOT and onset f0 is language-specific (different peak locations across languages) or language-independent (similar peak locations across languages). Finally,
Temporal and spectral dependencies in the processing of Spanish and English
contribution
Predictions
–60
temporal-based spectral-based
–40
–20
0
20
40
60
vot (ms)
Figure 2. Predictions of the temporal-based model and the spectral-based model with respect to the contribution of onset f0 across the VOT continuum
peak locations were analyzed to determine whether the mechanism of perceptual interaction of VOT and onset f0 is of a temporal nature (around 0 ms VOT) or a spectral nature (mid positive VOT values).
5.1 The database The perceptual database was collected by Llanos et al. (2013). It consisted of perceptual responses by 40 participants to a two-alternative forced-choice task (2AFC). Twenty participants were native speakers of Spanish (9 men, 11 women; mean age 28.3 years) tested in Madrid, Spain (Consejo Superior de Investigaciones Científicas – Centro de Ciencias Humanas y Sociales; CSIC-CCHS) and 20 participants were native speakers of American English (12 women, 8 men; mean age 21 years) tested in Lafayette, Indiana (Purdue University). English participants reported an average of 3.4 years of experience with a language other than English (Spanish, French, Japanese, German and Chinese). Spanish participants were speakers of Latin-American dialects (5 participants) or Central-North varieties of Castilian Spanish (15 participants). All Spanish participants reported having studied English and an average of 2.2 years residence in an English-speaking country. None of the Spanish participants reported having lived in an English-dominant environment during the year prior to the experiment. Participants reported no history of speech or hearing impairment.
Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz
The stimuli used in the 2AFC task consisted of a set of 52 synthetic syllables ranging from [ba] to [pha] orthogonally in terms of VOT (from –60 ms to 60 ms in 13 steps of 10 ms) and onset f0 (from 90 Hz to 150 Hz in 4 steps of 20 Hz). Syllables were synthesized using the Klatt Speech synthesizer (Klatt, 1980) as implemented in Praat 5.2 (Boersma & Weenink, 2009) and played to listeners in 11 randomized blocks at a comfortable listening level via a MATLAB 7.1 GUI (MathWorks, 2010). After listening to each token, listeners indicated their response by clicking one of the two on-screen buttons labeled “PA” and “BA”. Despite the fact that stimuli were syllables instead of words, all participants in the database appeared to perform in their respective Spanish or English mode of perception. This was confirmed by the significantly different perceptual VOT boundaries established for English (average VOT boundary = 22.4 ms) and Spanish participants (average VOT boundary = 2.8 ms) (Llanos et al., 2013). A previous analysis of this database (Llanos et al., 2013) also indicated the number of perceptual ‘users’ of VOT and onset f0: participants who relied significantly on both cues when making voicing decisions. Although all participants were VOT users, the number of those who used onset f0 as well was smaller: 10 in English and 9 in Spanish. Only the participants who used both VOT and onset f0 in their voicing decisions were included in the current analysis.
5.2 The model The model of perceptual data analysis was designed to quantify the degree of perceptual contribution of onset f0 at different VOT values. For each onset f0 user, the acoustic space was subdivided into thirteen VOT windows of 30 ms, each one centered at a different VOT value. For the two windows at the edges, centered at –60 and 60 ms of VOT, respectively, the window length was 20 ms. For each VOT window, two different logistic models were fitted to the proportion of PA responses, one with a single predictor and another one with two predictors. The single-predictor model collapsed PA responses over onset f0 for each VOT level in each window, and thus reflected voicing decisions as if they were based exclusively on VOT. In contrast, the two-predictor model reflected voicing decisions based on both VOT and onset f0. For each model, the goodness of fit was calculated in terms of its logistic deviance from the real proportion of PA responses, such that lower deviance indicates a better fit (Hosmer, Lemeshow, & Sturdivant, 2013). The degree of perceptual contribution of onset f0 for each VOT value was calculated by subtracting the deviance of the two-predictor model from the deviance of the single-predictor model in the corresponding VOT window. Since a greater contribution of onset f0 to the goodness of fit of the two-predictor model will lower the models’ deviance compared to that of the one-predictor model, larger values of the deviance difference will reflect a greater independent contribution of onset f0 to
Temporal and spectral dependencies in the processing of Spanish and English a. Acoustic space
Onset f0 (Hz)
160 140 120 100 80
–60
–40
–20
0 20 VOT (ms)
40
60
PA proportion
b. Proportion of PA responses
1 0.5 0 150 130
110
Onset f0 (Hz)
90
10
20 VOT
30 (ms)
Prob (pa)
c. Two predictor model deviance = 117.0492
0.8 0.6 0.4 0.2 140
120
Onset f0 (Hz)
100
10
15
25 20 VOT (ms)
30
d. One predictor model deviance = 146.6792 0.6
Prob (pa)
0.5 0.4 0.3 0.2 0.1 10
12
14
16
18 20 22 VOT (ms)
24
26
28
30
Figure 3. Example of model fitting and calculation of deviance (goodness of fit). (a) window of analysis centered at 20 ms VOT; (b) proportion of PA responses to tokens within the window of analysis by a single English-speaking subject; (c) decision surface from a two-predictor logistic model fit to these proportions, showing deviance as a measure of goodness of fit (117); (d) decision curve from a one-predictor logistic model fit to the same proportions collapsed over onset f0 values, showing greater deviance (146)
Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz
the voicing decisions. Figure 3 shows an example of this procedure applied to a VOT window centered at 20 ms for one English-speaking participant. In this case, the onset f0 effect can be observed in the measurably lower deviance of the two-predictor model (117, in Figure 3c) versus that of the one-predictor model (146, shown in Figure 3d). For each subject, the peak contribution of onset f0 in the VOT space was identified as the VOT value corresponding to the greatest difference in deviance between the oneand two-predictor models.
5.3 The analysis The first question to be assessed by the analysis was whether the English and Spanish locations of individual onset f0 peaks were significantly different, which would suggest a language-specific mechanism of VOT and onset f0 interaction. To address this question, a two-sample t-test was applied to the individual onset f0 peaks from both groups of participants. The second question to be addressed was whether the peak contribution of onset f0 was located close to 0 ms VOT, as predicted by the temporalbased account, or in a mid-positive VOT area, at a value significantly higher than 0 ms VOT, as predicted by the spectral-based account. The location of the peak contribution of onset f0 relatively to 0 ms VOT was estimated with a one-sample t-test which compared the actual distribution of onset f0 effect to that of the null hypothesis (distributional mean is located at 0 ms VOT) and provided 95% confidence interval for the VOT values significantly removed from 0 VOT.
6. Results Results of the two-sample t-test demonstrated that Spanish and English listeners did not differ significantly in terms of the VOT values associated with the peak contribution of onset f0 (t17 = –1.3219, p = 0.2), with a mean value of 25 ms of VOT (SD = 8.1 ms) for the Spanish group, and 30 ms of VOT (SD = 10.1 ms) for the English group. Since there was no significant difference between Spanish and English participants in terms of the peak location, both English and Spanish individual onset f0 peaks were submitted to the same one-sample t-test to assess whether the joint Spanish-English distribution was located around 0 ms of VOT, as predicted by the temporal-based account, or at a mid-positive VOT value, as predicted by the spectral-based account. Results showed that the Spanish-English distributional mean of onset f0 peak contribution was significantly different from 0 ms VOT (t18 = 3.44, p = 0.002) with a 95% confidence interval spanning a mid-positive VOT area from 22.8693 ms to 31.8675 ms. The distribution of onset f0 effect across the VOT continuum in the Spanish and English group can be seen in Figure 4, which plots the degree of onset f0 contribution obtained by applying the model described in Section 5.2 to all PA-responses collapsed across participants within each language (Spanish/English).
Temporal and spectral dependencies in the processing of Spanish and English Contribution of onset f0
180 160
Deviance subtraction
140 120 100
Spanish English
80 60 40 20 0 –60
–40
–20
0 vot (ms)
20
40
60
Figure 4. Actual onset f0 contribution across the VOT continuum: Spanish listeners (dotted line) and English listeners (solid line)
7. Discussion The first goal of the analysis was to determine whether the mechanism of perceptual interaction of VOT and onset f0 in English and Spanish was language-specific. The two-sample t-test revealed that the peak contribution of onset f0 was located at a similar VOT value for both Spanish and English listeners. In Spanish, the VOT range with the greatest onset f0 effect was around 25 ms of VOT, only a few milliseconds below the VOT boundary used by English listeners to distinguish between voiced and voiceless stops, and further away from the prototypical production of Spanish voiceless stops (14 ms VOT on average according to Dmitrieva et al., under review). The fact that Spanish listeners nevertheless rely on onset f0 in this area suggests that the perceptual use of onset f0 in Spanish is not significantly affected by the language-specific properties of input, which provides Spanish listeners with a significantly lower degree of experience with long lag VOTs as compared to English listeners (Dmitrieva et al., 2015). Similarly, the lack of onset f0 effect in the English processing of voicing lead stops could be attributed to the English participants’ lower degree of experience with the voicing lead VOT category. However, Spanish listeners are even less familiar with long lag stops than English speakers are with voicing lead stops. Nevertheless, Spanish participants still relied on onset f0 for the processing of long lag stops. This suggests that
Fernando Llanos, Olga Dmitrieva, Alexander L. Francis & Amanda Shultz
the lack of onset f0 effect in the English processing of voicing lead is more likely to be produced by the mechanism of perceptual interaction between VOT and onset f0 that operates in the rest of the VOT domain, and not by experience-based factors. In English, the VOT range with the greatest onset f0 effect was around 30 ms of VOT. It could be argued that English listeners rely on onset f0 more heavily at 30 ms VOT to compensate for the voicing ambiguity of those VOT values that lie between the two main VOT categories in English. Since such tokens cannot be easily classified as voiced or voiceless based on the VOT alone, onset f0 becomes more informative in guiding voicing decisions. However, this explanation would predict no onset f0 effect in the processing of the same VOT value (30 ms) in Spanish, in which such value is unambiguously voiceless, and this prediction is not supported by the results of the analysis (which show a high onset f0 effect in that VOT area). Thus, the results suggest that the mechanism of perceptual interaction of VOT and onset f0 is not language-specific. The second goal of the analysis was to determine whether the perceptual interaction of VOT and onset f0 was based on a temporal or a spectral dependency between the two cues. According to the results of the one-sample t-test, English and Spanish listeners exhibited a higher onset f0 effect in the area in which the spectral contribution of onset f0 is perceptually more salient (mid-positive VOT values), as predicted by the spectral-based account. This supports a mechanism of perceptual interaction in which the contribution of onset f0 significantly depends on the spectral contribution of VOT to the presence of low frequency energy nearby the consonant release. On the other hand, the lack of onset f0 effect in the shortest VOT range, where the temporal hypothesis predicted the peak onset f0 effect, suggests that the temporal component in the VOT and onset f0 interaction is not the determining factor in these cues’ joint perceptual use in voicing categorization.
References Boersma, P., & Weenink, D. (2009). Praat: doing phonetics by computer (Version 5.1. 05)[Computer program]. Retrieved May 1, 2009. Cho, T., & Ladefoged, P. (1999). Variation and universals in VOT: Evidence from 18 languages. Journal of Phonetics, 27(2), 207–229. DOI: 10.1006/jpho.1999.0094 Dmitrieva, O., Llanos, F., Shultz, A., & Francis, A. (2015). Phonological status, not voice onset time, determines the acoustic realization of onset f0 as a secondary voicing cue in Spanish and English. Journal of Phonetics, 49, 77–95. House, A.S., & Fairbanks, G. (1953). The influence of consonant environment upon the secondary acoustical characteristics of vowels. Journal of the Acoustical Society of America, 25, 105. DOI: 10.1121/1.1906982
Temporal and spectral dependencies in the processing of Spanish and English
Holt, L.L., Lotto, A., & Diehl, R. (2004). Auditory discontinuities interact with categorization: Implications for speech perception. Journal of the Acoustical Society of America, 116, 1763–1773. DOI: 10.1121/1.1778838 Hombert, J.M. (1978). Consonant types, vowel quality, and tone. In V. Fromkin (Ed.), Tone: A linguistic survey (pp. 77–111). New York NY: Academic Press. Hosmer Jr, D.W., Lemeshow, S., & Sturdivant, R.X. (2013). Applied logistic regression. Hoboken, NJ: Wiley.com. DOI: 10.1002/9781118548387 Idemaru, K., & Holt, L.L. (2011). Word recognition reflects dimension-based statistical learning. Journal of Experimental Psychology: Human Perception and Performance, 37(6), 1939. DOI: 10.1037/a0025641 Keating, P. (1984). Phonetic and phonological representations of stop consonant voicing. Language, 60, 286–319. DOI: 10.2307/413642 Kingston, J., Diehl, R.L., Kirk, C.J., & Castleman, W.A. (2008). On the internal perceptual structure of distinctive features: The [voice] contrast. Journal of phonetics, 36(1), 28–54. DOI: 10.1016/j.wocn.2007.02.001 Kingston, J., & Diehl, R.L. (1994). Phonetic knowledge. Language, 70, 419–454. DOI: 10.1353/lan.1994.0023 Klatt, D. H. (1980). Software for a cascade/parallel formant synthesizer. The Journal of the Acoustical Society of America, 67(3), 971–995. Lisker, L. (1978). In qualified defense of VOT. Language and speech, 21(4), 375–383. Lisker, L., & Abramson, A.S. (1964). A crosslinguistic study of voicing in initial stops: Acoustical measurements. Word, 20, 384–422. Llanos, F., Dmitrieva, O., Shultz, A., & Francis, A.L. (2013). Auditory enhancement and second language experience in Spanish and English weighting of secondary voicing cues. Journal of the Acoustical Society of America, 134(3), 2213–2224. DOI: 10.1121/1.4817845 Löfqvist, A., Baer, T., McGarr, N.S., & Story, S.R. (1989). The cricothyroid muscle in voicing control. Journal of the Acoustical Society of America, 85, 1314–1321. DOI: 10.1121/1.397462 McMurray, B., & Jongman, A. (2011). What information is necessary for speech categorization? Harnessing variability in the speech signal by integrating cues computed relative to expectations. Psychological review, 118(2), 219. DOI: 10.1037/a0022325 Petersen, N.R. (1983). The effect of consonant type on fundamental frequency and larynx height in Danish. Annual Report of the Institute of Phonetics of the University of Copenhagen, 17, 55–86. Repp, B.H. (1982). Phonetic trading relations and context effects: New experimental evidence for a speech mode of perception. Psychological Bulletin, 92, 81–110. DOI: 10.1037//0033–2909.92.1.81 Shultz, A.A., Francis, A.L., & Llanos, F. (2012). Differential cue weighting in perception and production of consonant voicing. Journal of the Acoustical Society of America, 132(2), EL95– EL101. DOI: 10.1121/1.4736711 Toscano, J.C., & McMurray, B. (2012). Cue-integration and context effects in speech: Evidence against speaking-rate normalization. Attention, Perception, & Psychophysics, 74(6), 1284–1301. DOI: 10.3758/s13414-012-0306-z Whalen, D.H., Abramson, A.S., Lisker, L., & Mody, M. (1993). F0 gives voicing information even with unambiguous voice onset times. Journal of the Acoustical Society of America, 93, 2152–2159. DOI: 10.1121/1.406678
Segmental and prosodic conditionings on gradient voicing assimilation in Spanish Rebeka Campos-Astorkiza The Ohio State University
This study experimentally analyzes sibilant voicing assimilation in Spanish, i.e. the assimilation in voicing of /s/ before a voiced consonant. We argue that it is only by exploring the effects of both prosodic and segmental material, and their interplay, that the nature of this assimilation can be understood. We show that the variable degree of sibilant voicing is dependent upon the location of stress, the manner of articulation of the triggering consonant, and the type of prosodic boundaries. Based on our results, this gradient assimilatory process is modeled as the result of blending of the glottal gestures for the adjacent voiceless and voiced sounds, and for this reason, the degree of voicing is subject to influence by segmental and supra-segmental information. Keywords: voicing; assimilation; gestural blending; Spanish
1. Description of the phenomenon The alveolar fricative /s/ undergoes voicing assimilation before a voiced consonant, both within words and across word boundaries, in those varieties of Spanish that retain this sound in pre-consonantal position, as illustrated in (1). Impressionistic descriptions of this phenomenon describe it as gradient, variable and stylistically determined (Navarro Tomás, 1977; Hualde, 2005), an observation that leads to two important issues. First, the factors that influence this gradient voicing should be explored in an effort to find out how gradience works and how it is manifested in the language. Second, a model capable of capturing this gradience in a principled manner should be developed. These are the two main goals of this paper, which goes beyond the basic observation that voicing assimilation is gradient, and offers evidence of the segmental and prosodic factors that affect the process and proposes a formal system that integrates these factors as determiners in the resulting amount of voicing assimilation. (1) [izla] ‘island’ [dezðe] versus [peste] ‘from; plague’ [laz#ɣamas] versus [las#kamas] ‘the ranges; the beds’
doi 10.1075/ihll.4.07cam © 2015 John Benjamins Publishing Company
Rebeka Campos-Astorkiza
Despite being a canonical phenomenon of the Spanish phonological system, only recently have a handful of instrumental studies allowed us to start to compose a more detailed, data-based picture of /s/ voicing assimilation. Using semi-directed speech, Schmidt & Willis (2011) present acoustic data from Mexican Spanish regarding the degree of /s/ voicing assimilation in different environments, including before a voiced consonant within the same word. The authors find an absence of voicing in 37% of the contexts where /s/ occurred before a voiced consonant and conclude that this assimilatory process is not categorical for Mexican Spanish. Strycharczuk (2012) uses acoustic analysis to explore /s/ voicing in Quito Spanish in intervocalic position, and before a sonorant, a voiced and a voiceless stop in read speech. She concludes that the assimilation is gradient for some speakers and categorical, though optional, for others. She bases her conclusions on the fact that speech rate influences the percent of voicing for some of her participants but not for others. Beyond this effect, Strycharczuk does not elaborate on any other possible determining factors. Garcia (2013) studies /s/ voicing in intervocalic and pre-consonantal positions in read and spontaneous speech in Highland Colombian Spanish. Focusing on her results for /s/ voicing before a voiced consonant, the author finds that the degree of voicing is highly variable and that it is influenced by speech rate but not by speech formality. Garcia’s main interest lies in the speech rate effects and she does not discuss any other possible factors influencing the degree of voicing. Furthermore, she groups together data from speakers of different dialects for her analysis, which is problematic since voicing assimilation may display dialectal differences. Finally, Bárkányi (2013) collects acoustic data from the northwestern Spanish city of Oviedo through a reading task, and explores the amount of /s/ voicing before a voiced consonant based on the duration of voicing in milliseconds during the fricative. Bárkányi concludes that voicing is optional or categorical by speaker. As for the determining factors, beyond speaker differences, the author finds less duration of voicing before sonorants than before obstruents. Even though the new findings presented by Bárkányi are welcome, there are some methodological problems that limit the impact of the results. More precisely, the use of the raw voicing duration is highly problematic since some kind of normalization is necessary to control for differences in rate (stemming from different speech rates or speakers). Most studies looking at voicing assimilation use the percent of voicing in relation to the overall duration of the fricative. Overall, these recent studies offer much welcome data in relation to pre- consonantal voicing of /s/, emphasizing the gradient nature of the phenomenon. However, there have been very limited attempts at understanding what factors condition the assimilation, beyond speaker differences and speech rate effects. The current study argues that it is only by exploring the effects of both prosodic and segmental material, and their interplay, that the nature of this assimilation can be fully understood. We show that the variable degree of sibilant voicing is dependent upon a series
Segmental and prosodic conditionings on gradient voicing assimilation in Spanish
of factors including the location of stress, the manner of articulation of the following/ triggering consonant, and the types of boundary. Furthermore, these results help us develop a model of this assimilatory process based on the blending of the laryngeal gestures for the adjacent voiceless and voiced sounds, which accounts for the main factors involved in the phenomenon.
2. V oicing assimilation as gestural blending: Predictions and supporting findings Previous studies have analyzed gradient assimilation processes in different languages as the result of increased gestural overlap, i.e. temporal co-occurrence of different articulatory gestures. According to these studies, gradient assimilatory effects derive from changes in gestural magnitude and timing which lead to greater overlap among adjacent gestures (Browman & Goldstein, 1989). Gestural blending is the result of increased overlap between gestures specified for the same articulator (Browman & Goldstein, 1989). Under this approach, voicing assimilation stems from the blending of two temporally overlapping laryngeal gestures (Munhall & Löfqvist, 1992). In fact, Romero (1999) presents articulatory data from Castilian Spanish /s/ voicing assimilation that supports the gestural blending account. The author reports results for the timing between the laryngeal gesture and the oral gesture for /s/. He finds that the laryngeal gesture peak occurs between the oral gestures for /s/ and a following voiced stop, rather than being synchronized with the oral gesture for /s/ as is the case in single consonants. Romero argues that these findings indicate that there is mutual influence between the two consonants, which results in gestural blending in the laryngeal configuration. Thus, taking Romero’s results into account and following previous analyses of assimilation, we model /s/ voicing assimilation in Spanish as the result of gestural blending due to an intense overlap between the conflicting glottal gestures for /s/ and a following voiced consonant, within the theory of Articulatory Phonology (Browman & Goldstein, 1989). This model offers a novel framework within which to analyze /s/ assimilation in Spanish and allows us to make predictions about factors that have been shown to affect gestural magnitude and organization as possible conditionings on the degree of voicing. We explore the role of stress and prosodic boundaries, which influence the duration and magnitude of gestures, and the manner of articulation of the voiced consonant triggering the assimilation, which determines the gestural specification of the relevant sound. Previous articulatory studies have found that gestural magnitude is increased in stressed positions (Pierrehumbert & Talkin, 1992; Beckman, Edwards, & Fletcher, 1992). In relation to voicing, Cooper (1991) found that stress influences the magnitude of laryngeal gestures in both time and space for English, with stressed positions
Rebeka Campos-Astorkiza
leading to larger, longer gestures. The increase in gestural magnitude of stressed elements affects the degree of overlap with adjacent (unstressed) elements resulting in more overlap with them – larger lengthened gestures temporally coincide for longer with surrounding gestures. Accordingly, we would expect more assimilation, i.e. more overlap, to be triggered by stressed than unstressed consonants. Applying this to Spanish /s/ assimilation, we predict that more voicing assimilation will take place when the triggering consonant is in a stressed syllable than when it is in an unstressed one (e.g. /rasgó/ versus /rásgo/). On the other hand, gestural magnitude is reduced in unstressed positions, with the highest amount of reduction occurring in those unstressed positions following a stressed syllable, i.e. post-stress positions. Several studies have reported more reduction in post-stress positions than in pre-stress ones (Arvaniti, 1991; de Jong, 1998). For Spanish, Cole et al. (1999) found more weakening of stops (spirantization) in post-stress than in pre-stress positions, lending further support to the observation that gestural magnitude is highly reduced in post-stress locations. Lesser magnitude, i.e. shorter gestures, would result in greater degree of overlap by adjacent gestures, leading to more assimilation. Consequently, we predict that the highest degree of voicing will be observed when /s/ occurs in post-stress position. As for the influence of prosodic boundaries on gestural organization, several studies have reported less temporal overlap among gestures separated by, or adjacent to, a prosodic boundary, (e.g. Byrd & Choi, 2010). Furthermore, prosodic boundaries of different strengths display differences in the magnitude of their effects, with degree of overlap decreasing as we move into higher prosodic boundaries (Byrd & Salzman, 1998; Parrell, Lee & Byrd, 2010). Earlier studies claimed that major prosodic boundaries could prevent processes from taking place in a categorical manner (Nespor & Vogel, 1986). However, a number of more recent studies have shown that in fact boundary effects are gradient. Their presence reduces or limits the magnitude of the process but does not completely block it in a categorical manner (e.g. Zsiga, 1995; Byrd & Saltzman, 1998). With these findings in mind, we expect the amount of /s/ voicing assimilation to be affected by the presence of prosodic boundaries, more precisely, a decrease in voicing is predicted as we move to higher prosodic boundaries. Additionally, the presence of a major phrase boundary such as an intonational phrase boundary is predicted to not block assimilation completely but rather limit the amount of voicing to its lowest. Finally, the manner of articulation of a consonant affects its gestural specification, which in turn determines the duration and magnitude of the consonantal gestures. Consequently, we would expect different manners of articulation to show different behavior with respect to gestural overlap and amount of assimilation (Recasens & Mira, 2012). Some of these predictions were initially explored in Campos-Astorkiza (2014), where the percent of voicing in Northern Peninsular Spanish is analyzed. As expected,
Segmental and prosodic conditionings on gradient voicing assimilation in Spanish
the results show that voicing assimilation is gradient. With respect to the role of stress, this factor was considered as a simple dichotomy between stressed or unstressed /s/, and it showed no effect on the percent of voicing. Similarly, there was no effect of word boundaries, and the same amount of voicing could be observed within and across words. However, the presence of an intonational phrase boundary had a significant effect on the degree of voicing limiting the assimilation, though not categorically. Finally, the manner of articulation of the following consonant proved to be a significant factor in the degree of assimilation: A following open or more lenited obstruent triggers more assimilation than a close or less lenited one. In this paper, we report an experiment that elaborates on the results of CamposAstorkiza (2014) and aims at advancing our understanding of the linguistic factors that condition the voicing assimilation of /s/. First, we further explore the role of stress in Northern Peninsular Spanish by including a more refined classification of stress patterns, including pre-stress versus post-stress positions and stressed versus unstressed, to test the effect of post-tonic positions. Second, we examine the effect of other manners of articulation of the triggering consonant, expanding to obstruents versus sonorants. Section 3 describes in detail the experimental methodology, and Section 4 reports the main results. Section 5 concludes with some discussion of the findings and the implications for the model and future research.
3. Experimental methodology 3.1 Experimental materials Three sets of stimuli were created according to three conditions. First, the voiceless versus voiced condition, which was used to establish the threshold of the unvoiced category (see Section 3.2), included words with /s/ preceding a voiceless or voiced stop (2a). This set included 8 different lexical items, 4 with a voiced obstruent and 4 with a voiceless one (see Appendix for the complete list of stimuli). The second set corresponded to the triggering consonant manner, which was used to explore the role of different types of manner including obstruent, nasal, and liquid, within and across words (2b). For the voiced obstruents and nasals, several places of articulation were included in the stimuli, but for the lateral just one place was included. In total, 12 different items were part of this set, half for within words and the other half for across words. (2) Sample experimental materials a. Voiced versus voiceless condition /sésge/ vs /péske/ ‘bias (3sg.sbjv); fish (3sg.sbjv)’ /músgo/ vs /búsko/ ‘moss; search (1sg.prs)’
Rebeka Campos-Astorkiza
b. Triggering Consonant Manner condition /mísmo/ vs /ísla/ vs /sésge/ ‘same; island; bias (3sg.subj)’ /los#mótes/ vs /los#lótes/ vs /los#bótes/ ‘nicknames; lots; cans’ The third and final set relates to the stress pattern condition, which included four stress patterns, with either a sonorant or an obstruent as the following consonant. In order to get four stress patterns, the relevant sequences, i.e. /s/ + voiced consonant, were placed across words, allowing us to modify the stress location with respect to the target /s/ and the triggering consonant. The word containing /s/ could be stressed on the syllable containing /s/ or the preceding syllable, and the following word could be stressed on the first (which contains the triggering consonant) or second syllable. This gave us the four patterns shown in Table 1. The labels used to refer to these patterns are ‘post-stress’, ‘pre-stress’, ‘stressed only’ and ‘pre-stress, stressed’. Note that the ‘post-stress’ pattern (e.g. /tápas duRítas/) is where /s/ is the furthest away from the stress, and this would be the weakest position in articulatory terms, as described in Section 2. The ‘pre-stress’ pattern (e.g. /tápas dúRas/) also includes /s/ in a post-stress location, however we expect this position to be less weak than the pattern labeled ‘post-stress’ because of the stress on the triggering consonant. This condition tests the role of the four stress patterns for obstruents and also sonorants, and it included 16 items. Table 1. Sample stimuli for the stress pattern condition Triggering obstruents
Triggering sonorants
‘post-stress’
/tápas duRítas/
/tápas lokáles/
‘pre-stress’
/tápas dúRas/
/tápas lókas/
‘stressed only’
/papás duRítos/
/papás lokwélos/
‘pre-stress, stressed’
/papás dúRos/
/papás lókos/
Unstressed /s/
Stressed /s/
All target words were inserted in similar positions within carrier sentences of comparable structure and length. These 36 sentences can be found in the Appendix. 8 native speakers of Northern Peninsular Spanish (6 females and 2 males)1 were recorded reading the stimuli sentences. Each sentence was repeated 5 times in random order, giving us 180 tokens per speaker, and all repetitions were analyzed. The participants
. Gender was not one of the factors considered in this study but based on the overall pattern there seem to be no differences between female and male participants.
Segmental and prosodic conditionings on gradient voicing assimilation in Spanish
were recorded using a head-mount Plantronics-DSP microphone (with an internal USB sound card) and a laptop computer as they read the sentences from the computer screen in a quiet room. Note that Northern Peninsular Spanish is characterized by a lack of /s/ weakening so that this fricative does not undergo aspiration or deletion in this variety (Hualde, 2005). All the speakers form a relatively homogenous group. They are all between 25–35 years of age, live in the same area of Northern Spain (Bilbao region), have some education past high school, either at the university level or some professional degree, and have already entered the market place, i.e. they are not students anymore. All the participants were monolingual speakers of Spanish, although they all had some L2 knowledge of Basque and English. The scope of this project is limited to the potential linguistic factors influencing voicing assimilation, however, this process could display variation based on social factors and for this reason, these factors were controlled for as much as possible.
3.2 Data analysis Three acoustic cues were measured in order to quantify the amount of voicing assimilation. Vowel duration was measured from the beginning of the vocalic formant structure to the beginning of the frication noise as seen in the spectrogram, paying special attention to changes in intensity of F2. The fricative duration was taken from when the frication noise started in the spectrogram until it receded. Finally, the duration of any glottal pulses during the fricative was taken by observing the presence of a voicing bar in the spectrogram and periodicity in the waveform. This method was chosen over others because it has been shown to be the most reliable way to measure fricative voicing in Spanish (Gradoville, 2011). The spectrogram and waveform in Figure 1 illustrate these three measurements.
Vowel
Voicing Fricative During Fricative
Figure 1. Waveform and spectrogram illustrating the three acoustic measurements for the word /mismo/
Rebeka Campos-Astorkiza
Using these acoustic measurements, four dependent variables were calculated, namely, vowel duration, fricative duration, percentage of voicing during frication and voicing category. Due to space limitations, only the results for percent voicing and voicing category are presented in the results section. The voicing duration was used to calculate the percentage of voicing during the fricative. This measurement (% voicing) allows for some normalization across different speakers and speech rates and for this reason, it is preferable to use it instead of the raw voicing duration (Recasens & Mira, 2012). Finally, the voicing category for each /s/ was coded as unvoiced, partially voiced or fully voiced based on the percentage of voicing, following Smith (1997). The unvoiced category includes those tokens that have less than 25% of voicing during /s/. This threshold is based on the distribution of the percentage of voicing for the voiceless sequences, i.e. tokens with /s/ followed by a voiceless consonant (see Cuartero, 2001). Figure 2 includes the distribution of the voiceless sequences and shows that most of the tokens fall below the 25% threshold. As for the two other voicing categories, partially voiced tokens are those that have between 25% and 90% of voicing. Fully voiced tokens are those with over 90% of voicing. 35
% of voicing
30 25 20 15 10 5 0
0
50 100 Number of tokens
150
Figure 2. Distribution of the percentage of voicing for voiceless sequences
4. Results In order to obtain a complete picture of the phenomenon, all four variables should be analyzed, including vowel and fricative duration. However, due to space limitations, only the results for those two directly related to voicing during /s/ are presented; the other two are left for future discussion.
4.1 Following consonant manner The two dependent variables reported here, i.e. percent voicing and voicing category, were subject to a series of statistical analyses to evaluate the effect of the following
Segmental and prosodic conditionings on gradient voicing assimilation in Spanish
consonant manner, word boundaries and stress patterns. To test for the effect of the following consonant manner and of word boundaries (data sets illustrated in (3a, b)), the percent voicing was first subjected to a one-factor ANOVA with following consonant manner (voiceless, voiced, nasal, lateral) as the independent variable and then to a two-factor ANOVA with following consonant manner (voiced, nasal, lateral) and word boundary (within, across) as independent variables. There are no tokens with a following voiceless consonant across words and for this reason, the two tests were necessary. Speaker was included as a random effect for both tests in order to obtain more robust effects and to control for possible interspeaker variation. Post-hoc pairwise comparisons with a Bonferroni adjustment were also performed for the different manners in order to evaluate whether they were different from each other. For the voicing category variable, crosstabs and Fisher’s exact tests were calculated for the following consonant manner and word boundary. The significance level for the statistical analyses was set at 0.05. The results from the one-factor ANOVA indicate that the manner of articulation of the following consonant has a significant effect on the percentage of voicing (F(3,766 = 25.508, p < .001). Table 2 includes the means for the percent voicing according to the four possible manners, i.e. voiceless stop, voiced stop, nasal or lateral. Table 2. Means and standard deviations for % voicing according to the manner of the following consonant (in ms.) Voiced obstruent
Nasal
Lateral
Voiceless obstruent
mean
71.2
55
57.6
13.7
SD
1.55
1.68
2.38
2.05
The results from the pairwise comparisons for the manners further show that voiceless stops are different from the three other manners, similarly to voiced stops, which also behave differently from the three other types. However, nasals and laterals are not different from each other, indicating that they lead to comparable amounts of voicing during the sibilant. Thus, the ranking in (3) illustrates the relative assimilatory strength of each manner, from the manner that triggers the highest amount of percent voicing, i.e. voiced obstruents, to that which results in the lowest degree of voicing, i.e. voiceless obstruent.
(3) voiced obstruent > nasal = lateral > voiceless obstruent
The results from the two-factor ANOVA show that, as expected, the following consonant manner has a significant effect on percent voicing (F(2,590) = 5.586, p = .016), but the word boundary has no effect on this variable. Furthermore, there is no interaction between word boundary and manner, indicating that all manners behave
Rebeka Campos-Astorkiza
s imilarly with respect to word boundary, more precisely, the percent voicing across and within words is not significantly different for any of the manners. Table 3 summarizes the mean percent voicing triggered by each manner, i.e. voiced obstruent, nasal and lateral, within and across words. These results for word boundary replicate those obtained in the experiment reported in Campos-Astorkiza (2014). Table 3. Mean and standard deviations for % voicing for each manner across and within words Voiced obstruent
Nasal
Lateral
mean
SD
mean
SD
mean
SD
within words
66.27
2.21
55.82
2.21
54.26
3.14
across words
77.82
2.55
53.56
3.12
64.61
4.41
The crosstabs and Fisher’s exact test results indicate that the manner of the following consonant has an effect on the voicing category (two-sided Fisher’s exact test, p I, III
¡¡ Te/Os me/nos √ Te lo/a/s (M; E) ¡¡ Os lo/a/s ¡¡ Te/Os le/s
I > III III
√ Me lo/a/s (M; E) ¡¡ Me le/s ¡¡ Nos lo/a/s ¡¡ Nos le/s
Initials indicate which child produced which type.
The sum is more than its parts
In terms of Perlmutter’s template, the children produced only a subset of allowed clitic clusters. They produced clusters se > 1P and se > 3P, 2P > 3P and 1P > 3P but there were no clusters that combined 2P > 1P nor se > 2P. Their production of acceptable clusters is consistent with the following sequencing knowledge: se > 2P/1P > 3P. As noted in Section 4.3, both children also produced a few cases of sequences that are disallowed by the template. Except for one case of *lo se (18c), the sequences produced that are disallowed by the template (five tokens of me se) are actually attested variants in many Spanish dialects or sociolects (see Heap, 2005).
4.4.3 Featural content of the clusters María produced 117 licit clitic clusters. Out of that total, 41% consisted of 1Ps-3P combinations, 35% were 2Ps-3P combinations and 24% consisted of se followed by 1P (10%) or by 3P (14%). Among the non-se 3P clitics, there were both singular forms (3s: 67 lo, 16 la and 6 le) and plural forms (3p: 8 los and 7 las). The number of tokens for each cluster type appears in Table 6. Table 6. Error-free clitic combinations: María CL1
1s
1s Total
CL 2
3p 3s
2s
2s Total
3p
3s
3-se 1s 3p
3-se Total
Grand Total
3s
Age 3
3
1
1
2
1
1
6
11
16
2
2
4
1
1
21
2;03
10
10
2
2
1
4
5
17
2;04
3
3
1
4
5
2
2
4
12
5
6
2
1 6
7
15
2;06
3
3
3
3
6
2;08
2
2
22
22
3
3
27
2;09
2
2
3
3
1
2
3
8
1
1
1
1
5
35
41
12 1 15
28
117
2;01 2;02
2;05
5
1
2;11
2
1
3
Grand Total
8
40
48
6
2
The same counts for Emilio show that he produced a total of 87 licit two-clitic clusters. The majority (77%) consisted of 1Ps-3P combinations, followed by 20% of 2Ps-3P clusters and only 2 cases (2.2%) of se-3P combinations. There were no plural clitics for any person or gender. The number of tokens for each cluster type appear in Table 7, including gender information (f = feminine; m = masculine) for 3P clitics.
María Cristina Cuervo & Ana Teresa Pérez-Leroux
Table 7. Error-free clitic combinations: Emilio CL1
1s
1s Total
CL 2
3sf 3sm
2s
2s Total
3sf 3sm
3p Total Grand Total 3sf 3sm
Age 2;02
1
2
2
2
2;03
1
2
2
2;04
1
1
1
3
3
3
2
2
2
3
4
4
17
17
5
5
2;09
19
33
11
11
2;11
2
2
3;00
1
1
2
2
51
67
18
18
2;05 2;06
1
2;07 2;08
Grand Total
14
16
22 1
1
2
46 2 3
1
1
2
87
Data for both children show a clear tendency of clusters to combine a participant clitic (singular 1P or 2P) with a 3P (mostly singular) accusative clitic. Only María produced a good number of INCH/DAT clusters involving se followed by a 1P or 3P dative clitic. There are no clusters which contain either nos or os (a participant plural clitic). A search of target contexts in our data indicates the gap is in the contexts, not in the forms (i.e. there are no cases in our coding of os omission in required contexts). In the general corpus of single clitic structures, we found only three cases of nos in María (3/1178) and six in Emilio (6/691); there were no cases of os in either corpus. These findings are consistent with the general absence or scarcity from early productions noted by Reglero & Ticio (2003). None of the children produced errors related to featural content, that is, there were no clusters violating the Person-Case Constraint or sequences of two identical clitics.
4.4.4 Opaque forms As mentioned in Section 2.2, the only case in Spanish of a problematic combination of clitics that is solved by the emergence of an opaque form is spurious se. When and how do children acquire this combination? There are cases of spurious se in the production of both children: María produced 7 examples while Emilio produced only 2. In one instance, María omits spurious se
The sum is more than its parts
(Las quito ‘I take them’ for Se las quito ‘I take them from her’) but in the produced 7 examples and the 2 by Emilio there are no errors in the production of spurious se. In terms of emergence and productivity, María’s first spurious se takes place at 2;04 and becomes productive at 2;05. Inchoative se appears at 2;01 and becomes productive at 2;03, while aspectual se appears in a cluster at 2;03 and becomes productive by age 2;05 (20). Emilio’s two cases of spurious se appear at 2;09 while inchoative se appears (with order errors) at 2;06 and the one case of aspectual se is produced (with an order error) at 2;08 (21). (20) María Que se lo cuente yo. ‘That I tell it to him’ Spurious se (2;05) Se me escapa el bebé. ‘The baby is escaping from me’ Inchoative se (2;03) El lobo se la comía. ‘The wolf ate her up’ Aspectual se (2;05) (21) Emilio Tú se la das a mamá. ‘You gave it to mommy’ Me se ha caído una. ‘I have dropped one’ Sí, lo se llevado. ‘Yes, she has taken it with her’
Spurious se (2;09) Inchoative se (2;06) Aspectual se (2;08)
Therefore, although the production of spurious se is not very frequent, there does not seem to be a delay in its emergence or productivity with respect to genuine (aspectual or inchoative) se. Nor did we observe any instance where children produce illicit but transparent clusters such as *le las or *les lo.
5. Discussion Our study has revealed that after a short-lived no-cluster stage, children produce clusters: –– –– –– –– ––
at rates that show adult-levels of productivity, with extremely low error rates, with no illicit feature combinations, with appropriate repair strategies (spurious se), and with little evidence of systematic gaps in the paradigm, as their uses approximate adult patterns.
Our results show that the children are not conservative in the sense of Tomasello (2003), but that they are conservative in the sense of Snyder (2007). That is, they favour errors of omission rather than commission. Omission errors occur at rates comparable to those of single clitics. The absence of errors in the positioning of clusters in obligatory and optional contexts (clitic climbing) indicates that before the age of 3 children have fully acquired the syntactic and morphological operations responsible
María Cristina Cuervo & Ana Teresa Pérez-Leroux
for multiple cliticization. In terms of internal ordering, the one consistent reordering pattern (me se) takes place within the range of dialectal variation in the language. While the possibility that the children might have been exposed to this use cannot be ruled out in principle, postulating this as the source of the children’s errors would go against both the logic of language acquisition and the accumulated evidence on the role of input. We believe the divergent patterns of the child learner and the range of micro-parametric variation both express the set of options that can be generated within the organization of the system. It is improbable that children produce an atypical expression because it exists in some other dialect not present in their community, though it is impossible to rule out incidental exposure. The reverse scenario is more plausible: namely, that the range of dialectal micro-variation exists in part because of the kinds of things children do. The alternative me se dialects exist because it is the kind of pattern children’s grammar can generate. The most frequent cluster pattern does not correspond to what we could predict from the single clitic frequencies (since se is quite common), nor does it necessarily correspond to the cluster input, since for instance, for María’s mother, se + ACC is the prevalent form. Another place where input and child use diverge is in the production of dative/accusatives. While these clusters are the most frequent type for both children and adults, it is not the case that the other two constructions (inchoative/dative and aspectual/accusative) are late to appear. It is striking that the percentage of successful realization is much higher with the low-frequency constructions than with the more common ones. In principle, this is not compatible with usage-base approaches, but subsequent analyses should relativize this observation to verb types. The status of the first clitic is a relevant feature: inchoatives and aspectual constructions have an “advantage” in that the first clitic is obligatory; in other words, the inchoative and the aspectual are clitic constructions in which the reflexive cannot be alternatively expressed by a lexical DP, as datives and accusatives can. Our data provides no evidence of a general avoidance for clusters. Although we found that children did not produce clusters with forms that combine a participant (marked person) and a plural feature (marked number), i.e. nos or os, these forms were never a target in the cluster contexts. There is one potential gap in this first system for clusters: the combination of 1 and 2 person clitics. Although one can never be sure, we can speculate that children at this stage still do not know the proper sequencing of first and second person clitics, that is, clitics expressing a participant feature. It is important to note that ordering of participant clitics varies crosslinguistically (e.g. Spanish dictates 2P > 1P, while Romanian dictates 1P > 2P) and whether a language has one or the other ordering seems to be a non-predictable property of clusters. In fact, the combination of 1P and 2P clitics can be disallowed altogether, a restriction known as the strong version of the Person Case Constraint (Bonet, 1991). An emerging generalization is that children tend not to produce clusters containing two marked features
The sum is more than its parts
(participant, plural, feminine, dative). For the moment, however, we cannot differentiate between this strong interpretation and a hypothesis based on input frequency. Children have remarkable distributional learning abilities that they can employ to detect abstract patterns in language (Marcus et al., 1999). In the case of clitic clusters, this distribution is a complex contingency dependent on abstract featural information, form and context. Children navigate this complexity with seeming ease, revealing a performance that is conservative but productive at the same time.
References Alba de la Fuente, A. (2012). Cltic combinations in Spanish: Syntax, processing and acquisition. Unpublished Doctoral Dissertation. University of Ottawa. Bonet, E. (1991). Morphology after syntax: Pronominal clitics in Romance. (Doctoral dissertation, MIT). Distributed by MITWPL. Bonet, E. (1995). Feature structure of Romance clitics. Natural Language and Linguistic Theory, 13, 607–647. DOI: 10.1007/bf00992853 Castilla, A.P., & Pérez-Leroux, A.T. (2010). Omissions and substitutions in Spanish object clitics: Developmental optionality as a property of the representational system. Language Acquisition, 17(1/2), 2–25. DOI: 10.1080/10489221003620904 Colantoni, L., &. Cuervo, M.C. (2013). Clíticos acentuados. In L. Colantoni & C. Rodríguez Louro (Eds.), Perspectivas teóricas y experimentales sobre el español de la Argentina (pp. 143–157). Madrid: Iberoamericana Vervuert. Connors, K., & Nuckle, L. (1996). Morphosyntax of French personal pronouns and the learning/acquisition dichotomy. In O.A. Jaeggli & C. Silva Corvalán (Eds.), Studies in Romance linguistics (pp. 225–282). Dordrecht: Foris. Cuervo, M.C. (2013). Spanish clitic clusters: three of a perfect pair. Borealis: An International Journal of Hispanic Linguistics, 2(2), 191- 220. DOI: 10.7557/1.2.2.2692 Cuervo, M.C. (2014). Alternating unaccusatives and the distribution of roots. Lingua, 141, 48–70. DOI: 10.1016/j.lingua.2013.12.001 Davis, M. (1995). Analyzing syntactic variation with computer‐based corpora: The case of Modern Spanish clitic climbing. Hispania, 78(2), 370‐380. DOI: 10.2307/345438 Dominguez, L. 2003. Interpreting reference in the early acquisition of Spanish clitics. In S. Montrul & F. Ordóñez (Eds.), Linguistic theory and language development in Hispanic languages (pp. 212–228). Somerville, MA: Cascadilla Press. Eisenchlas, S. (2002). What do Argentinian children know about clitics that linguists don’t? Unpublished Doctoral Dissertation. University of Queensland. Ezeizabarrena, M.J. (1997). Morfemas de concordancia con el sujeto y el objeto en el lenguaje infantil. In A. T. Pérez-Leroux & W. R. Glass (Eds.), Contemporary perspectives on the acquisition of Spanish, volume 1: developing grammars. Somerville, MA: Cascadilla Press Francetto, B., & Santos, M. (2012). The ontology of roots and the emergence of nouns and verbs in Kuikuro language and children’s speech. Ms. UFRJ. Fujino, H., & Sano, T. (2000). Some notes on the null object phenomenon in child Spanish. In S. C. Howell, Fish, S.A., & T. Keith-Lucas (Eds.), Proceedings of the Boston University Conference on Language Development 24 (pp. 308–318). Somerville, MA: Cascadilla Press.
María Cristina Cuervo & Ana Teresa Pérez-Leroux Gonzalez, G. (1970). The acquisition of Spanish grammar by native Spanish speakers. Unpublished Doctoral Dissertation. University of Texas at Austin. Grimshaw, J. (1997). The best clitic: Constraint interaction in morphosyntax. In L. Haegeman (ed.), Elements of grammar: A handbook in contemporary syntactic theory, (pp. 169–196). Dordrecht: Kluwer. DOI: 10.1007/978-94-011-5420-8_4 Grinstead, J., Lintz, P., Vega-Mendoza, M., De la Mora, J., Cantú-Sánchez, M., & Flores-Avalos, B. (2014). Evidence of optional infinitive verbs in the spontaneous speech of Spanishspeaking children with SLI. Lingua, 140, 52–66. DOI: 10.1016/j.lingua.2013.11.004 Harris, J. (1995). The morphology of Spanish clitics. In H. Campos (Ed.), Evolution and revolution in linguistic theory (pp. 168–197). Washington, DC: Georgetown University Press. Heap, D. (2005). Constraining optimality: Clitic sequences and feature geometry. In L. Heggie & F. Ordóñez (Eds.), Perspectives on clitic and agreement affix combinations. Amsterdam: John Benjamins. DOI: 10.1075/la.74.05hea Hernández-Pina, F. (1984). Teorías psico-sociolingüísticas y su aplicación a la adquisición del español como lengua materna. Madrid: Siglo XXI. Larrañaga, P., & Guijarro-Fuentes, P. (2012). Pronouns and clitics in early acquisition. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110238815 López Ornat, S. (1994). La adquisición de la lengua Española. Madrid: Siglo XXI. DOI: 10.1177/014272379601604706 MacWhinney, B. (2000). The CHILDES project: Tools for analyzing talk (Third Edition). Mahwah, NJ: Lawrence Erlbaum Associates. DOI: 10.1177/014272370002006006 Marcus, G.F., Vijayan, S., Bandi Rao, S., & Vishton, P.M. (1999). Rule-learning in seven-monthold infants. Science, 283, 77–80. DOI: 10.1126/science.283.5398.77 Monachesi, P. (1995). A grammar of Italian clitics. Unpublished Doctoral Dissertation. Tillburg University. Ordóñez, F. (2012). Clitics in Spanish. In J.I. Hualde, A. Olarrea, & E. O’Rourke (Eds.), The handbook of Hispanic liinguistics (pp. 423–451). Oxford: Blackwell. DOI: 10.1002/9781118228098. ch21 Pérez-Leroux, A.T., Cuza, A., & Thomas, D. (2011). Clitic placement in Spanish/English bilingual children. Bilingualism: Language and Cognition, 14(1), 221–232. DOI: 10.1017/s1366728910000234 Pérez-Leroux, A. T., & Kahnemuyipour, A. (2014). News, greatly exaggerated: Commentary on Ambridge, Pine & Lieven. Language, 90(3), e115–e125. Perlmutter, D.M. (1971). Deep and surface structure constraints in syntax. New York NY: Holt, Rinehart and Winston. Pescarini, D. (2010). Elsewhere in Romance: Evidence from clitic clusters. Linguistic Inquiry, 41, 427–444. DOI: 10.1162/ling_a_00003 Pescarini, D. (2011). Mapping Romance clitic sequences. Quaderni di lavoro ASIt, 12, 1–29. Reglero, L., & Ticio, M.E. (2003). The acquisition of clitics in child Spanish. In S. Montrul & F. Ordoñez (Eds.), Linguistic theory and language development in Hispanic languages (pp. 297–316). Somerville: Cascadilla Press. Richards, N. (2010). Uttering trees. Cambridge, MA: The MIT Press. DOI: 10.1017/s0022226711000223 Rivero, M.L. (2004). Spanish quirky subjects, person restrictions and the person-case constraint. Linguistic Inquiry, 35(3), 494–502. DOI: 10.1162/ling.2004.35.3.494
The sum is more than its parts
Rivero, M.L. (2008). Oblique subjects and person restrictions in Spanish: A morphological approach. In R. D’Alessandro, S. Fischer, & G. Hrafnbjargarson (Eds.), Agreement restrictions (pp. 215–250). Berlin: De Gruyter. DOI: 10.1515/9783110207835.215 Roeper, T., & Pérez-Leroux, A.T. (2011). Simplicity and complexity in child language and its explanation. Infancia y Aprendizaje, 34(3), 363–380. DOI: 10.1174/021037011797238612 Serra, M., Serrat, E., Solé, R., Bel, A. & Aparici, M. (2000). La adquisición del lenguaje. Barcelona: Editorial Ariel. Slobin, D.I. (1985). Why study acquisition crosslinguistically? In Slobin, D.I. (Ed.), The crosslinguistic study of language acquisition (pp. 4–24). Hillsdale, NJ: Lawrence Erlbaum Associates. DOI: 10.1017/s0272263100013528 Snyder, W. (2007). Child language: The parametric approach. Oxford: Oxford University Press. DOI: 10.1017/s0022226708005276 Tomasello, M. (2003). Constructing a language: A usage-based theory of language acquisition. Cambridge, MA: Harvard University Press. DOI: 10.1017/s0012162205220698 Vila, I. (1990). Adquisición y desarrollo del lenguaje. Barcelona: Graó. Yang, C. (2002). Knowledge and learning in natural language. Oxford: Oxford University Press. DOI: 10.1017/s1351324905213724
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?* José Camacho & Alena Kirova Rutgers University
We analyze the effect of overt agreement marking on nouns in the acquisition of the syntactic constraints of bare nominals among L1 Russian learners of L2 Spanish. Gender is grammatically active in both Russian and Spanish, but nominals have a different structure and distribution. We posit that gender serves as a path for Russian learners to acquire the distribution and the structure of nominals in Spanish. The results of the present study show that L1 Russian speakers do acquire the overall DP structure in Spanish, and that overt agreement has some effect on the acquisition of bare NP constraints. Results are consistent with the representation of feminine as different from masculine. Keywords: gender; gender concord; bare NPs; Russian; Spanish; L2 acquisition
1. Introduction Studies generally agree that the acquisition of L2 gender agreement presents difficulties, though different studies report non-convergent results, in part due to varying methodologies (Bruhn de Garavito & White, 2002; Hawkins & Franceschina, 2004; White, Valenzuela, Kozlowska–Macgregor & Leung, 2004; Franceschina, 2005; Cornips & Hulk, 2008; Montrul, Foote, & Perpiñán, 2008; Sagarra & Herschensohn, 2011; Grüter, Lew–Williams & Fernald, 2011; among others). Bruhn de Garavito & White (2002), for example, report that adult L2 Spanish speakers eventually acquire gender marking on definite determiners, but have more difficulty with indefinite determiners. Hawkins & Franceschina (2004), on the other hand, note that late bilinguals of an L1 without gender are less accurate than those of a gendered L1 in producing and perceiving g ender agreement. Processing studies also offer no consensus * Thanks to two anonymous reviewers and to the editors of the volume for invaluable comments.
doi 10.1075/ihll.4.09cam © 2015 John Benjamins Publishing Company
José Camacho & Alena Kirova
on whether L2 adult learners show sensitivity to gender agreement as manifested in determiner–noun discord (see Sagarra & Herschensohn, 2013 for a summary). Some studies report different results for production and comprehension in offline and online experiments, showing that L2 Spanish speakers perform at ceiling in offline comprehension, but show errors in elicited production (Grüter, Lew–Williams, & Fernald, 2012). Most research on the acquisition of gender agreement focuses on L1 speakers of a non-gendered language learning a gendered L2 (L1 English–L2 Spanish, for example). A different question arises when the L1 and the L2 both have explicitly marked gender, but show different distributional properties. Does the presence of overt gender facilitate the acquisition of the distribution properties of nominal structures in the L2? The availability of gender in both languages suggests that the L2 learner will have a representation for the feature in the L2. However, two further issues arise: first, whether the representation of the feature will be the same in each grammar (or even across noun classes in each language), and second, whether having a feature representation will help acquire other aspects of the nominal domain that may not be similar across both grammars. Regarding the first question, we compare two different perspectives on gender morphology: the traditional view that considers morphemes like -a in Spanish to be the morphological counterpart of the feature ‘feminine’, and Harris’ (1991) view that -a is a word class marker, only indirectly connected to gender. In terms of the second question, both Russian and Spanish have overt gender marking on nouns and adjectives, but they differ with respect to the overt structure of the DP. Descriptively, Russian generally lacks overt determiners, whereas Spanish generally has them. More interestingly, Russian bare NPs can appear in any word order, whereas Spanish bare NPs are restricted to postverbal positions. If we assume that Russian and Spanish share abstract gender features and gender concord as a basic grammatical property but that the structure of bare NPs is qualitatively different in the two languages (in ways we will specify below), we ask whether gender marking (lexical assignment of gender features to nouns) and concord (the process of matching gender features across the DP) can aid the learner in figuring out the underlying structure of bare nominals in Spanish and the constraints associated with the distribution of bare NPs. Specifically, overt evidence of agreement between nominals and adjectives in Russian and between nominals, adjectives and overt determiners in Spanish could help the learner to hypothesize the existence of null determiners in Spanish, and to discover the word order constraints on bare nominals that follow from having a null determiner. Looking at the role of overt gender morphology in the context of L2 acquisition provides us with a window into the internal structure of DPs and the representation of abstract features such as gender.
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
2. On the L2 acquisition of DP structure Two areas targeted by studies of L2 acquisition of the DP include the status of overt determiners when the L1 lacks them, and the acquisition of gender in an L2 with explicit gender when the L1 lacks this feature. Regarding the first area, L2 learners whose L1 lacks determiners must acquire several properties of the DP. On one hand, they must learn the distinction between mass and count nouns, and its correlate with determiner choice. On the other hand, they must hypothesize which determiner is associated with a certain meaning; thus, for example, they must figure out that ‘a’ is indefinite but ‘the’ is definite. Additionally, they must learn under which conditions a given determiner can be interpreted as generic or existential (see Snape, 2006; Montrul & Ionin, 2010, 2012; Cuza, Guijarro, Pires, & Rothman, 2013; Perez–Cortes, 2014; among others). For example, Snape (2006) argues that Japanese L1 speakers who learn English can distinguish between count and mass nouns, but have difficulties with definite ‘the’ in mass plurals and with indefinite ‘a’ in count singulars. Other studies have noted that speakers of a determinerless L1 fluctuate between a specific and definite interpretation for determiners in the L2, and this fluctuation has been argued to be due to lack of access to Universal Grammar (UG) (Ionin, Ko, & Wexler, 2004) or due to pragmatic deficits (cf. Deprez, Sleeman, & Guella, 2011). Montrul & Ionin (2012) found a low percentage of monolingual–target judgments among L2 learners and heritage speakers of S panish for generic NPs, a pattern that they ascribe to L1 transfer. Perez–Cortes (2014) found the same patterns for heritage speakers in Spanish, but their patterns in English were not consistent with those of English monolinguals, suggesting a more nuanced explanation based on a number of factors, including representational strategies as a product of bilingualism and differential mapping between syntax and interpretation across English and Spanish. As we have seen, the L2 acquisition of grammatical gender presents challenges, particularly for speakers of an L1 without gender. Specifically, as we have indicated, Bruhn de Garavito & White (2002) note that whereas low proficiency L2 learners of Spanish do fairly well in producing gender on definite determiners, they are much less successful with gender in indefinite determiners. Advanced proficiency learners, however, show mastery of the formal gender feature, with any remaining errors assumed to be due to production–specific performance problems. Results from online, time–constrained tasks also show diverse results. On one hand, Sagarra & Herschensohn (2012) find sensitivity to gender violations even in low–proficiency learners (consistent with Bruhn de Garavito & White’s (2002) results), but others find significant differences between advanced L2 learners and monolinguals (Guillelmon, 2001; Gillon-Dowens, Vergara, Barber, & Carreiras, 2010; among others).
José Camacho & Alena Kirova
Grüter, Lew–Williams & Fernald (2011) combined online and offline comprehension and production tasks to see whether the different results from previous studies were due primarily to a production difficulty or a lexical access limitation. Their results indicate performance at ceiling in offline comprehension but significant differences between L1 and L2 speakers in elicited production, primarily with regards to gender assignment and not gender agreement (the grammatical process of sharing features across categories). From this they conclude that the syntactic aspects of gender, as manifested in the grammatical operation of agreement, are acquired (but not necessarily the lexical assignment of gender to categories). Previous research has not specifically focused on the intersection of gender and determiner structure: the structural representation of gender may have clear implications for the acquisition of determiner structure, depending on the analysis of how gender is represented. Thus, the general process of L2 acquisition can shed light on the representation of gender in DPs, both within a language and across languages.
3. The structure of nominals and agreement Since Abney’s (1987) and Szabolci’s (1983, 1987) seminal work, the assumed structure for nominals includes a functional layer with a determiner D head and a DP projection, as in (1)a). Longobardi (1994) argued that D is required for nominals to be interpreted as arguments, hence it is universally required. It follows that determinerless languages must also have D, presumably deleted at Phonological Form (PF), as in (1)b). (1) a. D0
DP
DP b. NP N
Ø
NP N
Others (Fukui, 1988; Chierchia, 1998; Baker, 2003; Bošković, 2008; Bošković & Gajewski, 2011; a. o.) have proposed that determinerless languages lack a functional D projection, as in (2), and that semantic interpretation is a parametric matter, following Chierchia (1998). (2) NP N
Bošković (2008) and Bošković & Gajewski (2011), for example, argue that Serbo– Croatian lacks a determiner layer, noting several systematic differences between
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
Serbo–Croatian and English that they connect to the (un)availability of a D–layer. For example, Serbo–Croatian allows for extractions from NPs, but English does not (see (3)a versus (3)b), a fact that Bošković attributes to the presence of a D–layer in English. This layer is a strong phase which blocks movement from inside. Serbo–Croatian lacks the functional DP–layer, which constitutes a strong phase, hence extraction from within the NP is possible. (3) a. *From which cityi did Ivan meet [NP girls ti] b. Iz kojg gradai je Ivan sreo [DP djevojke ti] from what city is Ivan meet girl ‘Ivan met girls from which city?’ Although Russian does not usually have determiners, it does have overt demonstratives, possessives and quantifiers, which can be argued to be determiners. However, we will argue that they are always NPs, suggesting that possessives and quantifiers are adjectival (but cf. Caruso, 2011 for an argument that they are Ds).
3.1 Nominal structure in Spanish Spanish has overt determiners, including definite and indefinite articles, quantifiers and demonstratives as illustrated in (4). (4) El/un/algún/este amigo. the(m)/a(m)/some(m)/this(m) friend(m) ‘The/a/some/this friend.’ As for bare NPs, they can appear under three restricted conditions (Contreras, 1986; Casielles, 1997; Benedicto, 1998; a. o.). First, they can only be postverbal (see (5)a versus (5)b). Second, they must be plural (see (5)b versus (5)c). Finally, they are interpreted existentially, as in (5)b-(6)a, and not generically, as in (5)b–(6)b.1 (5) a. *Huracanes aparecen en el Caribe. ‘Hurricanes appear in the Caribbean.’ b. En esta época aparecen huracanes en el Caribe. in this season, appear hurricanes in the Caribbean ‘During this season hurricanes appear in the Caribbean.’ c. *En esta época apareció huracán en el Caribe. in this season appeared hurricane in the Caribbean
. Bare modified NPs are possible preverbally: huracanes así aparecen en el Caribe (Laca, 1999), and in headlines, which are well–known to have different properties in this area (see for instance Weir, 2009).
José Camacho & Alena Kirova
(6) a. During this season, there are hurricanes in the Caribbean. b. #Hurricanes have the property of appearing in the Caribbean during this season. We assume a uniform DP structure for Spanish nominals, as in (7), though we are not committed to the notion that only DPs can be arguments. The null determiner in (7) must be licensed. Intuitively, the verb does so under c–command, hence the word order restriction. Specifically, in Benedicto’s (1998) proposal, the null D is a variable existentially bound by the V+T complex that rises to a position above T. (7) a.
DP
el
DP
b. NP
huracán
ø
NP
huracanes
3.2 Nominal structure in Russian The basic Russian noun phrase consists of the noun, and all other elements are optional, regardless of the type of noun (see (8)). In this sense, demonstratives and possessives only appear in marked contexts. (8) a. kniga book
(count noun)
b. voda water
(mass noun)
c. vremya (abstract noun) time The distribution of nominals in Russian raises the question of whether they are DPs with a null D (as in (1)), or NPs (as in (2)), and as mentioned earlier, the evidence is mixed. However, even if Russian nominals have a null determiner, it is clear that the distribution of bare NPs in this language is not subject to the constraints shown for Spanish, in particular with respect to word order and interpretation. Russian nominals can appear in either SV or VS word order, as illustrated in (9).2
. Nevertheless, “[w]ord order is not free, but rather determined by such discourse functions as topic and focus” (Kallestinova (2007: 2). However, in cases where a clause is an answer to a broad focus question, word order is flexible, and the type of nominal is not constrained (see (i)–(iii)), unlike in Spanish.
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
(9) a. Uchenik reshil zadachku. (SVO) pupil.nom solved problem.acc ‘A/the pupil solved a problem.’ b. Uchenik zadachku reshil. (SOV) pupil.nom problem.acc solved c. Zadachku reshil uchenik. (OVS) problem.acc solved pupil.nom d. Zadachku uchenik reshil. (OSV) problem.acc pupil.nom solved e. Reshil uchenik zadachku. (VSO) solved pupil.nom problem.acc f. Reshil zadachku uchenik. (VOS) solved problem.acc pupil.nom
3.3 Distribution of gender in Russian and Spanish Both Spanish and Russian nominals are marked for gender (and number) and agree with other nominal categories. In Spanish, nominals are classified either as masculine or feminine, which typically correspond to -o and -a morphemes respectively, as illustrated in (10). Not all -o/–a animate Ns are masculine/feminine respectively (cf. el fantasma ‘the(m) ghost’), nor are all masculine/feminine referents mapped onto an -o/–a ending: other endings are also typically associated with masculine or with feminine gender, in the sense that they appear with masculine or feminine determiners or adjectives (see Harris, 1991). (10) a. El perro blanco the(m) dog(m) white(m) ‘The white dog (masc.)’ b. La perra blanca the(f) dog(f) white(f) ‘The white dog (fem.)’ More importantly, nouns ending in -e can be masculine or feminine, as in (11). For Harris, this epenthetic vowel is inserted due to the syllable structure requirements of Spanish.
(i)
V čom delo? Čto proisxodit?
‘What is the matter? What is going on here?’
(ii) Lajut sobaki. bark–3sg dogs–nom.pl
‘(The) dogs bark.’
(iii) Sobaki lajut. dogs–nom.pl bark–3sg
(Kallestinova 2007: 19)
José Camacho & Alena Kirova
(11) Unspecified morphemes a. el puente, el diente the(m) bridge(m) the(m) tooth(m) b. la fuente, la muerte the(f) fountain(f) the(f) death(f) Gender is often determined by the sex of the referent with animate nouns that refer to humans, as seen in (12). Nouns ending in -e and consonants are frequently underdetermined for gender (see (12)), which is overtly indicated by the determiner. Sometimes nouns ending in a consonant are interpreted as masculine, with a corresponding -a noun for feminine, as in (12). Whether speakers prefer la juez ‘the judge(fem)’ or la juez–a ‘the judge (fem)’ depends on geographical and/or individual speaker variation. (12) Nouns with human referents a. el cajero, la cajera the(m) cashier(m) the(f) cashier(f) b. el cantante, la cantante the(m) singer the(f) singer c. el juez, la juez /la jueza the(m) judge the(f) judge/the(f) judge(f) Determiners and adjectives also show gender morphology similar to that of nouns, and agree with the noun (see (10)). Russian gender in nominal phrases patterns very similarly to Spanish. Grammatical gender in Russian has three values: masculine (46%), feminine (41%) and neuter (13%) (Corbett, 1991: 78). Masculine nouns usually end in a consonant (kostyum, ‘suit(m)’; stol, ‘table(m)’), feminine nouns in -a (rubashka, ‘shirt(f)’; komnata, ‘room(f)’) and neuter nouns in -o (zerkalo, ‘mirror(n)’; kreslo, ‘armchair(n)’). Palatalized consonant endings can be M or F (lokot, ‘elbow(m)’; dver, ‘door(f)’). As in Spanish, gender on animate, human nouns corresponds to the referent’s sex. Additionally, gender triggers agreement with several elements in the nominal structure (demonstratives, quantifiers, including numerals, possessives, adjectives – see (13)), as well as other elements in the sentence (participles, past tense verbs). (13) a. Moi noviy kostyum my(m) new(m) suit(m) ‘My new suit’ b. Etot krasniy stol this(m) red(m) table(m) ‘This red table’ c. Moya belaya rubashka my(f) white(f) shirt(f) ‘My white shirt’
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
d. Eta bolshaya komnat(a) this(f) big(f) room(f) ‘This big room’ e. Moyo chistoye zerkalo my(n) clean(n) mirror(n) ‘My clean mirror’ f. Eto udobnoye kreslo this(n) comfortable(n) chair(n) ‘This comfortable chair’ To summarize, the differences between Russian and Spanish nominals are presented in Table 1. Table 1. Comparison between nominals in Russian and Spanish Unrestricted bare NPs
Preverbal NPs
Unrestricted interpretation
Φ–features
Φ–feature agreement
Russian
yes
yes
yes
yes
yes
Spanish
no
no
no
yes
yes
4. The representation of gender: Implications for L2 acquisition The traditional analysis of gender assumes that gender is represented by a morpheme (for example, -o/–a in Spanish). Harris (1991), on the other hand, argues that -o/–a mark word class, whereas gender is either a lexically-marked feature, or a feature assigned by default, and that there are redundancy rules that relate gender and word class. As an example, Harris proposes the representations in (14) (adapted from Harris, 1991: 43) for different words. Gender is lexically specified for some of them (like libra) but not for others. Those that lack lexical specification are assigned gender through a default rule. In the same sense, word class is lexically represented for libre and día and assigned by default for other categories. Finally, the redundancy rule in (15) maps “f ”, the lexical gender feature, to the word marker -a (represented as ]a in (15)). (14) Word representations in Harris (1991) libro (M) libra (F) libre día (M) ‘book’ ‘pound’ ‘free’ ‘day’ Stem /libr/ /libr/ /libr/ /di/ Category N N A N Lex. gender f Class ]ø ]a
José Camacho & Alena Kirova
(15) Feminine Marker Rule (Harris, 1991: 44) f → ]a Importantly for us, -o/a contrast with -e in that the first set of morphemes are word markers (either lexically specified or assigned by redundancy), whereas -e is a default vowel present due to syllabic requirements. Harris’ proposal rests on an asymmetric representation of gender: on one hand, only “f ” is lexically marked, and on the other hand, the rule in (15) only applies explicitly to “f ” and to ]a (the setting for ]o would result from a default/elsewhere rule). Thus, we have three possible alternative analyses for Spanish: (16) a. Traditional analysis: gender maps to nominal categories in Spanish, and importantly for our purposes, to D: -o/–a directly map to D (either as morphemes representing M/F or as morphological correlates of an abstract gender feature), whereas -e does not directly map to D. b. Harris’ analysis: gender maps to D, but -o/–a represent a word marker indirectly connected to gender through a morphological rule like (15). c. Gender only maps to N, all other instances of gender across the DP are due to agreement. Each of these possibilities makes a different prediction for the acquisition of gender and the structure of DP. The traditional analysis in (16)a predicts that only -o/–a should provide clues about the existence of D even for bare NPs in Spanish, since those morphemes (or the abstract features related to them) are present in D, whereas -e is not. Harris’ analysis in (16)b, on the other hand, predicts that only feminine might provide a clue to DP structure, because only the feminine feature is lexically marked on nouns, since masculine is the default. No direct connection exists between gender and -o or -e. Finally the analysis in (16)c predicts no differential effect of -o/–a/–e when L2 learners acquire DP structure. From a different point of view, an L1 Russian speaker will have to acquire at least three different properties of DPs in order to converge on the grammar of monolingual Spanish: (a) the existence of an additional D layer (i.e. moving from the structure in (2) to the one in (1)); (b) the fact that DPs with null determiners have a restricted distribution determined by word order; and (c) that DPs with null determiners are interpreted existentially. Given the different analyses with regards to the representation of gender discussed earlier, the question is whether Russian speakers can make use of overt gender agreement to set the distribution of bare nominals in Spanish. Specifically, the L2 learner will have to acquire the existence of a null determiner (as in (1)). If gender is projected on D and -o/–a represent gender, we should find that nominals with overtly marked gender morphology (–o/–a) provide a more visible clue than those without gender morphology (–e) regarding the existence of null D in Spanish.
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
The hypotheses we will test in this study are presented in (17). (17) H1: Russian L1 speakers are sensitive to the different types of determiners (overt and null) in Spanish. H2: If gender is represented in D and -o/–a endings directly represent gender, Russian L1 speakers should rate preverbal DPs ending in -o/–a higher than preverbal NPs ending in -o/–a, but show no such difference for DPs/NPs ending in -e. H3: If only -a is indirectly related to gender (versus -o/–e), -a should have an effect on nominal structure and distribution (i.e. L2 speakers should show a more target–like pattern for the distribution of bare NPs with -a than with -o/–e).
5. The study Thirty–six native speakers of Russian learning Spanish and 6 native speakers of Spanish participated in the study (20–38 years old). All participants were comparable in terms of education level and socioeconomic status (determined through background questionnaires). The native speakers of Russian were recruited in Moscow, Russia and all completed a modified version of the Diploma de Español como Lengua Extranjera (DELE) to test for proficiency (maximum score of 50). According to the results of the Spanish proficiency test, of the 36 L2 Spanish participants, 9 had a low proficiency level (score range: 20–25), 8 were of intermediate proficiency (scores range: 30–35), and 19 had a high level of proficiency in Spanish (scores above 40).3 The native speakers were recruited among postgraduate students at the Department of Spanish and Portuguese at Rutgers University. All participants were asked to judge the acceptability of sentences preceded by a short description of a context (the first sentence in each of the examples in (18)–(20)). After seeing the short description, they saw the test sentence for 3.5 seconds and were asked to rate it on a scale of –2 to 2. Three variables were manipulated in the sentences: subject word order (VS versus SV, as in (18)), gender marking (–o/a versus -e, as in (19)) and type of nominal (bare NP versus DP, as in (20)). The overall distribution of items is presented in Table 2, 12 of which (NPs in SV order) are ungrammatical in monolingual Spanish. Distractors included items with
. Proficiency did not reveal any statistically significant effects as a covariate in the ANOVAs. In other words, we found a sensitivity to different types of determiners and to the word order constraints even in the group that included low proficiency L2 learners.
José Camacho & Alena Kirova
the same structure as the test items, but either with inanimate plural nouns or with possessives and quantifiers.4 (18) a. En la escuela están aprendiendo sobre primeros auxilios. Ayer vinieron los enfermeros. (VS) ‘In school they are learning about first aid. Yesterday the(m) nurses(m) came.’ b. *En este aeropuerto no hay mucho tráfico. Pero las viajeras llegaron ayer. (SV) ‘In this airport there isn’t much traffic. But the(f) travelers(f) arrived yesterday.’ (19) a. Varios amigos y amigas se fueron de viaje hace dos semanas. Las viajeras llegaron esta mañana. (–o/a ending) ‘Several male and female friends went on a trip two weeks ago. The(f) travelers(f) arrived this morning.’ b. Esta mañana no había ningún tren en circulación. Sin embargo, los visitantes llegaron a la estación. (–e ending) ‘This morning there were no trains circulating. Nevertheless, the(m) visitors arrived at the station.’ (20) a. Esta mañana no había ningún tren en circulación. Sin embargo, llegaron visitantes a la estación. (NP) ‘This morning there were no trains circulating. Nevertheless, visitors arrived at the station.’ b. Esta semana hay una cumbre de distintos países en Cartagena. Ayer vinieron los presidentes a la reunión. (DP) ‘This week there is a summit of different countries in Cartagena. Yesterday the(m) presidents came to the meeting.’ Table 2. Distribution of test items DP
NP
Total
SV
VS
SV*
VS
–o
4
3
4
3
14
–a
4
5
4
4
17
–e
8
8
4
6
26
Distractors
18 Grand total
75
. The distribution of items is uneven due to an improper construction of three of the items in -o.
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
As can be seen in Table 2, there were 75 items in total, with 57 test items and 18 distractors. Out of the 57 test items, 45 items were grammatical and 12 items were ungrammatical (the ones under column SV* in Table 2). All items were presented in at random using PsychoPy software.5 Participants were given instructions and a practice set before starting the experimental task.
5.1 Results Our research questions and hypotheses concerned whether the L1 Spanish and L1 Russian-L2 Spanish speakers would behave differently depending on the following variables: position (whether the nominals appeared in preverbal or postverbal positions in the sentence), gender marking (whether the nominals were marked for gender (-a/-o endings) or unmarked for gender (-e ending)), and nominal type (whether the nominals were full DPs or bare NPs). In order to address these questions we ran a 4-way repeated-measures ANOVA with a 2(Position: VS versus SV) x 2(Gender Marking: -a/o ending versus -e ending) x 2(Nominal Type: bare NP versus DP) x 2(Group: L1 versus L2 group) factorial design. This test revealed a significant main effect for Nominal Type (F(1,40) = 10.68, p = .002), and a significant interaction between Nominal Type and Group (F(1,40) = 10, p = .003). That is to say, the L1 control group and the Russian learners of L2 Spanish differed significantly in their ratings of full DPs and bare NPs. This finding is graphically reflected in Figure 1, which shows average ratings for DPs and NPs. The L2 speakers seem to have no preference for either type (mean rating of .6 for DPs versus .59 for NPs), whereas the L1 group correctly rates DPs much higher than NPs (1.05 versus .32 respectively). On the other hand, no significant main effects for Group (F(1,40) = .18, p = .67), Position (F(1,40) = .75, p = .39) or Gender Marking (F(1,40) = .57, p = .46) were found – a result that seems to indicate that the two groups’ performance was not affected by whether the nominals were marked or unmarked for gender and whether they appeared preverbally or postverbally. However, there was a significant interaction between Nominal Type and Position (F(1,40) = 13.32, p = .002), which motivated us to take a closer look at the data and we made a number of observations as a result. First of all, there was a divergence between the L2 and the L1 groups preverbally. On average, the L2 group rated DPs as .8 and NPs as .6, whereas the L1 showed a much higher rating for DPs (1.3) than for NPs (.1). Postverbally, however, L2 speakers prefer NPs to DPs (.4 versus .6 respectively), whereas the L1 group rated DPs slightly higher than NPs (.8 versus .6 respectively). Note that the ratings in the column marked with
. “PsychoPy is an open–source application to allow the presentation of stimuli and collection of data for a wide range of neuroscience, psychology and psychophysics experiments.” (psychopy.org)
José Camacho & Alena Kirova 2 1.5 1 0.5 L2
0 –0.5
DP
L1
NP
–1 –1.5 –2
Figure 1. Average ratings for Nominal Type: DPs versus NPs (–2 to 2) 2.0 1.5 1.0 0.5 L2
0.0 –0.5
DP-V
*NP-V
V-DP
V-NP
L1
–1.0 –1.5 –2.0
Figure 2. Average ratings (–2 to 2) by Nominal Type and Position
an asterisk (NP-V*) would be ungrammatical in the target language; nevertheless, the average ratings, albeit lower, are not negative, possibly reflecting a bias against negative numbers on the scale. Results broken down by position are presented in Figure 2. Although the other variables did not interact significantly, as we mentioned, it is instructive to compare the mean ratings for these categories, as seen in Figure 3. First, in preverbal position, DPs were rated similarly regardless of gender marking whereas NPs (ungrammatical in monolingual Spanish) were rated higher by L2 speakers than by L1 speakers. Second, L2 speakers rated preverbal NPs marked for gender lower than when they were not marked, suggesting a trend in the direction
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
predicted by H2 (that gender is represented on Ds and directly on -o/–a endings). Postverbally, the L2 group rated gender-marked DPs (-o/-a) slightly lower than unmarked DPs(-e) (.33 versus .45 respectively); for the L1 group, they were almost the same (.95 versus .92 respectively). NPs with -o/-a endings were rated lower than NPs with -e ending by the L2 speakers (.48 versus .7 respectively), but higher by the L1 speakers (.98 versus .66 respectively). Recall that all of the postverbal options are grammatical in Spanish. One could wonder why native speakers rate different types of nominals differently, and whether this reflects a true difference in representation of those nominals, or whether it reflects a task effect. At this point, we don’t have a clear answer for this question. 2 1.5 1 0.5 L2
0 –0.5
DPoa-V DPe-V *NPoa-V *NPe-V V-DPoa V-DPe V-NPoa V-NPe
L1
–1 –1.5 –2
Figure 3. Average ratings (-2 to 2) for nominals by Type, Gender marking and Position
Additionally, we were interested in the alternative analysis of nominals proposed by Harris (1991), who rejects the traditional masculine -o versus feminine -a gender distinction and proposes that these endings mark word classes, not gender. Recall that Harris groups -o/–e endings together versus the -a ending, since only the latter is assigned lexical gender that is mapped to the word class through a rule. In this chapter, then, we looked at two types of gender marking – gender marking (the traditional distinction between marked -o/–a versus unmarked -e nouns) and Harris’ (1991) word class marking (the default -o/–e versus the marked feminine gender -a). We have reported the results on gender marking above, and below we will report the findings with respect to word class marking. In order to address the question of whether the notion of word class played a role in the observed patterns, we ran a 4-way repeated-measures ANOVA with a 2(Word Marking) x 2(Position) x 2(Nominal Type) x 2(Group) factorial design. Importantly
José Camacho & Alena Kirova
for our research hypotheses, Word Marking approached significance (F(1,40) = 3.66, p = .06), and there was a significant interaction between Word marking, Nominal Type and Group (F(1,40) = 12.02, p = .001). As seen in Figure 4, L2 speakers rated DP(–a) lower than DP(–o/–e), whereas L1 speakers had the opposite trend: higher ratings for DP(–a) than for DP(–o/–e). Regarding NPs, which would be ungrammatical in monolingual Spanish, L2 speakers had very similar ratings for NP(–a) and NP(–o/–e), whereas L1 speakers had noticeably lower ratings for NP(–a) than for NP(–o/–e). 1.6 1.4 1.2 1.0 0.8
L2
0.6
L1
0.4 0.2 0.0 –0.2
Dpa
DPo/e
*NPa
*NPo/e
–0.4
Figure 4. Preverbal ratings by type and word marking
From this data, it would seem that L2 learners have noticed that DP(–a) has different properties than DP(–o/–e), but they haven’t acquired the monolingual pattern. For NPs, on the other hand, the pattern for the L2 speakers goes in the same direction as the native speakers, but the difference in rating that the L1 group shows is still missing. In this sense, L2 learners have not yet acquired the target pattern, since the difference in ratings between the two groups is statistically significant.
6. Discussion Overall, the results suggest that L2 speakers are sensitive to different types of determiners (H1), since participants rated bare NPs and DPs significantly differently. Furthermore, speakers were sensitive to the word order constraints on nominal type, confirming that L2 learners have begun to acquire the structure and distribution of nominals in Spanish.
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
In the context of the nominal structures discussed for Russian and Spanish in (1)–(2) above, (21) provides three possible acquisition scenarios. 21(a) assumes partial transfer from Russian: bare nominals are NPs (as in Russian), but overt determiners are DPs (as in Spanish). 21(b) is the representation of monolingual Spanish, and 21(c) represents full transfer from Russian: all nominals are analyzed as NPs, with the overt determiner treated as a specifier. (21) a. Partial b. Convergent transfer (NP) representation DP el
NP NP huracán
huracán
DP los
c. Full L1 transfer NP
DP NP
Ø
huracanes
NP NP
huracanes
los
NP NP
huracanes
huracanes
Our results are clearly incompatible with the scenario in 21(c), as this option does not predict asymmetric ratings based on nominal type or on word order. (21)a-b are consistent with our results: since both representations have distinct structures for bare NPs and overt DPs, speakers may posit different distributions for each type. However, if we assume Benedicto’s (1998) analysis that null Ds are variables (see (7)), we can account for the L2 speakers’ preference for NPs over DPs in postverbal position: in postverbal position, V+T always c–commands the null D, whereas preverbally it does not. The relatively high ratings for preverbal NPs could be seen either as a sign of competing grammars (one with NP, one with null D), or perhaps as a result of generalizing over several speakers may have a single consistent grammar. Turning to the role of gender representation in the representation and acquisition of nominal structures, H2 was not confirmed as L2 speakers did not rate preverbal DPs ending in -o/–a higher than preverbal NPs ending in -o/–a, although overt gender marking seems to be an important clue as to overall nominal structure. Regarding the hypothesis of whether -a is indirectly connected to gender, but not -o/–e (H3), our results are ambiguous. We found evidence that -a nominals were rated differently than –o/–e nominals, but the patterns were consistently different for the L2 than the L1 group. Suppose that -a is (indirectly) connected to feminine gender, as Harris proposes. Because gender is subject to concord with D, the existence of null D will be more salient for -a nouns than for nouns in -o/–e. This would be consistent with the higher L1 group ratings for preverbal DP(–a) than for DP(–e), and with the lower ratings for NP(–a) than for NP(–o/–e). However, since feminine gender is a lexically assigned feature, transfer of feminine from Russian to Spanish is not straightforward: a feminine word in Russian may not be feminine in Spanish. Hence, the lower ratings for
José Camacho & Alena Kirova
reverbal DP(–a) than for DP(–o/–e) observed for L2 speakers may reflect this fact: p whereas masculine is the default in Spanish, feminine needs to be assigned lexically, and learners may not have completely acquired that lexical mapping. For example, Kirova’s (2014) findings indicate that L1 Russian-L2 Spanish speakers transfer and rely heavily on their L1 lexical gender values (F, M) when determining the gender of the L2 nouns. Along the same lines, Klassen’s (Forthcoming) study of the representation of gender in L1 and L2 systems mismatched in number of gender classes finds that L1 Spanish-L2 German speakers have shorter naming latencies for nouns that are gender–congruent in Spanish and German, providing support for Salamoura & Williams’s (2007) gender–integrated representation hypothesis (our study did not look at congruency, although this is an area with important implications for our analysis). This picture is compatible with two different explanations. The asymmetry between different types of nominal marking could be morphological: the nature of the contrast relates to the different status of the morphological rule for mapping the feature “f ” to ]a and the default rule that assigns ]o to all other unmarked nouns. However, this analysis does not directly connect word class and nominal type in any obvious way. The second possibility would be that nouns in -a have different syntactic representations than nouns in -o/–e; for - a nouns, the abstract feature “f ” would be represented both on N and on D, whereas for -o/–e, gender is absent from the syntactic structure, and only realized morphologically, as represented in (22). In syntax (22), la casa is represented as having “f ” assigned to N, and an unvalued feature on D, whereas el techo (22) is represented without any gender. After agreement (22), the determiner is valued for gender in la casa, but el techo still lacks any gender marking. Morphological rules (22) map “f ” to the word marker ]a, and a default rule assigns word marker ]o to all other nominal heads (other rules insert lexical roots into terminal nodes). Concord in masculine could be achieved either by a morphological concord rule, or by assuming that N → ]o applies to all the categories of the extended nominal projection (D, N, Adj), following Grimshaw’s (2000) notion of extended category. (22)
a. La casa ‘the house (F)’ (I) Syntax (II) Agreement DP DP D [G_]
NP N [f]
(III) Morphology f → ]a [la] [kasa]
D [ Gf ]
NP N [f]
b. el techo ‘the roof (M)’ (I) Syntax DP D
NP N
(III) Morphology N → ]o [el] [techo]
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
7. Conclusion In this paper, we undertook the exploration of the L2 acquisition of Spanish nominal structures by L1 Russian speakers. We focused on two distinct areas: bare NPs/full DPs, an area where both languages differ, and agreement, where the two languages share similar classification systems and concord across the NP. Our results suggest that NPs and DPs are treated differently by the two groups, and show a different rating depending on the position: while both groups rate pronominal DPs higher than NPs, the difference is much higher for the L1 than for the L2 speakers. This suggests that the L2 learners are beginning to acquire the structure of DPs in Spanish but have not yet fully acquired the word order constraints that regulate null D. In terms of gender, we found that the traditional division between masculine/ feminine–ending words (–o/–a) and words not overtly marked for overt gender (–e) did not yield a significant effect in the acquisition of nominal structure. On the other hand, when words were classified as having a lexically assigned word marker (–a) or not (–o, a default marker -e), we did find a significant effect of word–marking on Nominal Type and Word Order. These results proved to be more compatible with an analysis of feminine as a lexically assigned gender, whereas -o is a default word class marker, as in Harris (1991). Following this proposal, we suggest that nouns like casa ‘house’ include an abstract [f] gender feature that maps to the word marker a, whereas techo ‘roof ’ lacks a gender feature in the syntax, and is assigned a default marker in the morphology. Because only -a is indirectly connected to gender, its presence makes the structure of the DP more salient, hence it aids in the process of L2 acquisition.
References Abney, S. (1987). The English noun phrase in its sentential aspect. Unpublished Doctoral Dissertation. MIT. Baker, M. (2003). Lexical categories. Cambridge: Cambridge University Press. Benedicto, E. (1998). Verb movement and its effects on determinerless plural subjects. In A. Schwegler, B. Tranel, & M. Uribe-Etxebarria (Eds.), Romance linguistics: Theoretical perspectives (pp. 25–40) Amsterdam: John Benjamins. DOI: 10.1075/cilt.160.04ben Bošković, Ž. (2008). What will you have, DP or NP? In E. Elfner & M. Walkow (Eds.), NELS 37: Proceedings of the 37th Annual Meeting of the North East Linguistic Society (pp. 101–114). Amherst, MA: GLSA. Bošković, Ž., & Gajewski, J. (2011). Semantic correlates of the NP/DP parameter. In S. Lima, K. Mullin, & B. Smith, B. (Eds.), NELS 39: Proceedings of the 39th Annual Meeting of the North East Linguistic Society (pp. 121–134). Amherst, MA: GLSA. Bruhn de Garavito, J., & White, L. (2002). L2 acquisition of Spanish DPs: The status of grammatical features. In A. T. Pérez–Leroux & J. M. Liceras (Eds.), The acquisition of S panish morphosyntax: The L1/L2 connection (pp. 153–178). Dordrecht: Springer. DOI: 10.1007/978-94-010-0291-2_6
José Camacho & Alena Kirova Caruso, D. (2011). Nominal phrases in Croatian as DPs. In K. Otaki, H. Takeyasu, & S. Tanigawa (Eds.), Online Proceedings of GLOW in Asia Workshop for Young Scholars. Retrieved from 〈http://faculty.human.mie–u.ac.jp/~glow_mie/Workshop_Proceedings/02Caruso.pdf〉 (4 November 2013). Casielles, E. (1997). Topic, focus and bare nominals in Spanish. Unpublished Doctoral Dissertation. University of Massachusetts. Chierchia, G. (1998). Reference to kinds across language. Natural Language Semantics, 6, 339–405. DOI: 10.1023/a:1008324218506 Contreras, H. (1986). Spanish bare NPs and the ECP. In I. Bordelois, H. Contreras, & K. Zagona (Eds.), Generative studies in Spanish syntax (pp. 25–49). Dordrecht: Foris. Corbett, G. (1991). Gender. Cambridge: Cambridge University Press. DOI: 10.1017/s0022226700015449 Cornips, L., & Hulk, A. (2008). Factors of success and failure in the acquisition of grammatical gender in Dutch. Second Language Research, 24, 267–296. DOI: 10.1177/0267658308090182 Cuza, A., Guijarro–Fuentes, P., Pires, A., & Rothman, J. (2013). The syntax–semantics of bare and definite plural subjects in the L2 Spanish of English natives. International Journal of Bilingualism, 5, 634–652. DOI: 10.1177/1367006911435594 Déprez, V., Sleeman, P., & Guella, H. (2011). Specificity effects in L2 determiner acquisition: UG or pragmatic egocentrism? In M. Pirvulescu (Ed.), Selected Proceedings of the 4th Conference on Generative Approaches to Language Acquisition North America (GALANA 2010) (pp. 27–36). Somerville, MA: Cascadilla Proceedings Project. Franceschina, F. (2005). Fossilized second language grammars. The acquisition of grammatical gender. Amsterdam: John Benjamins. DOI: 10.1075/lald.38 Fukui, N. (1988). Deriving the difference between English and Japanese: A case study in parametric syntax. English Linguistics, 5, 249–270. DOI: 10.9793/elsj1984.5.249 Gillon-Dowens, M., Vergara, M., Barber, H.A., & Carreiras, M. (2010). Morphosyntactic processing in late second–language learners. Journal of Cognitive Neuroscience, 22, 1870–1887. DOI: 10.1162/jocn.2009.21304 Grimshaw, J. (2000). Extended projections and locality. In P. Coopmans, M. Everaert, & J. Grimshaw (Eds.), Lexical specification and insertion (pp. 115–33). Amsterdam: John Benjamins. DOI: 10.1075/cilt.197 Grüter, T., Lew–Williams, C., & Fernald, A. (2011). Grammatical gender in L2: Where is the problem? In N. Danis, K. Mesh, H. Sung (Eds.), Proceedings of the 35th annual Boston University Conference on Language Development (pp. 246–258). Somerville, MA: Cascadilla Press. Grüter, T., Lew-Williams, C., & Fernald, A. (2012). Grammatical gender in L2: A production or a real-time processing problem? Second Language Research, 28, 191–215. DOI: 10.1177/0267658312437990 Guillelmon, D., & Grosjean, F. (2001). The gender marking effect in spoken word recognition: The case of bilinguals. Memory and Cognition, 29, 503–511. DOI: 10.3758/bf03196401 Harris, J. (1991). The exponence of gender in Spanish. Linguistic Inquiry, 22, 27–62 Hawkins, R., & Franceschina, F. (2004). Explaining the acquisition and non–acquisition of determiner–noun gender concord in French and Spanish. In P. Prévost & J. Paradis (Eds.), The acquisition of French in different contexts (pp. 175–206). Amsterdam: John Benjamins. DOI: 10.1075/lald.32.10haw
Does agreement affect the syntax of bare nominal subjects in Russian–Spanish bilinguals?
Ionin, T., Ko, H., & Wexler, K. (2004). Article semantics in L2–acquisition: The role of specificity. Language Acquisition, 12, 3–69. DOI: 10.1207/s15327817la1201_2 Kallestinova, E.D. (2007). Aspects of word order in Russian. Unpublished Doctoral Dissertation. University of Iowa. Kirova, A. (2014). The role of values in gender feature transfer in Russian L1 Spanish L2 speakers. Poster presented at the 44th Linguistic Symposium on Romance Languages, Western Ontario, Canada. Klassen, R. (Forthcoming). The representation of asymmetric grammatical gender systems in the bilingual mental lexicon. Laca, B. (1999). Presencia y ausencia de determinante. In I. Bosque & V. Demonte (Eds.), Gramática descriptiva de la lengua española (Vol. 1; pp. 891–928). Madrid: Espasa. Longobardi, G. (1994). Reference and proper names. Linguistic Inquiry, 25, 609–666. Montrul, S., Foote, R., & Perpiñán, S. (2008). Gender agreement in adult second language learners and Spanish heritage speakers: The effects of age and context of acquisition. Language Learning, 58, 503–553. DOI: 10.1111/j.1467–9922.2008.00449 Montrul, S. & Ionin, T. (2010). Transfer effects in the interpretation of definite articles by Spanish heritage speakers. Bilingualism: Language and Cognition, 13, 449–473. DOI:10.1017/S1366728910000040. Montrul, S., & Ionin, T. (2012). Dominant language transfer in Spanish heritage speakers and L2 learners in the interpretation of definite articles. The Modern Language Journal, 96, 70–94. DOI: 10.1111/j.1540–4781.2012.01278 Sagarra, N., & Herschensohn, J. (2011). Asymmetries in gender and number agreement processing in late bilinguals. In L. A. Ortiz-López, (Ed.), Selected Proceedings of the 13th Hispanic Linguistics Symposium (pp. 169–177). Somerville, MA: Cascadilla Proceedings Project. Sagarra, N., & Herschensohn, J. (2012). The role of proficiency and working memory in gender and number agreement processing in L1 and L2 Spanish. Lingua, 120, 2022–2039. DOI: 10.1016/j.lingua.2010.02.004 Sagarra, N., & Herschensohn, J. (2013). Processing of gender and number agreement in late Spanish bilinguals. International Journal of Bilingualism, 17, 607–27. DOI: 10.1177/1367006912453810 Salamoura, A., & Williams, J.N. (2007). The representation of grammatical gender in the bilingual lexicon: Evidence from Greek and German. Bilingualism: Language and Cognition, 10, 257–275. DOI: 10.1017/s1366728907003069 Snape, N. (2006). The acquisition of the English determiner phrase by Japanese and Spanish learners of English. Unpublished Doctoral Dissertation. University of Essex. Szabolcsi, A. (1983). The possessor that ran away from home. The Lingusitic Review, 3, 89–102. DOI: 10.1515/tlir.1983.3.1.89 Szabolcsi, A. (1987). Functional categories in the noun phrase. In I. Kenesei (Ed.), Approaches to Hungarian 2 (pp. 167–89). Szeged: JATE. Weir, A. (2009). Article drop in English headlines. Unpublished MA Dissertation. University College London. White, L., Valenzuela, E., Kozlowska–Macgregor, M., & Leung, I. (2004). Gender agreement in non–native Spanish: Evidence against failed features. Applied Psycholinguistics, 25, 105–133. DOI: 10.1017/s0142716404001067
Perfecting the past Joanne Markle LaMontagne & María Cristina Cuervo University of Toronto
In studies on Spanish first language (L1) acquisition the Present Perfect (PP) appears together with the Present tense, or immediately after, and Peninsular Spanish children use it productively before the Preterit. Though the PP has become the default form of past reference in many varieties of Peninsular Spanish, this finding is surprising given the Perfect’s morphological and semantic complexity. In a corpus study of longitudinal spontaneous data of two L1 Peninsular Spanish children (María, 1;09–3;11; Emilio, 1;09–3;10), we explore the emergence and distribution of tenses, and the types of uses of the PP. Results are consistent with the absence of a complex, indirect referential meaning for the PP and suggest that children have a simpler meaning for it, and thus having the form does not necessarily imply having the meaning. Keywords: Present Perfect; Peninsular Spanish; L1 acquisition; tense and aspect
1. Introduction Spanish presents the possibility of talking about events that took place in the past by using the Preterit, as in (1), or alternatively, by using the compound Present Perfect (PP), as in (2). (1) ¿Dormiste bien (ayer)? 2sg-sleep-pret well (yesterday) ‘Did you sleep well (yesterday)?’ (2) ¿Has dormido bien (hoy)? 2sg-have-prs sleep-ptcp well (today) ‘Have you slept well (today)?’ Although both forms involve past reference and anteriority, certain denotational differences between the two, such as direct and indirect reference to the moment of speech, are not entirely obvious. This contrast has been also analyzed as one governed by aspectual properties. In the present study we focus on the PP, the morphologically and semantically more complex form, which in Peninsular Spanish is produced before the Preterit in
doi 10.1075/ihll.4.10mar © 2015 John Benjamins Publishing Company
Joanne Markle LaMontagne & María Cristina Cuervo
first (L1) language acquisition, (Hernández Pina, 1984; López Ornat, 1994; Berman & Slobin, 1994; Aguado, 1995; Bel, 2002; among others), and is the most frequent adult form for past reference (Schwenter & Torres Cacoullos, 2008). In spite of previous studies, a gap in research remains with respect to the acquisition of tenses as a system within which each tense carves out a particular space, and whether acquisition of the PP form entails acquisition of its complex temporal and aspectual meaning. In order to investigate these questions, we develop a study on the acquisition of the Spanish PP that focuses on the meaning of the tense, rather than only on its form, as it emerges from the examination of the types of uses found in longitudinal spontaneous data from two Peninsular children. In Section 2 we present the approach to the meaning of the PP that we assume. We combine elements from various approaches to the PP, and discuss findings from the acquisition literature in this light. In Section 3 we introduce our study on L1 acquisition of the Spanish PP; we present the results and discussion in Section 4. Some general conclusions and lines for future research appear in Section 5.
2. The meaning of the Present Perfect and previous studies in acquisition 2.1 Tense or aspect? Crosslinguistically, the PP has been studied extensively, particularly in English (see McCawley, 1971; Comrie, 1976; Binnick, 1991; Klein, 1992; Iatridou, 2003; Iatridou, Anagnostopoulou & Pancheva, 2003; Portner, 2003; von Fintel & Iatridou, 2005; among others). Though many theories have been proposed, the PP is continuously puzzling because it shares properties with both temporal and aspectual forms, thus raising questions about whether it encodes both tense and aspect, or whether its aspectual properties are better understood as consequences of temporal properties. Semantically, the PP has been treated from the perspective of tense or aspect in the literature. Following the Reichenbachian (1947) framework of temporality, the PP is a complex tense that works as a vector, which refers to another tense. Under this view, the PP expresses temporal precedence between the event time (E) (the time (or interval) at which the event holds), the speech (S) (the moment of speech) and reference (R) times. The reference time is relative to a reference point or interval at which the underlying event described in the clause is considered. For instance, in the notation E_R, S, the event time of the verbal predicate trabajar mucho (‘to work a lot’) precedes (E_) the reference (R) and speech (S) times, both of which coincide (R,S), as in (3). (3) Ana ha trabajado mucho. Ana 3sg-has-pres work-ptcp a lot ‘Ana has worked a lot.’
Perfecting the past
Similarly, under Rojo & Veiga’s (1999) approach to temporal relations in Spanish, the PP expresses a past event (E), one that is anterior to a reference point (R) that is simultaneous with the moment of speech (S), or E_R,S. Alternatively, when treated as aspect, the PP is said to denote a resultant state that follows from a prior event (Parsons, 1990; Klein, 1992, 1994; Kamp & Riley, 1993; Vlach, 1993; Giorgi & Pianesi, 1998; among others). In (4), the telic event of Felipe finishing the homework is mapped onto a state, meaning that Felipe is now in the state that results from having finished the homework. (4) Felipe ha terminado la tarea. Felipe 3sg-has-pres finish-ptcp the homework ‘Felipe has finished the homework.’ In this study we consider the denotation of the PP first and assume that the core difference in meaning between the PP and Preterit is temporal, and that this distinction can be further operationalized and made more explicit via adverbial modification.
2.2 Spanish temporal relations and the Perfect time span In their analysis of the Spanish tenses, Bello (1847) and Rojo & Veiga (1999) propose that temporal relations can be described as anterior (A), simultaneous (S), or posterior (P) to the moment of speech, or to a given reference point. Under this view, anteriority to the moment of speech can be expressed directly, as in the Preterit, or indirectly, as in the PP, as shown in Figure 1.
Moment of speech
Anterior Preterit
A
S
Simultaneous Present
P
A S Present Perfect
Juan comió tres tortas. Juan ha comido tres tortas. Figure 1. Temporal relations in Spanish (adapted from Bello (1847) and Rojo & Veiga (1999))
Posterior Future
P
A
S
P
Joanne Markle LaMontagne & María Cristina Cuervo
Rojo & Veiga (1999) maintain that the Preterit expresses a relation of direct anteriority to the moment of speech (Juan comió tres tortas/‘Juan ate three cakes’) whereas the PP – antepresente (‘before present’) for Bello (1847) – expresses indirect anteriority to the moment of speech via another reference point (Juan ha comido tres tortas/‘Juan has eaten three cakes’). Essentially, the PP conveys a past event that is anterior to a reference point that is simultaneous with the moment of speech, and thus includes a double reference component. In terms of Cowper’s analysis (2003, 2005), structurally, the Spanish Preterit is monoclausal and the PP is bi-clausal. Under this view, the Preterit expresses the feature Precedence, while the past participle component of the PP bi-clausal structure expresses the features Precedence and Event (the tense of the auxiliary corresponding to another inflectional head). When we combine Rojo & Veiga’s (1999) and Cowper’s (2003, 2005) frameworks, the Preterit expresses a relation of pure Precedence (anteriority) to the moment of speech, while the PP expresses Precedence (anteriority) and Simultaneity.1 In addition to expressing anteriority, Iatridou (2003), Iatridou et al. (2003), and von Fintel & Iatridou (2005) propose that the PP introduces a time span, known as the Perfect Time Span (PTS). The PTS is a modified version of the Extended Now Theory (XN) (McCoard, 1978; Dowty, 1979; Vlach, 1993; among others), which argued that the Perfect introduces a time interval that extends into the past from the reference time, or moment of speech. In the revised approach, the PTS is anchored by a right (RB) and left boundary (LB) of which the LB extends towards the past and can be set by an adverbial, such as desde (‘since’), and the RB or bracket extends to, and can include, the moment of speech and is set by the Present tense, as in Figure 2. The diagram in Figure 2 integrates Rojo & Viega’s (1999) analysis of temporal relations in Spanish and Cowper’s (2005) Precedence to Simultaneity framework, along with a representation of the PTS that is argued for in Iatridou (2003) and related work. The moment of speech affirms the relational nature of the PP as having a double reference component (i.e. Precedence to Simultaneity) and being past of the present, while the lower portion of the diagram indicates how precedence relations in Spanish align with the PTS, a time interval that is set by the Present and can be referred to implicitly, or explicitly by adverbials in PP utterances.
. This can also be expressed with Reichenbach’s Times if we assume there is a difference between two times being the same, as in E = R of the Preterit, and being simultaneous in the sense of one including another, as in E < R ⊃ S for the PP (see Zagona (2007) for an analysis of part-whole relations between Times).
Perfecting the past Moment of speech
Simultaneous Present
Anterior Preterit Me visitó ayer. Me ha visitado varias veces desde la semana pasada.
Posterior Future
Anterior Present Perfect
(.......(.......(.......(.......) PERFECT TIME SPAN
Left boundary (adverbial)
Right boundary (tense)
Figure 2. A representation of the Perfect Time Span and Spanish temporal relations (adapted from Rojo & Veiga (1999), Cowper (2003, 2005), Iatridou (2003), Iatridou et al. (2003) and von Fintel & Iatridou (2005))
In Figure 2, the PTS is explicit and contained by the RB/bracket extending from the present moment into the past and set at the LB by the adverbial complement desde la semana pasada (Me ha visitado varias veces desde la semana pasada/‘S/he has visited me many times since last week’), which indicates the starting point of the PTS. In contrast, the Preterit does not include a time span and instead expresses an event as directly anterior to the moment of speech. In Figure 2, the event of visiting is further temporally specified via the adverbial complement ayer (Me visitó ayer/‘S/he visited me yesterday’). The PTS is a parenthesis, not a reference, and can divide into subintervals. These subintervals characterize the granular nature of the PTS and the aspectual consequences it yields. For instance, when a verbal predicate possesses the subinterval property, a universal interpretation is achieved (Iatridou, 2003). This means the predicate holds throughout some interval that extends from a certain past moment up to and including the moment of speech, exhibits atelic Aktionsart (i.e. an unbounded situation that lacks an inherent endpoint), and expresses the lexical semantics of states or activities, as in (5) with stative verb vivir, ‘to live’. (5) Isabela ha vivido en Quebec desde Navidad. Isabela 3sg-has-pres live-ptcp in Quebec since Christmas ‘Isabela has been living in Quebec since Christmas.’
Joanne Markle LaMontagne & María Cristina Cuervo
Under the PTS approach, (5) maintains there is a time span whose RB/bracket is set by the Present and whose LB is Christmas, and for every interval of which it is true that Isabela lives in Quebec. So, Isabela has lived in Quebec since Christmas and continues to live there up to and including the moment of speech. When the predicate is located within the PTS by an existential operator, however, an existential (or experiential) perfect interpretation holds. This means that the event expressed by the verbal predicate has occurred in a given time span for at least one subinterval, as in (5) in which Isabela has lived in Quebec since Christmas for at least one subinterval. For an existential perfect reading to hold, the underlying eventuality of the verbal predicate can exhibit any type of Aktionsart. Sentence (6), with an achievement predicate (ganar/‘to win’), expresses that the PTS from Christmas to the Present consists of at least five subintervals at which time it is true that Isabela wins a competition. Contrary to the universal perfect, Isabela’s winning five competitions is not required to hold true at the present moment. (6) Isabela ha ganado cinco concursos desde Navidad. Isabela 3sg has-pres win-ptcp five competitions since Christmas ‘Isabela has won five competitions since Christmas.’ In this study we combine Iatridou’s PTS approach with Rojo & Veiga’s (1999) and Cowper’s (2005) Precedence to Simultaneity frameworks to offer an alternative analysis of the Spanish PP, one that also revises the interval component of this form. In particular, we assume that the PP makes indirect reference to the moment of speech, unlike the Preterit, which makes direct reference to it. We also assume that the PP denotes a time interval/span, whereas the Preterit does not.
2.3 Previous findings in L1 acquisition of the Perfect Studies on the L1 acquisition of Spanish have found that the PP appears together with the Present tense, or immediately thereafter, and that Peninsular Spanish children use the PP productively before the Preterit (Hernández Pina, 1984; López Ornat, 1994; Berman & Slobin, 1994; Aguado, 1995; Bel, 2002; among others). While by age 2;0 this pattern is true in certain varieties of Peninsular Spanish, the opposite tendency is found in Latin American children, who by 2;0 produce the Preterit, then the PP. With respect to the distribution patterns of tenses in young children, Aguado (1995) reports that in a study of the spontaneous speech of L1 Spanish Peninsular children (N = 35, mean age 2;06) in recorded structured play sessions, the Present tense is the most prevalent form (64.4%), as in Cortes and Vila (1991), followed by the PP (8.9%), Imperfect (1.2%), and Preterit (0.4%). Bel (2002) reports similar findings for Spanish and Catalan, and shows that present, past and future forms are used appropriately with present, past and future references.
Perfecting the past
In a crosslinguistic study that includes the examination of Frog Story (Mayer, 1969) narratives by Spanish-speaking children (aged 3;0–9;0 from Argentina, Chile, and Spain), Berman & Slobin (1994) observe that the Perfect is completely absent from the Latin American children’s samples, but is used extensively by Peninsular children from age 3;0. Three-year-old children from Madrid use the PP to express past actions that can be inferred from pictures depicting result-like events, rather than actions with no explicit visual support, producing, for instance, Aquí se ha subido (‘Here he has climbed’) based on an illustration of a boy sitting in a tree ([S3a-3;6], Berman & S lobin, 1994:250). By age 5;0 children begin to use the PP for events that are not visually depicted, i.e. ones that can only be inferred from preceding situations (Berman & Slobin, 1994:251). Developmentally, the authors propose, for the PP of Peninsular children, an “expansion of the temporal distance between a completed past situation and the current state of affairs at utterance (=narrative) time” (1994:251). In a similar vein, dialectal differences have also been reported in the L1 acquisition of the PP in Scottish- and American-English speaking children (Gathercole, 1986). Gathercole (1986) observed that in PP contexts in spontaneous speech, pre-school aged Scottish children produced the Perfect 60% of the time, whereas American children produced it less than 1% of the time. These patterns reflect the frequency of use in these particular grammars. Gathercole argues that these findings stress the importance of frequency of input, particularly when in conjunction with other factors, such as cognitive and syntactic simplicity. Although she investigated the relative frequency of the PP and the Preterit in different contexts or semantic categories (“uses of the perfect”), and assessed the possibility of children interpreting or using the PP in simpler ways, she did not directly tie the ranking of cognitive simplicity to an explicit analysis of the meaning of the Perfect tenses. To sum up, L1 studies have observed that children can differ in the timing and course of acquisition of the PP along dialectal lines. The early emergence and use of the PP in children correlates positively with frequency of input and there seems to be a course of acquisition in terms of the meaning of the Perfect used by children, although the details of the different meanings or how one developmental stage differs from the other are left rather vague. Now, if we centre on the analysis of the Perfect presented in previous sections, it is clear that the PP is a morphologically and semantically more complex tense than the Preterit. In particular, the Preterit is a morphologically and semantically simple tense that denotes direct anteriority to the moment of speech (¿Te portaste bien?/‘Did you behave well?’; Rojo & Veiga, 1999; Cowper, 2003, 2005). The PP, in contrast, is composed of the morphologically complex form of two verbs (i.e. auxiliary haber+past participle) and semantically locates the time of the event (E) in relation to a reference time (R) that is distinct from the speech time (S), even if it is simultaneous with it
Joanne Markle LaMontagne & María Cristina Cuervo
(Rojo & Veiga, 1999; Cowper, 2005) or it is a time span that includes it (Iatridou, 2003 and related work). To the morphological and semantic complexity of the PP, some authors have added explicit syntactic complexity in terms of a bi-clausal analysis (Cowper, 2003; cf. Zagona, 2007). Given this analysis, it is completely puzzling that Peninsular Spanish children (and English speaking Scottish children) use the PP productively before the Preterit. Thus, in this study we investigate the acquisition of the PP in Peninsular children with a special focus on the task of mapping tense forms to temporal meanings.
3. Study 3.1 Specific research questions: The puzzle We depart from two observations from previous studies: (7) a. Peninsular children acquire the PP early and before the Preterit b. The PP is very frequent in Peninsular Spanish (adults) Given the complexity of the PP discussed above, finding (7a) is puzzling even in light of finding (7b). The puzzle is solved if, in spite of appearances, children are different from adults in their use of the PP. Alternatively, children may indeed be matching the input from adults but (some of the uses of) the PP of Peninsular adults may not have the meaning assigned by the analysis in Section 2. Therefore, we ask the question of whether early uses of the PP involve the acquisition of a complex, indirect referential system. In other words, we address the issue of whether the use of the PP by young children is evidence of the mapping of the PP as a form to its complex temporal denotation. In order to explore this question we first studied the relative distribution of the various tenses used by the children. With respect to PP instances, we analyzed their meaning in terms of uses of the PP (as defined in the literature) and modification by temporal adverbials.
3.2 Participants We conducted a CHILDES (MacWhinney, 2000) corpus study and analyzed longitudinal spontaneous data from the López-Ornat (1994) and Vila (1990) databases of two L1 Peninsular Spanish children, María (1;9–3;11) from Madrid and Emilio (1;9–3;10) from Girona. The children were recorded at home during bath, play and mealtime sessions with at least one Spanish-speaking parent.
3.3 Methods
Perfecting the past
In order to study the distribution of tenses in Emilio and María, we first consulted recent studies on variation (see Schwenter & Torres Cacoullos (2008, henceforth S&TC); Howe (2013)) which provide evidence that the PP is the default or ‘neutral’ form of past reference in certain varieties of Peninsular Spanish (e.g. Alicante, Madrid, Valencia) and that the Preterit is the default form of past reference in Latin American varieties, in countries such as Argentina, El Salvador, Mexico, and Peru. According to these studies, in Peninsular Spanish the PP is preferred 54% of the time while Preterit is preferred 46% of the time, as is illustrated in Table 1. Table 1. Distribution of Preterit and Present Perfect forms (frequency %) across a variety of Spanish dialects, as reported in S&TC and Howe (2013) Present Perfect
Preterit
Argentina
6% (N = 47)
94% (N = 783)
El Salvador
22% (N = 838)
78% (N = 2616)
Mexico
15% (N = 331)
83% (N = 1903)
Peru
27.1% (N = 972)
72.9% (N = 2616)
Spain
54% (N = 965)
46% (N = 827)
In Emilio and María’s samples we coded all conjugated verbs in the indicative according to the following: PRES(ent), PERF(Present Perfect), IMPerf(ect), PRET(erit), PLUPerf(ect), and FUT(ure). With respect to the emergence and production of tense, we measured emergence by the age at which a particular temporal form first appeared and productiveness by the age at which two different verbal forms appeared with two different inflections. The second component we investigated in Emilio and María’s data was the meaning of the PP, specifically the types of uses of this form. Crosslinguistically we find four prototypical uses of the PP: perfect of result, recent past (‘hot news’) perfect, existential (experiential) perfect, and universal (continuous, persistent situation) perfect (McCawley, 1971; Comrie, 1976; Said, 1976; Portner, 2003; Iatridou, 2003, among others). Spanish exhibits all four types (Howe, 2013), as seen in (8)–(11). (8) Perfect of result Se ha secado la tinta del bolígrafo. Se 3sg-has-pres dry-ptcp the ink of-the pen ‘The pen has dried up.’ (9) Recent past or ‘hot news’ perfect Esta mañana hemos ido a nadar. This morning 1pl-have-pres go-ptcp to swim ‘This morning we have gone swimming.’
Joanne Markle LaMontagne & María Cristina Cuervo
(10) Existential (or experiential) perfect Desde 2010 he visitado a mis padres dos veces. Since 2010 1sg-have-pres visit-ptcp my parents two times ‘Since 2010 I have visited my parents twice.’ (11) Universal (or continuous, persistent situation) perfect a. He vivido en Bogotá desde enero de 2010. 1sg-have-pres live-ptcp in Bogota since January of 2010 ‘I have lived in Bogota since January 2010.’ b. Marta todavía no ha llegado. Marta yet not 3sg-have-pres arrive-ptcp ‘Marta has not arrived yet.’ The perfect of result focuses on the resultant state that arises after an event or change of state takes place and expresses the outcome that is relevant at the present time, as in (8) in which the ink has dried up and is still dried up at the moment of speech. The recent past (or ‘hot news’) perfect reports an event or change of state that recently occurred, as in (9) in which we went swimming on the same day as the utterance. The existential (or ‘experiential’) perfect asserts that the subject has been involved in an event or has been in a certain state during a specific timespan in her lifetime, as in (10) in which the speaker has visited her parents twice since 2010. Again, with the existential perfect instances of the specific event must occur during the PTS, but are not required to hold true at the RB/bracket, i.e. the present moment. Finally, the universal (or continuous, persistent situation) perfect expresses a situation that holds throughout the PTS and one that must continue up to and include the RB/ bracket (present moment), as in (11a) in which the speaker has lived in Bogota and continues to live there. For a universal interpretation to hold, the verbal predicate generally exhibits atelic Aktionsart, as in the stative verb vivir (‘to live’) and activity verbs. In addition, negative polarity, as in (11b), also favours the PP and this interpretation. In our study, we coded every instance of the PP by type of use as in Gathercole’s (1986) semantic categories, which are based on the four prototypical uses of Perfects: RES (perfect of result), REC (recent perfect or ‘hot news’ perfect), EXP (experiential, or Iatridou’s ‘existential’ perfect), and PERS (perfect of persistent situation, or Iatridou’s ‘universal’ perfect). Finally, a third area of study included an examination of the PP in combination with temporal adverbial complements. Specifically, we were interested in finding out whether adverbials are indicative of a secondary reference in children’s PP uses, that is, a time span that extends from the present towards the past, or a reference time that includes the moment of speech. From their study on the Preterit and PP contrast in adult Peninsular and Mexican Spanish, S&TC observe that although approxi-
Perfecting the past
mately 75% of all PP and Preterit tokens occur without any co-occurring temporal expression, the PP combines most frequently with proximate (hoy/‘today’), frequency (siempre/‘always’), and ya (‘already, finally, now’) adverbials in both corpora, although at different rates. Adverbials were coded by type, including: connective (luego/‘later, then’), durative (desde/‘since’), frequency (siempre/‘always’), proximate (hoy/‘today’), and specific (este año/ ‘this year’). Ya was coded separately because it can also combine with other temporal adverbials, such as después ya (‘then’), etc. In their variationist study, S&TC find that utterances which contained proximate and frequency adverbials in Peninsular Spanish (N = 118) were much more likely (91%) to be instances of the PP than the Preterit; utterances with ya (N = 91) also favoured the PP, at 75%. In contrast, in Mexican Spanish proximate and frequency adverbs (N = 109) combined with the Perfect 49% of the time, while ya (N = 175) co-occurred with the Perfect only 10% of the time. In line with S&TC, in our study we coded each case of PP+temporal adverbial modifier by type, including: Conn(ective), Dur(ative), Freq(uency), Prox(imate), Spec(ific) and Ya (ya, ya+any other combination).
4. Results 4.1 Emergence, productiveness, and relative frequency of tenses In both Emilio and María’s data we find that the Present and PP tenses emerge around 1;10, and that Emilio uses the Present productively by 2;01 and María by 1;10. As for the PP, Emilio uses this form productively by 2;00 and María does so by 1;11. While the Imperfect emerges next for Emilio (2;07) and is used productively by 3;00, we find an opposing pattern in María’s data in that the Preterit (1;10) emerges before the Imperfect (2;00) and the Preterit is used productively by 2;00 and the Imperfect by 2;10. The Preterit first appears in Emilio’s data by 2;09 and then productively by 3;10. These findings are summarized in Table 2. Table 2. Age of emergence and productivity of tenses for Emilio and María Emilio
María
Emergence
Productivity
Emergence
Productivity
PRES
1;10
2;01
1;10
1;10
PERF
1;10
2;00
1;10
1;11
IMPerf
2;07
3;00
2;00
2;10
PRET
2;09
3;10
1;10
2;00
Joanne Markle LaMontagne & María Cristina Cuervo
With respect to the frequency of indicative tensed verbs in Emilio’s data, we observe that overall the Present is the preferred tense across all three stages of development (75.1%), followed by the PP (12.6%), Imperfect (3.1%), and Preterit (0.7%). We also observe a PP and Preterit trade-off: as Emilio’s use of the Imperfect and Preterit increases throughout the first to third stages, his PP use decreases. These findings are summarized in Table 3. Table 3. Distribution of indicative tensed verbs across three stages of development for Emilio (%, tokens) Tense Stage
PRES
PERF
IMPerf
PRET
PLUPerf
FUT
Total
1;10–2;06
70.9 (168)
24.1 (57)
0 (0)
0 (0)
0 (0)
5.0 (12)
25.4 (237)
2;07–2;11
80.8 (273)
10.3 (35)
0.6 (2)
0.6 (2)
0 (0)
7.7 (26)
36.3 (338)
3;00–3;11
72.5 (258)
7.0 (25)
7.6 (27)
1.4 (5)
0.3 (1)
11.2 (40)
38.2 (356)
Total
75.1 (699)
12.6 (117)
3.1 (29)
0.7 (7)
0.1 (1)
8.4 (78)
931
María’s data exhibit the same pattern of distribution of indicative tensed verbs across all three stages of development and within the same range as that of Emilio’s. The Present is the preferred tense (70%), followed by the PP (13.7%), Imperfect (4.1), and Preterit (3.3%). The overall percentage of PP and Preterit remains quite constant from stage one to stage three, while use of the Imperfect increases gradually. These findings are summarized in Table 4. Table 4. Distribution of indicative tensed verbs across three stages of development for María (%, tokens) Tense Stage
PRES
PERF
IMPerf
PRET
PLUPerf
FUT
Total
1;09–2;03
72.6 (594)
14.5 (119)
1.2 (10)
2.0 (16)
0 (0)
9.7 (79)
29.8 (818)
2;02–2;11
67.5 (596)
13.2 (117)
4.1 (36)
5.1 (45)
0 (0)
10.1 (89)
32.2 (883)
3;01–3;11
70.0 (731)
13.3 (139)
6.4 (67)
2.8 (29)
0.9 (9)
6.6 (69)
38.0 (1044)
Total
Perfecting the past 70.0 (1921)
13.7 (375)
4.1 (113)
3.3 (90)
0.3 (9)
8.6 (237)
2745
Focusing now on the four past tenses attested, the children’s reference to the past starts off with a strong preference for the PP (exclusive in Emilio), with a few instances of Imperfect and Preterit in María (Stage 1: Emilio PP 100%; María PP 82%, PRET 11%, IMPerf 7%). Interestingly, the PRET and the PP seem to cover the same semantic space at this stage, as evidenced by María’s use of one tense or the other in the same situation and the same verb:2 María 2;00
Utterance #519 Utterance #520 Utterance #521 Utterance #522
S’am bañao Se cayó e patito Se rompieron S’a caído
PP PRET PRET PP
‘They’ve taken a bath’ ‘The ducky fell down’ ‘They broke’ ‘It has fallen down’
By the third stage (3;00 – 3;11), the semantic space of past reference has developed into a more complex system in which different tenses are used to express various temporal relations (Stage 3: Emilio PP 43%, PRET 9%, IMPerf 47%, PLUPerf 2%; María PP 57%, PRET 12%, IMPerf 27%, PLUPerf 4%). The use of the Pluperfect provides evidence that the children are capable of making an indirect reference (anteriority to a time anterior) to the moment of speech.
4.2 Types of uses of the Present Perfect With respect to the distribution of types of uses of the PP we find that the most frequent types of uses in Emilio’s data are RES (59.0%) and REC (37.6%), followed by PERS (2.6%) and EXP (0%), as in (12)–(14). Over the course of Emilio’s three stages of development we see that while his RES use decreases, his REC Perfects increase. These findings are summarized in Table 5. Table 5. Distribution of the types of uses of the Present Perfect across three stages of development for Emilio (%, tokens) Types of uses of the PP Stage
EXP
PERS
REC
RES
Total
. Further to this, in María we found a few cases of use of PP preceded by cuando ‘when’ and luego ‘then’ in sequencing events, a use that, although found in Peninsular Spanish, typically corresponds to the PRET.
Joanne Markle LaMontagne & María Cristina Cuervo
1;10–2;06
0 (0)
3.5 (2)
31.5 (18)
65.0 (37)
48.7 (57)
2;07–2;11
0 (0)
2.8 (1)
42.9 (15)
54.3 (19)
29.9 (35)
3;00–3;11
0 (0)
0 (0)
44.0 (11)
52.0 (13)
21.4 (25)
Total
0 (0)
2.6 (3)
37.6 (44)
59.0 (69)
117
(12) Se ha escondido. SE 3sg-has-pres hide-ptcp ‘S/he has hidden.’
RES (3;11)
(13) Yo he visto vacas. I 1sg-have-pres see-ptcp cows ‘I have seen cows.’
REC (3;10)
(14) No lo he quitado yo. No it 1sg-have-pres remove-ptcp I ‘I haven’t removed it.’
PERS (2;09)
In María’s data we observe a similar pattern for the most frequent types of uses. REC (48.0%) and RES (43.0%), as in (15)–(16), are the most prominent uses and account for nearly 90% of all of María’s PPs. Again, we see a decrease in RES uses over time and an increase in REC. María’s data, unlike Emilio’s, includes both PERS (6.1%) and EXP (3.0%) uses, as in (17)–(18). These uses, however, account for less than 10% of María’s PP data. These findings are summarized in Table 6. Table 6. Distribution of uses of the Present Perfect across three stages of development for María (%, tokens) Types of uses of the PP Stage
EXP
PERS
REC
RES
Total
1;09–2;03
0 (0)
0 (0
32.0 (38)
68.0 (81)
31.7 (119)
2;02–2;11
0.8 (1)
5.1 (6)
35.9 (42)
58.1 (68)
31.2 (117)
3;01–3;11
7.2 (10)
12.2 (17)
72.0 (100)
8.6 (12)
37.1 (139)
Total
3.0 (11)
6.1 (23)
48.0 (180)
43.0 (161)
375
(15) Yo te he hecho un café. I you 1sg-have-pres make-ptcp a coffee ‘I have made you a coffee.’
REC (2;03)
(16) Se ha caído. SE 3sg-has-pres fall-ptcp ‘It has fallen.’
RES (2;06)
Perfecting the past
(17) No has tomado tu sopa. No 2sg-have-pres take-ptcp your soup ‘You haven’t eaten your soup.’
PERS (3;09)
(18) He salido yo, allí en Madrid. 1sg-have-pres go out-ptcp I, there in Madrid ‘I have gone out, there in Madrid.’
EXP (3;01)
It is important to note that in both children, most cases of PERS correspond to negated statements compatible with a negation of a REC or RES (see examples (14) and (17)) rather than a true PERS corresponding to the negation of the existential EXP (as in I have never seen a whale) or what can be thought of as a state introduced by negation that continues to the moment of speech (as in He has not arrived yet). Overall, it seems that both Emilo and María’s types of uses of the PP point to pure anteriority or Precedence meaning.3 Even if we do not analyze the contrast PP-PRET based on aspectual features, it is important to establish whether the use of these tense forms is restricted by aspectual class. Accordingly, we studied the aspectual class of the verbs in PP and PRET, and in each type of PP use. We found that both tenses are used with all four aspectual classes: activities (correr ‘run’, llorar ‘cry’), accomplishments (tomarse ‘drink up’, lavar ‘wash’), achievements (romperse ‘break’, caerse ‘fall down’) and states (ser, estar ‘be’, tener ‘have’, querer ‘want’). The same applies to the four types of PP use, with the only unsurprising exception of there being no state verbs in the Perfect of Result.
4.3 Temporal adverbial modification If the children’s use of the PP does not entail a secondary reference, we expect an absence of temporal adverbials. In general this is what we find. Overall, only 11.6% of Emilio and María’s combined total PP uses contain temporal adverbial modifiers. Adverbials appear in small numbers, their use increasing with age. In both children we observe that Ya and Prox adverbials, such as ahora ‘now’, este año ‘this year’, and hoy ‘today’, comprise the most prominent types. Table 7 shows that temporal adverbials modify 9.4% of Emilio’s total PPs. Again, Ya and Prox adverbials are the most prominent types, as in (19)–(21). We see an increase in the use of Ya from stage 1 to 2, but a decrease in stage 3.
. A preliminary look at 3,705 tokens from María’s and Emilio’s parents reveals that overall REC and RES consist of approximately 90% of their PP uses, a finding that is consistent with this type of PP in Peninsular children. EXP and PERS account for the remaining PP uses in the parents’ data.
Joanne Markle LaMontagne & María Cristina Cuervo
Table 7. Distribution of temporal adverbial types for Present Perfect across three stages of development for Emilio (%, tokens) Temporal adverbial type Stage
Conn
Freq
Dur
Prox
Spec
Ya
Total
1;10–2;06
0 (0)
0 (0)
0 (0)
0 (0)
0 (0)
100.0 (2)
3.5 (2)
2;07–2;11
0 (0)
0 (0)
0 (0)
0 (0)
0 (0)
100.0 (7)
20.0 (7)
3;00–3;11
0 (0)
0 (0)
0 (0)
50.0 (1)
0 (0)
50.0 (1)
8.0 (2)
Total
0 (0)
0 (0)
0 (0)
9.0 (1)
0 (0)
91.0 (10)
9.4 (11)
(19) Ya he hecho pipí. Already 1sg-have-pres do-ptcp pee ‘I already went pee.’
Ya (2;04)
(20) Ya te he dicho que sí. Already you 1sg-have-pres tell-ptcp that yes ‘I’ve already you told you yes.’
Ya (2;09)
(21) Hoy me ha venido a buscar papá. Today me 3sg-has-pres come-ptcp to pick up dad ‘Today dad has come to pick me up.’
Prox (3;10)
As for María, though she has four times the number of tokens as Emilio, we find very few adverbials in her data, which only consist of 12.2% of her total Perfect use. Table 8 shows a steady increase in María’s use of Conn, Freq, Dur, Prox, and Spec adverbials, as in (22)–(26), and a slight decrease with the most prominent type Ya, as in (21), from stage 2–3. Table 8. Distribution of temporal adverbial types for Present Perfect across three stages of development for María (%, tokens) Temporal adverbial type Stage
Conn
Freq
Dur
Prox
Spec
Ya
Total
1;09–2;03
0 (0)
0 (0)
0 (0)
12.5 (1)
0 (0)
87.5 (7)
6.7 (8)
2;02–2;11
0 (0)
0 (0)
0 (0)
10.0 (1)
0 (0)
90.0 (9)
8.5 (10)
3;01–3;11
7.1 (2)
10.7 (3)
7.1 (2)
32.1 (9)
14.3 (4)
28.6 (8)
20.1 (28)
Total
4.3 (2)
6.5 (3)
4.3 (2)
23.9 (11)
8.7 (4)
52.2 (24)
12.2 (46)
(21) Yo ya he terminado. I already 1sg-have-pres finish-ptcp. ‘I’ve already finished.’ (22) No, yo ahora me he
cambiado
Ya (2;11)
Perfecting the past
No I now me 1sg-have-pres change-ptcp de nombre. of name
Prox (3;09)
‘No, I have now changed names.’ (23) … cuando has estao de vacaciones. when 2sg-has-pres be-ptcp of vacations ‘… when you were on holidays.’
Spec (3;09)
(24) Ah, es que nunca me has visto. Freq (3;07) Ah, 3sg-be-pres that never me 2sg-have-pres see-ptcp ‘Ah, it’s that you’ve never seen me.’ (25) Yo muy poco rato he estao. I very little time 1sg-be-pres be-ptcp ‘I have been for very little time.’
Dur (3;11)
(26) Y bueno, luego me he ido a casa a jugar con And well after me 1sg-have-pres go-ptcp home to play with mis juguetes. my toys ‘And well, then I have gone home to play with my toys.’
Conn (3;06)
Overall, we observe that there is little temporal modification in both children. This, in itself, may be indicative of an absence of secondary reference, or is at least compatible with it. The relatively few adverbials that do appear in PP utterances are mostly ya (‘already’) and hoy (‘today’). Ya locates the event time before a reference time, which can be the speech time and therefore it does not require a complex time reference. Proximate adverbials such as ahora (‘now’), hoy (‘today’), and particularly este año (‘this year’) are more probable modifiers of a Reference time (a time span) which includes the speech time, that is, they may be indicators of a complex meaning of the PP. Their incidence, however, is extremely low, with only 1 case for Emilio (less than 1% of his PP production) and 11 cases for María (3% of her PP production). In both children all or most of these adverbials appear in the last stage studied (3;00 to 3;11).
5. Conclusions Based on analyses of the complex temporal reference that Perfect tenses express, we set out to explore the hypothesis that the early acquisition of the PP reported for children speaking Peninsular Spanish entails acquisition of the form, but does not necessarily mean children have acquired the complex meaning. Our analysis is supported by the
Joanne Markle LaMontagne & María Cristina Cuervo
initial absence and slow emergence of alternative past tense forms from which the PP would distinguish itself. Overall, we found that the PP was the most frequent form of past reference (70%) in the children’s data and that aspectual class did not restrict PP or Preterit uses. In contrast, S&TC show that in their corpus of Peninsular Spanish adults the PP and Preterit are almost equally used at a relative frequency of 54% and Preterit 46% respectively. We then studied the PP in Emilio and María in terms of four uses defined in previous literature and associated each use with a certain meaning explicitly in terms of the analysis of tenses adopted. REC and RES account for 94% of Emilio’s and María’s PP uses. These types of uses do not entail a secondary reference time, and are consistent with a hodiernal or non-specific perfective past interpretation. With respect to adverbial modification, neither child uses the PP in combination with Left Bracket adverbials, such as desde (‘since’), which introduce a past time span. Instead, the few adverbials they do use consist mostly of proximate (ahora/‘now’) and ya (‘already’) adverbials. Thus, the event time is presented as recent or, in S&TC’s words, as a default “hodiernal perfective with no necessary pragmatic inferences of current relevance” (S&TC: 23). Nevertheless, S&TC’s claim that the PP is the default form of past reference in this variety cannot adequately explain what adults, or children, are doing in the absence of an account of the types of meanings of the PP. Our findings are, in general, consistent with the absence of a complex, indirect referential meaning. We believe this simpler meaning associated with children’s PP arises because there is no actual reference to a time distinct from speech time: either there is no reference to a time span extending to the past or to a reference time simultaneous with, but distinct from, speech time. We acknowledge, however, that one of the limitations of this corpus study is the impossibility of knowing with certainty whether the absence of some PP uses really means that the children have not yet acquired all four types. In sum, the results are consistent with there being a mismatch between form and (adult) meaning. This “mapping problem” has antecedents in the L1 acquisition of Spanish in the domain of subjunctive mood (Pérez-Leroux, 2008). If this is true, then the frequency of the input would have direct impact in the acquisition of the PP as a form, but not of its complex meaning. Alternatively, however, it might be that children’s production matches their input more closely, and that in Peninsular Spanish the adult PP is also used with the simplex meaning of reference to recent or indefinite past. Instead of the PP expressing a time span and a past event whose reference is simultaneous with the speech time (present), as in Bello (1847), Rojo & Veiga (1999) and Iatridou (2003), the PP of Peninsular Spanish children and adults might express a direct relation of precedence in reference to the speech time, one that is consistent with the Preterit. A preliminary comparison between children and adults confirmed that REC and RES are the most frequent types of PP uses overall, yet it appears that Emilio and Maria’s PPs are less complex than
Perfecting the past
those of their parents whose PP uses extend into a more distant past and include more experiential and universal PP meanings. In general, these findings are consistent with those of Gathercole (1986) who found that Scottish children mirror their parents’ most frequent types of PP uses. Though there is no explicit analysis of the four types of uses of PPs in S&TC, the authors attribute the frequent PP use in adults to irrelevant, today, and indeterminate uses of the PP, which, like Emilio’s and María’s, seem to be consistent with a PP that lacks reference to a moment distinct from the speech time, or a time span. In order to assess this alternative thoroughly it is necessary to study parental input more closely. In addition, a comprehensive study should compare these varieties to children’s PP uses and meanings in dialects in which the Preterit is employed more frequently than the PP. These are the next steps in our research. Ultimately, the analysis of the meaning must be associated with syntactic structure. There are at least two possibilities for this association: (i) simplicity of meaning corresponds directly to simplicity of structure (less evaluated Times, fewer syntactic projections) or (ii) simplicity of meaning may arise from the same levels of complexity of syntactic structure, the difference lying in the array of features that are specified or projected in each phrase. This issue is left for future research.
References Aguado, G. (1995). El desarrollo del lenguaje de 0 a 3 años. Madrid: Ciencias de la educación preescolar y especial. Bel, A. (2002). Early verbs and the acquisition of tense feature in Spanish and Catalan. In A.-T. Pérez-Leroux & J.M. Liceras (Eds.), The acquisition of Spanish morphosyntax: The L1/L2 connection. (pp. 1–34). Dordrecht: Kluwer. DOI: 10.1007/978-94-010-0291-2_1 Bello, Andrés. (1847). Gramática de la lengua castellana destinada al uso de los americanos. Madrid: Edaf (1984 edition). DOI: 10.1515/zrph.2005.708 Berman, R. & Slobin, D. (1994). Relating events in narrative: A crosslinguistic developmental study. Hillsdale, NJ: Lawarence Erlbaum Associates. DOI: 10.1017/s0272263100014443 Binnick, R. (1991). Time and the verb: A guide to tense and aspect. Oxford: Oxford University Press. DOI: 10.1017/s002222670000044x Comrie, B. (1976). Aspect. Cambridge: Cambridge University Press. DOI: 10.1017/s0022226700011178 Cowper, E. (2003). Tense, mood and aspect: A feature-geometric approach. Ms. University of Toronto. Cowper, E. (2005). The geometry of interpretable features: INFL in English and Spanish. Language, 81(1), 10–46. DOI: 10.1353/lan.2005.0012 Dowty, D. (1979). Word meaning and Montague grammar. Dordrecht: Kluwer. DOI: 10.1007/978-94-009-9473-7 Gathercole, V.C. (1986). The acquisition of the present perfect: Explaining differences in the speech of Scottish and American children. Journal of Child Language, 13, 537–560. DOI: 10.1017/s0305000900006875
Joanne Markle LaMontagne & María Cristina Cuervo Giorgi, A., & Pianesi, F. (1998). Tense and aspect: From semantics to morphosyntax. Oxford: Oxford University Press. DOI: 10.1017/s1040820700002614 Hernández Pina, F. (1984). Teorías psicosociolingüísticas y su aplicación a la adquisición del español como lengua materna. Madrid: Siglo veintiuno editores. DOI: 10.2307/342474 Howe, C. (2013). The Spanish perfects. Pathways of emergent meaning. Houndmills: Palgrave MacMillan. Iatridou, S. (2003). A little bit more on the English perfect. In A. Alexiadou, M. Rathert, & A. von Stechow (Eds.), Perfect Explorations (pp. 133–152). Berlin: Mouton de Gruyter. DOI: 10.1515/9783110902358 Iatridou, S., Anagnostopoulou, E., & Pancheva, R. (2003). Observations about the form and meaning of the perfect. In A. Alexiadou, M. Rathert, & A. von Stechow (Eds.), Perfect explorations (pp. 153–204). Berlin: Mouton de Gruyter. DOI: 10.1515/9783110902358 Kamp, H., & Riley, U. (1993). From discourse to logic. Dordrecht: Kluwer. Klein, W. (1992). The present perfect ‘puzzle’. Language, 68, 525–552. DOI: 10.2307/415793 Klein, W. (1994). Time in language. London: Routledge. López Ornat, S. (1994). La adquisición de la lengua española. Madrid: Siglo XXI. DOI: 10.1177/014272379601604706 MacWhinney, B. (2000). The CHILDES project: Tools for analyzing talk (Third Edition). Mahwah, NJ: Lawrence Erlbaum Associates. DOI: 10.1177/014272370002006006 Mayer, M. (1969). Frog, where are you? New York, NY: Dial Press. McCawley, J. (1971). Tense and time reference in English. In C. Fillmore & T. Langendoen (Eds.), Studies in linguistic semantics (pp. 97–113). New York, NY: Holt, Rinehart and Winston. DOI: 10.1017/s0022226700003571 McCoard, R. (1978). The English perfect: Tense choice and pragmatic inferences. Amsterdam: North-Holland. DOI: 10.2307/325264 Parsons, T. (1990). Events in the semantics of English: A study of subatomic semantics. Cambridge, MA: The MIT Press. Pérez-Leroux, A.T. (2008). Subjuntivo y conciencia de la subjetividad en la adquisición infantil del lenguaje y la teoría de la mente. Revista de Logopedia, Foniatría y Audiología, 28, 90–98. DOI: 10.1016/s0214-4603(08)70048–2 Portner, P. (2003). The (temporal) semantics and (modal) pragmatics of the perfect. Linguists and Philosophy, 26, 459–510. Reichenbach, H. (1947). Elements of symbolic logic. New York, NY: Macmillan. DOI: 10.2307/2268978 Rojo, G., & Veiga, A. (1999). El tiempo verbal: Los tiempos simples. In I. Bosque & V. Demonte (Eds.), Gramática descriptiva de la lengua española (pp. 2867–2934). Madrid: Espasa. Schwenter, S., & Torres Cacoullos, R. (2008). Defaults and indeterminacy in temporal grammaticalization: The ‘perfect’ road to perfective. Language Variation and Change, 20, 1–39. DOI: 10.1017/s0954394508000057 Vila, I. (1990). Adquisición y desarrollo del lenguaje. Barcelona: Graó. Vlach, F. (1993). Temporal adverbials, tenses, and the perfect. Linguistics and Philosophy, 16, 231–283. DOI: 10.1007/bf00985970 von Fintel, K., & Iatridou, S. (2005). Since Since. Ms. MIT. Zagona, K. (2007). Some effects of aspect on tense construal. Lingua, 117, 464–502. DOI: 10.1016/j.lingua.2005.08.004
The protracted acquisition of past tense aspectual values in child heritage Spanish* Alejandro Cuza & Lauren Miller Purdue University
This study examines the knowledge of past tense aspectual distinctions in Spanish among 19 Spanish-English bilingual children born and raised in the United States. We compare their results with those of 12 of the children’s parents, who are long-term immigrants of Mexican background. We predicted more difficulties among the bilingual children with increasing age as well as strong correlations between performance and language dominance. As expected, the bilingual children showed low production of the imperfect form in characterizing situations, crucially with eventive predicates, but no deficits with the use of the preterit. In contrast to what was expected, target performance was not correlated with language dominance, and we found no correlation between performance and developmental age. However, at the individual level, the older children outperformed the younger children despite more prolonged contact with English. This contrasts with previous research claiming L1 attrition throughout the life span of the bilingual child to account for heritage speakers’ difficulties. Regarding the children’s parents, they all behaved at ceiling and showed no signs of attrition. Keywords: Child heritage Spanish; tense and aspect; crosslinguistic influence effects; L1 attrition
1. Introduction The acquisition of past tense aspectual distinctions in Spanish occurs without incident in non-pathological first (L1) language development, as evidenced by a proportional use of preterit and imperfect tense forms by the age of 3 (Hernández-Pina, 1984; Pérez-Pereira, 1989; Sebastian & Slobin, 1994). This success in the acquisition
* We would like to thank the two anonymous reviewers for their helpful comments and suggestions as well as the audience of the 2013 Hispanic Linguistics Symposium. We would also like to thank all the parents and children for their time and assistance.
doi 10.1075/ihll.4.11cuz © 2015 John Benjamins Publishing Company
Alejandro Cuza & Lauren Miller
process, however, is hardly the case in second (L2) language learners (Cuza, 2010; Montrul & Slabakova, 2002; Salaberry, 1999; Salaberry & Shirai, 2002), adult heritage speakers (Montrul, 2002) or bilingual children (Cuza, Pérez-Tattam, Barajas, Miller, & Sadowski, 2013; Silva-Corvalán, 2003, 2014) as far as aspectual selection is concerned. As we will discuss in detail shortly, one recurring explanation for the observed aspectual deficits among adult and child bilinguals is related to the role of crosslinguistic influence as well as limited input and use. Crosslinguistic influence refers to the positive (facilitation) or negative (interference) effects that one linguistic system might have on another as a result of their typological differences or similarities (Jarvis & Pavlenko, 2008; Odlin, 1989).1 Montrul (2002), for example, argues that the morphosyntactic variability that adult Spanish heritage speakers typically display with past tense aspectual distinctions stems from incomplete acquisition during the early stages of development as a result of limited input and transfer from English, a language that differs from Spanish in regards to the instantiation of aspectual features. Montrul defines incomplete acquisition as the outcome of an interrupted process of language development stemming from input reduction during early childhood (Montrul, 2008). However, when looking at bilingual children of different ages, Cuza et al. (2013) found patterns of incomplete acquisition in the case of the imperfect as well as evidence of child L1 attrition in the target use of the preterit. Child L1 attrition refers to the loss of previously developed grammatical properties as a child ages, as well as potential convergence with L2 forms, due to increased contact with the dominant L2 accompanied by reduced minority language use (see Polinsky, 2011 for recent discussion). These results lead the authors to conclude that both L1 attrition and incomplete acquisition may characterize the acquisition of tense and aspect morphology in child heritage Spanish. More recent research argues for a featural reassembly of L1 functional properties leading to the formation of a featural matrix that is intrinsically variable – albeit not incomplete – stemming from low language activation and use in a language contact scenario (Putnam & Sánchez, 2013). This approach focuses on the ongoing process of language reanalysis throughout the lifespan rather than a static outcome at any one time. Despite the breadth of research mentioned above, the precise nature of heritage language acquisition remains unclear due to several common limitations of previous research. First, it has focused primarily on adult heritage speakers at the university
. Crosslinguistic influence effects have been found from the L1 into the L2 but also from the L2 into the L1 among long-term immigrants and bilingual children and in almost all grammatical areas but primarily in syntax and phonology (Au, Knightly, Ju, & Oh, 2002; Müller & Hulk, 2001). It is a precursor of L1 attrition as well as incomplete acquisition.
The protracted acquisition of past tense aspectual values in child heritage Spanish
level, and, with the exception of very few recent studies, claims about incomplete acquisition or L1 attrition among Spanish heritage speakers have been made without examining the developmental process these bilinguals undergo during earlier stages of development, either cross-sectionally or longitudinally. The exploration of the child-adult connection is essential to testing existing claims, and to provide a reliable linguistic explanation about the nature and source of heritage speakers’ difficulties. Furthermore, elicited production data remain largely underexplored, with more studies looking at interpretation or grammatical intuition, task types which, due to the specific nature of heritage language learners, do not reliably measure grammatical competence. We aim to address these shortcomings and add to previous research by examining the elicited production of aspectual properties in a group of nineteen (n = 19) simultaneous Spanish-English bilingual children born and raised in the United States. Their results are compared to those of twelve (n = 12) of the children’s parents, who were born and raised in Mexico but have been living in the United States for an average of 14 years. Following previous research in child bilingual acquisition, we implement an elicited production task (Crain & Thornton, 1998) and investigate the potential correlations between target aspectual production and two main factors affecting the language acquisition process: crosslinguistic influence (Montrul, 2002; Müller & Hulk, 2001) and language dominance (Argyri & Sorace, 2007; Paradis, 2001; Pérez-Leroux, Cuza & Thomas, 2011). We predict that English-dominant children will have more difficulties with target aspectual use than Spanish-dominant or balanced bilingual children. Specifically, we expect a protracted development of the imperfect form, a form not available in English. Moreover, we anticipate finding an effect by lexical class. For instance, participants may have more difficulty with the use of the imperfect with telic predicates (accomplishments and achievements), which, in day-to-day use, appear more often in the preterit, or with the use of the preterit with stative verbs, which more often appear in the imperfect. In what follows we summarize the main aspectual differences between English and Spanish, followed by a literature review of previous research in aspectual development in Section 3. Section 4 outlines the study and the results. The discussion is provided in Section 5, followed by the conclusions in Section 6.
2. Past tense aspectual values in Spanish and English English and Spanish behave differently in regard to their representation of past tense aspectual values in that, in the past tense, Spanish expresses the difference between imperfective and perfective aspect morphologically, whereas the English past tense
Alejandro Cuza & Lauren Miller
morpheme -ed does not differentiate these two meanings (Comrie, 1976; de Miguel, 1992; Smith, 1997). The two morphemes used in Spanish to express this difference are illustrated in (1a) and (1b): (1) a. Dora comió helado ayer. Dora ate-pret ice cream yesterday ‘Dora ate ice cream yesterday.’
(perfective aspect)
b. Dora comía helado todas las noches después Dora ate-imp ice cream all the nights after de cenar. (imperfective aspect) eating dinner
‘Dora ate ice cream every night after dinner.’
In (1a) the intended meaning in Spanish is that of a completed/episodic event in the past, and therefore the preterit should be used (-ió). In (1b), however, the intended meaning is that of a repeated/characterizing action in the past, and thus the imperfect form should be used. In English, as represented in the translation of (1a) and (1b), there is no morphological differentiation between the two aspectual classes, and the preterit tense is used in both cases. However, it is also possible to express the imperfect meaning with the periphrastic expression used to or would (Dora used to eat/would eat ice cream every night after dinner). This is what is known in the literature as Grammatical Aspect (Comrie, 1976). English and Spanish also diverge in the use of the past progressive. In Spanish, an ongoing activity in the past can be expressed either using the imperfect form or the past progressive form. In English, only the past progressive is allowed, as represented in (2a) and (2b). (2) a. Ramiro cruzaba la calle cuando vio Ramiro was crossing the street when saw a María. to Mary
(past ongoing)
‘Ramiro was crossing the street when he saw Mary.’
b. Ramiro estaba cruzando la calle cuando vio Ramiro was crossing the street when saw a María. to Mary
(past ongoing)
‘Ramiro was crossing the street when he saw Mary.’
Although grammatical aspect is useful to describe the “different ways of viewing the internal temporal constituency of the situation” (Comrie, 1976:3), researchers have also described aspectual distinctions based on the lexical properties of tense heads. This classification is referred to as Lexical Aspect (Vendler, 1967), which divides verbs
The protracted acquisition of past tense aspectual values in child heritage Spanish
into four classes according to their specific lexical temporal properties: states, activities, accomplishments and achievements. States (have, know) and activities (eat, play) are considered atelic predicates with no intrinsic endpoint. Accomplishments (feed the dog) and achievements (fall, break) on the other hand are considered telic as they have a clear endpoint. This classification is useful but, again, limited as verbs often move from one aspectual class to the other depending on their internal arguments (Schmitt, 1996; Verkuyl, 1972). Verkuyl (1972) and others propose that aspectual notions are determined compositionally; that is, they depend on the specific compositional relation between the verbal head and its complements. For example, the verb to eat has an activity lexical meaning in (3a) but an accomplishment meaning in (3b). (3) a. Dora comió esta mañana. (activity) Dora ate this morning ‘Dora ate this morning.’ b. Dora comió una galleta esta mañana. (accomplishment) Dora ate a cookie this morning. ‘Dora ate a cookie this morning.’ In (3a), the verb to eat has an activity aspectual class but it is coerced into an accomplishment in (3b) due to the insertion of the direct object una galleta (‘a cookie’). This is known in the literature as Compositional Aspect. The interaction between the three types of aspect exacerbates difficulties that any learners may have, and may lead to reliance on frequent or prototypical combinations of lexical, compositional and grammatical aspect in Spanish, such as stative verbs appearing only with imperfect morphology or telic verbs (accomplishments and achievements) appearing only in preterit morphology (Montrul, 2002). Given these differences between the English and Spanish aspectual systems, Spanish-English bilinguals have to learn the specific aspectual properties of each specific tense head, and make the necessary form/meaning connections. This is a challenging task for English-speaking learners of Spanish, heritage speakers and child bilinguals, as they tend to transfer English aspectual values and the morphological representation from English, which can select either aspectual meaning, and is, therefore, the default or unmarked form. In what follows, we discuss recent research on the bilingual acquisition of these structures, outline our research questions and postulate our hypotheses.
3. The acquisition of aspectual values in Spanish Research on the acquisition of tense and aspect in Spanish monolingual children shows a proportional use of each morpheme by age 3 (Hernández Pina, 1984; PérezPereira, 1989; Sebastian & Slobin, 1994). For example, Pérez-Pereira (1989) conducted
Alejandro Cuza & Lauren Miller
an in-depth study of the acquisition of verbal morphology with 109 monolingual Spanish children between the ages of 3 and 6 and found stable and accurate production of the preterit and imperfect by the age of 4. The children were presented with a short preamble and were asked to fill in a blank with the grammatically correct word based on what they had heard. With regards to the accurate use of verbal morphology, the author’s results showed stable uses of the preterit and imperfect by age 4 with 98% of imperfect tokens correct and 74% of preterit tokens correct. The higher accuracy with the imperfect may be due to the regularity of its morphology. The preterit, in comparison, is highly irregular and may require higher amounts of input for complete mastery. Other studies have replicated these results, finding that monolingual acquisition of tense and aspect does not pose any special difficulty (Hernández Pina, 1984; Sebastian & Slobin, 1994). Research with L2 learners of Spanish (Cuza, 2010; Montrul & Slabakova, 2002; Montrul & Slabakova, 2003; Pérez-Leroux et al., 2008; Salaberry, 1999), Spanish heritage speakers (Montrul, 2002) and Spanish-English bilingual children (Cuza et al., 2013; Silva-Corvalán, 1994, 2014), however, suggest that, for these populations, tense and aspect can be more challenging. Montrul (2002), for example, did not find the same level of mastery among Spanish heritage speakers vis a vis L2 learners. The author tested 16 simultaneous Spanish-English bilinguals, 15 L1 Spanish child L2 learners of English, 8 L1 Spanish late L2 learners of English and 20 monolingual S panish speakers, who, at the time of testing, were all adults. The author presented the participants with four tasks: a cloze task, an oral narrative, a truth value judgment task and an acceptability judgment task. In all tasks, simultaneous bilinguals were less accurate than child L2 learners of English who were in turn less accurate than those who learned English as adults (who performed like Spanish monolinguals). Montrul concluded that age of onset of bilingualism is negatively correlated with maintenance or acquisition of the minority language. Stative verbs in the preterit were found to be the most difficult, followed by achievements in the imperfect and other uses of the imperfect. The author also found a high amount of individual variation, but concludes that results do suggest the presence of incomplete acquisition and L1 attrition. However, it is difficult to draw conclusions about processes that occurred during early childhood without testing children, as mentioned earlier. Cuza et al. (2013) examined the production of tense and aspect morphology cross-sectionally among a group of Spanish-English bilingual children via a story retelling task (Little Red Riding Hood). They compared the children’s production of preterit and imperfect forms to that of adult heritage speakers, monolingual children and adult monolinguals. When production across different age groups was compared, the results suggested that both incomplete acquisition and L1 attrition play a role in determining heritage speakers’ competence. Specifically, the preterit tense, which is acquired earlier in monolingual contexts, was fully acquired by the older
The protracted acquisition of past tense aspectual values in child heritage Spanish
bilingual children, while the adult bilinguals had regressed to a usage of the preterit tense similar to that of the younger bilinguals, suggesting that they had undergone L1 attrition of preterit morphology. However, the use of the imperfect tense remained low across all three bilingual groups, in contrast to monolinguals who show a steady increase in the use of the imperfect morphology with age, suggesting incomplete acquisition with the imperfect which is acquired later than the preterit in monolingual contexts. Despite these informative results, the lack of elicited production data prevents previous research in the acquisition of tense and aspect among heritage speakers of Spanish from drawing any definitive conclusions. Elicited production is necessary since, as Silva-Corvalán (1994) states, the absence of a structure in spontaneous production is not necessarily evidence of its absence in a speaker’s grammar. In addition, elicited production tasks are especially important given that heritage speakers often underperform on metalinguistic and written tasks but quite often do very well with oral production (Montrul, Davidson, de la Fuente, & Foote, 2013). Moreover, it remains to be seen, when using controlled production tasks, which combinations of lexical and grammatical aspect, if any, cause difficulties for bilingual children of different ages. Such detailed data would allow us to understand not only the acquisition of challenging structures in heritage Spanish and how to address these difficulties, but also the nature of language acquisition in general and the variables that affect it such as transfer, input, language dominance and age. For the aforementioned reasons, the current study is necessary to address this considerable gap. An important dimension in child bilingual acquisition is the role of language dominance and its correlation with crosslinguistic influence and the development of target linguistic representations in the minority language (Argyri & Sorace, 2007; Genesee & Nicoladis, 2007; Pérez-Leroux et al., 2011; Yip & Matthews, 2006). PérezLeroux et al. (2011) examined the acquisition of clitic climbing among 23 SpanishEnglish bilingual children living in Toronto. Results from a repetition task showed that those children who arrived to Toronto at a later age (sequential bilinguals), and who were more Spanish-dominant, maintained higher levels of pre-verbal object clitic placement, typical of the monolingual Spanish norm (i.e. Jazmin lo quiere ver esta noche ‘Jazmin wants to see him tonight’). The simultaneous bilinguals, in contrast, reported to be English-dominant and showed a preference for the post-verbal option (Jazmin quiere verlo esta noche ‘Jazmin wants to see him tonight’). Although dominance might be a determining factor, it does not always provide clear-cut results. For example, Argyri & Sorace (2007) examined the acquisition of different complementizer phrase (CP) structures among 16 Greek-dominant children living in Greece and 16 English-dominant children living in the UK. Results show unidirectional transfer from English into Greek among the English-dominant bilinguals but no transfer effects from Greek into English among the Greek-dominant bilinguals. These results
Alejandro Cuza & Lauren Miller
suggest that dominance may not be the sole determiner of crosslinguistic influence (Hulk & Müller, 2000; Yip & Matthews, 2000).
3.1 Research questions and hypotheses Taking into account previous research on the source of heritage speakers’ difficulties, we examine the following research questions: RQ1: To what extent do Spanish-English bilingual children have knowledge of preterit versus imperfect distinctions in Spanish? And if difficulties are found, in which contexts do they occur? RQ2: Will older children show more difficulties than younger children due to a greater length of exposure to English as a dominant societal language leading to greater variability in the aspectual domain? RQ3: Can the difficulties, if any, be accounted for in terms of crosslinguistic influence from English and language dominance? That is, will English-dominant children show more difficulties than Spanish-dominant children or balanced bilinguals?
Based on these research questions, we posit the following hypotheses: H1: Bilingual children will show low production of the imperfect form in contexts where the imperfect should be used. Specifically, we predict more deficits with the use of the imperfect with telic predicates as those verbs are usually found in the preterit. Conversely, there will be no difficulties with the use of the preterit, as this is the default option in English. H2: Older children will show more difficulties than younger children. This is due to the fact that older children have longer exposure and use of English as dominant language in the school system and, consequently, less exposure to and use of Spanish. H3: There will be strong correlations between target performance and patterns of language dominance. We expect more variability among English-dominant children than Spanish-dominant children.
In what follows, we present the study as well as the results and discussion.
4. The study 4.1 Participants Nineteen (n = 19) Spanish-English bilingual children participated in the current study. The children were born and raised in the United States, and were exposed to both languages at an early age via television, friends, parents and siblings. They all lived in Northwestern Indiana at time of testing and came from low socioeconomic
The protracted acquisition of past tense aspectual values in child heritage Spanish
ackgrounds. Age at time of testing ranged from 5;5 to 11;1 (M = 8;2, SD = 1.60). b The children attended English-only schools but had received some Spanish instruction as part of an after-school Spanish program. The patterns of minority language use at home were very positive with most children, with 70% reported as speaking Spanish to the mother. The rest of the children were reported to speak both English and Spanish or only English. In addition, most of the parents (79%) reported initiating a topic with their children only in Spanish or most often in Spanish. About a third of the children (36%) were reported to speak English with their siblings, while 26% were reported to speak Spanish and 26% were reported to speak both. All of the children’s parents except one were born and raised in Mexico and had been living in the United States for more than 10 years. They completed a parental language history questionnaire and a child language background questionnaire (adapted from Pérez-Leroux et al., 2011). The language history questionnaire inquired as to place of birth, age of arrival to the U.S., length of residence and patterns of language use. The questionnaire also included a self-assessment of the parents’ linguistic proficiency in the four skills of language via a Likert scale ranging from ‘basic’ (1) to ‘excellent’ (4). The child language background questionnaire elicited information on each child’s patterns of language use and bilingual dominance, among other variables. Twelve (n = 12) of the children’s parents completed the elicitation task, serving as a control baseline. The parent data was intended to help us determine whether the difficulties children had, if any, were a replication of the parents’ input. Parents had a mean age of 34 years at time of testing and were born and raised in Mexico. Their mean length of residence (LOR) in the U.S. was 14 years. Most of the parents indicated having a basic knowledge of English (mean score, 1.4/4) and good/fluent knowledge of Spanish (mean score, 3.4/4). A reviewer points out that, methodologically speaking, testing the parents as baseline is a limitation given the fact that they might have undergone L1 attrition themselves after 14 years living in the U.S. However, these parents all spoke Spanish in their daily life as they had very limited knowledge of English. They lived in a community where Spanish was part of their day-to-day interaction with family members and friends. Thus, L1 attrition among these immigrants is very unlikely.2 Table 1 summarizes the participants’ demographic information:
. While no evidence of attrition was found among the parents, even if there had been, they would still be the valid baseline for their children, since whatever input they provide to their children is the child’s “target” grammar (see Montrul & Sánchez-Walker (2013) for recent discussion).
Alejandro Cuza & Lauren Miller
Table 1. Participants’ demographic information
Age at testing
Bilingual Children 5;5–11;1 (n = 19) (M = 8;2; SD = 1.60) Parents (n = 12)
Place of birth Mean age of arrival
Mean LOR
United States
Birth
Birth
20.3
14
23–39 Mexico (M = 34.3; SD = 4.70)
4.2 Tasks Data was collected via a question-after-story task, whose purpose was to elicit preterit and imperfect past tense forms in contexts where either one or the other was most appropriate. Following previous research (Cuza, 2008, 2010), we organized the test tokens taking into consideration predicate type (statives, activities, accomplishments, achievements) and situation type (characterizing, episodic), for a total of eight conditions. There were a total of 45 test tokens (5 tokens per 9 conditions) plus 4 practice items.3 Each token consisted of a preamble followed by a question. The participant was required to respond to the question based on the information provided in the preamble and a photo, as represented in (4). (4) Preamble: Normalmente Dora dibuja estrellas pero ayer no. ‘Normally, Dora draws stars, but not yesterday.’ Prompt: ¿Ayer, qué? (a photo of Dora with a drawing of a heart was presented) ‘Yesterday, what?’ Target: Ayer, Dora dibujó un corazón. ‘Yesterday, Dora drew a heart.’ The task was administered to the participants orally and visually via text and photos with the aid of Microsoft PowerPoint. The testing was conducted by the investigators in the school setting, the participant’s home or in a private office. Table 2 outlines the structures under examination. All responses where digitally recorded and later coded for statistical analysis. Accurate responses (expected use of the preterit or the imperfect according to the preamble) were coded as 1 and non-accurate responses (unexpected use of the preterit, the imperfect or the present) were coded as 0. For the quantitative analysis, we used the proportion correct per participant for each condition. Whenever the child did not respond to one of the items, we excluded that item from the total number of items in that condition.
. We also tested the use of the imperfect with a past progressive interpretation, in addition to the other eight conditions, which we do not report for the purpose of the present study.
The protracted acquisition of past tense aspectual values in child heritage Spanish
Table 2. Structures under examination (without preamble) Predicate type
Characterizing
State
(5) De niña, Dora era una bailarina. (6) Ayer, Dora no quiso caramelos.
Activity
‘As a little girl, Dora was a ballerina.’
‘Yesterday, Dora didn’t want candy.’
(7) De niña, Dora jugaba al baloncesto
(8) Ayer, Dora tocó la guitarra.
‘As a little girl, Dora played basketball.’
‘Yesterday, Dora played the guitar.’
Accomplishment (9) De niña, Dora hacía su cama todos los días.
Achievement
Episodic
(10) Ayer, Dora pintó un corazón.
‘As a little girl, Dora made her bed every day.’
‘Yesterday, Dora painted a heart.’
(11) Antes, Dora siempre perdía.
(12) La mamá de Dora llamó.
‘Before, Dora always lost.’
‘Dora’s mother called.’
Language dominance was determined based on parental reports and scalar ratings for English and Spanish ranging from “not fluent” (1) to “completely fluent” (4), following Pérez-Leroux et al. (2011). To obtain the dominance score for each child, we subtracted the child fluency ratings given for English from those given for Spanish. Thus, scores above 0 (positive range of the scale) were interpreted as Spanishdominant, scores under 0 were interpreted as English-dominant (negative range of the scale) and 0 scores were interpreted as balanced.
5. Results and discussion 5.1 Preterite versus imperfect forms As predicted, the bilingual children showed very low proportions of imperfect form use in characterizing contexts. With the preterit, both the children and adults showed high levels of production except with stative predicates; with this predicate type, both groups showed low levels of preterit use. These results are represented in Figure 1. The decreased use of the preterit with stative verbs in episodic contexts might stem from the fact that some of the verbs used, such as querer (‘to want’), change their meanings in the preterit. For example, Dora no quería caramelos (‘Dora did not want any candy’) means that she did not want any candy but there is a possibility that she had some. With the preterit, however, the latter meaning is not possible. It requires the interpretation that she did not want any candy, nor did she have any. Thus, it is possible that both the children and parents used the imperfect instead of the preterit to imply the corresponding aspectual meaning.
Alejandro Cuza & Lauren Miller
Bilingual children
imp characterizing
Achievement
Accomplishment
Activity
State
Achievement
Accomplishment
Activity
Parents
State
1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00
pret-episodic
Figure 1. Proportion correct of imperfect and preterit form use by predicate and situation type per group
The proportions of preterit and imperfect forms produced were submitted to an analysis of variance (ANOVA) with repeated measures, with group as the betweensubjects variable and eventuality type and situation type as the within-subject variables. Overall ANOVA results showed significant main effects for group (F1,29 = 41.4, p < .000). As predicted, the bilingual children showed much lower levels of target preterit form use with episodic conditions and target imperfect form use with characterizing conditions when compared to the parents. To further examine the observed means between groups and conditions, we conducted independent sample t-tests. Regarding the use of the imperfect in characterizing situations, we found highly significant differences between the two groups across all predicate types (states, p < .000; activity, p < .000; accomplishment, p < .000; achievements, p < .000). Regarding the use of the preterit, results showed no significant differences between the two groups except with states (states, p < .001; activity, p = .307; accomplishment, p = .307; achievements, p = 0.88). It is clear that, as far as the bilingual children are concerned, difficulties with past tense aspectual morphology lie with the imperfect, confirming Hypothesis 1. The low levels of imperfect form use were related for the most part to an overextension of the preterit but also to the use of the present tense. As a reviewer points out, the use of the present instead of the imperfect should not be considered as a non-target response or “error” given its own imperfective aspectual value. However, given the context provided by the preamble, which clearly indicated that the situation occurred
The protracted acquisition of past tense aspectual values in child heritage Spanish
in the past via a temporal adverb, we strongly believe that a marked preference for the present among some of the bilingual children might be considered as a case of avoidance, especially if this occurred crucially in cases where the imperfect was expected rather than the preterit. This is confirmed by the fact that only 5% of the children used the present in contexts where the preterit was required versus 19% in contexts where the imperfect was expected. Thus, the use of the present instead of the imperfect or the preterit was unrelated to lexical class, which is not characteristic of the reportative present. Table 3 represents the proportion of preterit, imperfect, present, and other forms used by group in each predicate type and situation type. Table 3. Proportion of morpheme production by situation type per group EVENTIVE
CHARACTERIZING
PRET
IMP
PRES
OTHER
PRET
IMP
PRES
OTHER
Children
0.69
0.17
0.05
0.09
0.34
0.41
0.19
0.06
Parents
0.85
0.11
0
0.04
0.01
0.92
0.05
0.03
The existing differences we have found between the parents and children contradict previous research arguing that the difficulties that heritage speakers have stem from the acquisition of an a priori attrited contact variety (Missing Input Competence Divergence Hypothesis) (Rothman, 2007). As far as aspectual differences are concerned, with this particular type of population, this is not the case, as all of the parents performed at ceiling.
5.2 Older versus younger children Regarding the relationship between age and morpheme production, a Pearson correlation analysis showed no strong correlations between performance and age in months (r = .324, p = .177), disconfirming Hypothesis 2. In contrast to what was predicted, younger children and older children did not behave significantly differently from each other. However, a look at the individual data shows that older children have an advantage with both the preterit and imperfect forms. This is represented in Figure 2. As shown in Figure 2, the use of the preterit is stable (M = 72%, range 40%–100%) and remains around the 70% target rate among most participants past the age of 76 months (6;4 years of age). Although this is a much lower target proportion compared with monolingual children of a similar age (Pérez-Pereira, 1989; Sebastian & Slobin, 1994), we found no sharp decrease in their production with increasing age. In the case of the imperfect, the children’s behavior is less linear than with the preterit (M = 42%, range 0%–95%), but, again, we do not see a sharp decrease with age either, except
Alejandro Cuza & Lauren Miller 1.00 0.90 0.80 0.70 0.60 0.50
IMP
0.40
PRET
0.30 0.20 0.00
65.5 75.7 75.5 76.4 79.0 81.0 89.5 90.8 92.0 92.7 107.1 102.7 108.7 109.3 109.3 112.3 123.7 129.6 142.2
0.10
Figure 2. Average proportion of preterit and imperfect form use in episodic and characterizing contexts by age in months
in the case of one child. Overall, we see an improvement in the level of morpheme production despite more extensive exposure to English after the age of six through immersion in the school system. Thus, intense exposure to English does not seem be a barrier preventing the native morphosyntactic system from continuing to develop towards a more adult-like state. It simply develops more slowly, leading to a protracted development, as we have seen with other morphosyntactic structures including differential object marking (Cuza, Miller & Pérez-Tattam, 2014). These results are also along the lines of recent data documenting a protracted development in the acquisition of tense and aspect morphology among Spanish-English bilingual children in the U.S. (Silva-Corvalán, 2014). Although longitudinal data is necessary to confirm this trend, the fact that there is an improvement in the target production of both the imperfect and preterit forms with developmental age questions previous research arguing for child L1 attrition in the lifespan of the child (Polinsky, 2011). The nature of our elicited production task, as mentioned, also allowed us to analyze performance based on predicate type in order to better understand the nature of the difficulties found, and this analysis revealed differential outcomes according to the type of predicate. For example, participants often overextended the preterit to contexts where the imperfect should have been used primarily with activity and accomplishment predicates, as predicted. This is not unusual within a usage-based approach to language acquisition given that activities and accomplishment predicates don’t usually appear in the imperfect form in day-to-day use. Similarly, some participants overextended the imperfect or the present to stative episodic contexts where the preterit is required, as has been found in previous research (Montrul, 2002). This is not
The protracted acquisition of past tense aspectual values in child heritage Spanish
s urprising as stative verbs are most frequently formed in the imperfect. In cases where bilingual children are unsure, they may resort to the use of the preterit, the unmarked form, due to a clash between aspectual class and morphosyntactic form, transfer from English, or the form in which that verb appears most frequently due to input reduction. In this context, it is important to keep in mind that there were instances where both the preterit and the imperfect were possible depending on the aspectual meaning of the phrase (Ayer, Dora no quiso/quería caramelos ‘Yesterday, Dora didn’t want candy’) as discussed earlier.
5.3 English-dominant versus Spanish-dominant children As mentioned in the introduction, we were interested in examining the potential correlation between target aspectual use and language dominance. To do so, we looked at the target proportion of preterit and imperfect use across all predicate types and compared the participants’ target scores with their parent-reported dominance in Spanish. Most of the children (53%, 10/19) were reported to be English-dominant, 21% (4/19) as Spanish-dominant and 26% (5/19) as balanced. A Pearson correlation analysis showed no significant correlation between the proportion correct across all the contexts examined and the individual dominance scores (r = –.241, p = .320), disconfirming Hypothesis 3. Performance was not related to language dominance, which confirms previous research (Kupisch, 2007). The lack of strong correlations may stem from the fact that most of the children were Englishdominant. In addition, they all came from very similar sociocultural backgrounds, and were very homogeneous in regards to patterns of language use at home. Future research with a larger group of children from more heterogeneous backgrounds and with differing language dominance profiles is necessary to investigate this issue further.
6. Conclusions The present study examined the acquisition of past tense aspectual morphology among a group of Spanish heritage children living in the U.S. Our main goal was to investigate the role of language dominance and crosslinguistic influence in the use of preterit versus imperfect morphology. We were also interested in analyzing the effects of predicate and situation type, and whether the difficulties these children have, if any, were the result of child L1 attrition due to increasing exposure to English throughout the lifespan. In contrast with previous research, we implemented an elicited production task and controlled for language dominance via parental reports.
Alejandro Cuza & Lauren Miller
We have found significant difficulties in the target production of aspectual morphology among the bilingual children. The difficulties lied almost exclusively in the use of the imperfect, especially with activity and accomplishment predicates. The use of the preterit is overall stable and does not present a problem for child bilinguals. Bilingual children tend to overextend the preterit when unsure, as this is the default/ unmarked representation in English, leading to infelicitous constructions. Another strategy, which does not seem to be directly related to transfer from English, is the use of the most frequent morpheme with which a particular verb occurs in the input when the child is unsure, possibly an effect of reduced input. In contrast with what we expected, the older children outperformed the younger children but not significantly. This casts doubt on previous research claiming child L1 attrition as an inherent characterization of heritage language development. This does not seem to be the case, as far as the controlled production of tense and aspect morphology is concerned. In addition, our results do not support the claim that the difficulties heritage speakers have stem from the acquisition of an already attrited input (Rothman, 2007). Parents behaved at ceiling in their aspectual production, except with the use of the preterit with stative verbs where they showed some variability due to ambiguity in the intended interpretation. Thus, difficulties with tense and aspect cannot be attributed to the acquisition of an already attrited contact variety. We argue that the protracted development observed in our data might be explained in terms of crosslinguistic influence from English. Future research would benefit from a monolingual control group matched by age to see if monolingual children show similar patterns of performance, acquisition order and error types in comparison to the bilingual children. Although research in L1 acquisition shows target use of preterit versus imperfect morphology by age 4, it is possible that our task was too metalinguistically centered or cognitively complex for the younger children. Furthermore, future research should examine the extent to which the difficulties observed in this study are present at younger stages of development (3;0 to 4;0), and crucially, before the age of 6;0 when immersion in the school system usually starts. Finally, although cross-sectional methodology is valid and, by all means, reliable with participants from similar socio-cultural and linguistic background (as is the case in our study), future research would benefit from longitudinal analyses to completely disambiguate issues related to L1 attrition or incomplete acquisition.
References Argyri, E, & Sorace, A. (2007). Crosslinguistic influence and language dominance in older bilingual children. Bilingualism: Language and Cognition, 10(1), 77–99. DOI: 10.1017/s1366728906002835
The protracted acquisition of past tense aspectual values in child heritage Spanish
Au, T., Knightly, L, Jun, S., & Oh, J. (2002). Overhearing a language during childhood. Psychological Science, 13, 238–243. DOI: 10.1111/1467–9280.00444 Comrie, B. (1976). Aspect. Cambridge: Cambridge University Press. DOI: 10.1017/s0022226700007544 Crain, S., & Thornton, R. (1998). Investigations in Universal Grammar: A guide to experiments on the acquisition of syntax and semantics. Cambridge, MA: The MIT Press. DOI: 10.1017/s0022226701238651 Cuza, A. (2008). The L2 acquisition and L1 attrition of the interpretation and use of aspectual properties in Spanish among English-speaking L2 learners and long-term Spanish immigrants. Unpublished Doctoral Dissertation. University of Toronto. Cuza, A. (2010). The L2 acquisition of aspectual properties in Spanish. Canadian Journal of Linguistics, 55(2), 1001–1028. DOI: 10.1353/cjl.2010.0007 Cuza, A., Pérez-Tattam, R., Barajas, E., Miller, L., & Sadowski, C. (2013). The development of tense and aspect morphology in child and adult heritage Spanish: Implications for heritage language pedagogy. In J. W. Schwieter (Ed.), Innovative research and practices in second language acquisition and bilingualism (pp.193–220). Amsterdam: John Benjamins. DOI: 10.1075/lllt.38.12cuz Cuza, A., Miller L., & Pérez-Tattam, R. (2014). Bilingual success above and beyond the dominance shift: Evidence from the Spanish personal -a in child and adult heritage Spanish. Paper presented at the 44th Linguistic Symposium on Romance Languages. University of Western Ontario. May 2–4. de Miguel, E. (1992). El aspecto en la sintaxis del español: Perfectividad e impersonalidad. Madrid: Ediciones de la Universidad Autónoma de Madrid. Genesee, F., & Nicoladis, E. (2007). Bilingual first language acquisition. In E. Hoff & M. Shatz (Eds.), Handbook on language acquisition (pp. 324–343). Oxford: Blackwell. DOI: 10.1002/9780470757833.ch16 Hernández Pina, F. (1984). Teorías psicosociolingüísticas y su aplicación a la adquisición del español como lengua materna. Madrid: Siglo XXI. DOI: 10.2307/342474 Hulk, A., & Müller, N. (2000). Bilingual first language acquisition at the interface between syntax and pragmatics. Bilingualism: Language and Cognition, 3, 227–244. DOI: 10.1017/s1366728900000353 Jarvis, S., & Pavlenko, A. (2008). Crosslinguistic influence in language and cognition. New York, NY: Routledge. DOI: 10.1558/sols.v6i3.603 Kupisch, T. (2007). Determiners in bilingual German–Italian children: What they tell us about the relation between language influence and language dominance. Bilingualism: Language and Cognition, 10, 57–78. DOI:10.1017/S1366728906002823 Montrul, S. (2002). Incomplete acquisition and attrition of Spanish tense/aspect distinctions in adult bilinguals. Bilingualism: Language and Cognition, 5, 39–68. DOI: 10.1017/s1366728902000135 Montrul, S., & Slabakova, R. (2002). The L2 acquisition of morphosyntactic and semantic properties of the aspectual tenses preterit and imperfect. In A. T. Pérez-Leroux & J. Liceras (Eds.), The acquisition of Spanish morphosyntax (pp. 113–149). Dordrecht: Kluwer. DOI: 10.1007/978-94-010-0291-2_5 Montrul, S., & Slabakova, R. (2003). Competence similarities between native and near-native speakers: An investigation of the preterit/imperfect contrast in Spanish. Studies in Second Language Acquisition, 25, 351–398. DOI: 10.1017/s0272263103000159
Alejandro Cuza & Lauren Miller Montrul, S. (2008). Incomplete acquisition in Spanish heritage speakers: Chronological age or interface vulnerability? In H. Chan, H. Jacob, & E. Kapia (Eds.), Proceedings of the 32nd Annual Boston University Conference on Language Development, (pp. 299–310). Somerville, MA: Cascadilla Press. Montrul, S., Davidson, J., de la Fuente, I., & Foote, R. (2013). Early language experience facilitates gender agreement processing in Spanish heritage speakers. Bilingualism: Language and Cognition, 17(1), 118–138. (OnlineFirst). DOI: 10.1017/s1366728913000114 Montrul, S., & Sánchez-Walker, N. (2013). Differential object marking in child and adult S panish heritage speakers. Language Acquisition, 20(2), 109–132. DOI: 10.1080/10489223.2013.766741 Müller, N., & Hulk, A. (2001). Crosslinguistic influence in bilingual acquisition: Italian and French as recipient languages. Bilingualism: Language and Cognition, 4(1), 1–21. DOI: 10.1017/s1366728901000116 Odlin, T. (1989). Language transfer: Crosslinguistic influence in language learning. Cambridge: Cambridge University Press. DOI: 10.1177/026765839100700305 Paradis, J. (2001). Do bilingual two-year-olds have separate phonological systems? International Journal of Bilingualism, 5(1), 19–38. DOI: 10.1177/13670069010050010201 Pérez-Leroux, A.T., Pirvulescu, M. & Roberge, Y. (2008). Null objects in child language: Syntax and the lexicon. Lingua, 118, 370–398. DOI:10.1016/j.lingua.2007.07.002 Pérez-Leroux, A.T. Cuza, A., & Thomas, D. (2011). From parental attitudes to input conditions: A look at Spanish-English bilingual development in Toronto. In K. Potowski & J. Rothman (Eds.), Bilingual youth: Spanish in English-speaking societies. Amsterdam: John Benjamins. DOI: 10.1075/sibil.42.10per Pérez-Pereira, M. (1989). The acquisition of morphemes: Some evidence from Spanish. Journal of Psycholinguistic Research, 18, 289–312. DOI: 10.1007/bf01067038 Polinsky, M. (2011). Reanalysis in adult heritage language: A case for attrition. Studies in Second Language Acquisition, 33, 305–328. DOI: 10.1017/s027226311000077x Putnam, M., & Sánchez, L. (2013). What’s so incomplete about incomplete acquisition? A prolegomenon to modeling heritage language grammars. Linguistic Approaches to Bilingualism, 3(4), 476–506. DOI: 10.1075/lab.3.4.04put Rothman, J. (2007). Heritage speaker competence differences, language change and input type: inflected infinitives in heritage Brazilian Portuguese. International Journal of Bilingualism, 11(4), 359–389. DOI: 10.1177/13670069070110040201 Salaberry, R. (1999). The development of past tense verbal morphology in classroom L2 Spanish. Applied Linguistics, 20, 151–178. DOI: 10.1093/applin/20.2.151 Salaberry, R., & Shirai, Y. (2002). The L2 acquisition of tense-aspect morphology. Amsterdam: John Benjamins. DOI: 10.1075/lald.27.04sal Schmitt, C. (1996). Aspect and the syntax of noun phrases. College Park, MD: University of Maryland dissertation. Sebastián, E., & Slobin, D. (1994). Development of linguistic forms: Spanish. In R.A. Berman & D. I. Slobin (Eds.), Relating events in narrative: A crosslinguistic developmental study (pp. 239–284). Hillsdale, NJ: Lawrence Erlbaum Associates. DOI: 10.1017/s0272263100014443 Silva-Corvalán, C. (2014). Bilingual language acquisition: Spanish and English in the first six years. Cambridge: Cambridge University Press. DOI: 10.1017/cbo9781139162531
The protracted acquisition of past tense aspectual values in child heritage Spanish
Silva-Corvalán, C. (2003). Linguistic consequences of reduced input in bilingual first language acquisition. In S. Montrul & F. Ordóñez (Eds.), Linguistic Theory and Language Development in Hispanic Languages (pp. 375–397). Somerville, MA: Cascadilla Press. Silva-Corvalán, C. (1994). Language contact and change. Oxford: Oxford University Press. DOI: 10.1017/s0047404500020753 Smith, C. (1997). The parameter of aspect. Dordrecht: Kluwer. Vendler, Z. (1967). Verbs and times. Philosophical Review, 56, 143–160. DOI: 10.2307/2182371 Verkuyl, H.J. (1972). On the compositional nature of the aspects. Dordrecht: Foris. DOI: 10.1007/978-94-017-2478-4 Yip, V., & Matthews, S. (2000). Syntactic transfer in a Cantonese-English bilingual child. Bilingualism: Language and Cognition, 3(3), 193–208. DOI: 10.1017/s136672890000033x Yip, V. & Matthews, S. (2006). Assessing language dominance in bilingual acquisition: A case for mean length utterance differentials. Language Assessment Quarterly, 3, 97–116. DOI: 10.1207/s15434311laq0302_2
part iii
Language contact and language variation
Morphological adjectival intensifier variation in Lima, Peru Álvaro Cerrón-Palomino Arizona State University
This variationist study is the first addressing the alternation of adjectival intensifiers -ísimo (standard) and -azo (non-standard) in Spanish, concretely, in the Limeño variety of Peru.Using online digitalized corpora, online journals, magazines, and e-networks, I conducted a multivariate analysis of this alternation. The results show that language use in colloquial texts of e-networks favor the use of -azo, as well as male speakers, which suggests covert prestige. Semantically negative adjectives, commonly used in colloquial texts, also favor the non-standard variant. Interestingly, stative/resultative constructions of the form estar + participle with an experiencer argument promote the use of -azo, suggesting that it is more specialized than -ísimo, which has a broader distribution. Keywords: intensifier; variation; style; e-networks; Peru
1. Introduction Intensifier alternation is ideal for the study of language variation change mainly due to three features they possess: (a) they are versatile and numerous, (b) they can change quickly, and (3) they may entail recycling of different forms (Tagliamonte, 2011: 320). These characteristics entail that the coexistence of competing variants may be the outcome of old and new layers in the process of change. In addition, fast-changing forms may be very sensitive to social evaluation. This variationist study focuses on the alternation between the morphological intensifiers -ísimo and -azo affecting qualifying adjectives in the Spanish of Lima, Peru, as (1) and (2) show.1
. The forms -ísimo and -azo are only the masculine singular forms, but this study covers all the instances of gender and number inflection.
doi 10.1075/ihll.4.12cer © 2015 John Benjamins Publishing Company
Álvaro Cerrón-Palomino
(1) ¡Las dos están riquísimas! Pero como es encuesta, voto por Yamila, pero ahí no más con las justas gana, porque la uruguaya también está buenaza. (Forosperu.net forum, 06/18/2013) ‘Both of them are very hot! But since it is a poll, my vote is for Yamila, but she wins by a nose, because the Uruguayan (girl) is really hot, too.’ (2) Qué stress no poder manejarlo otra vez. Por lo menos, saldrá e ngreidísimo, porque le acabo de conseguir un timón Momo bravazo como para compensarle haberlo tenido lejos tantos días… (Todoautos.pe forum, 12/04/2008) ‘How stressful [it is] not to drive it again. Nonetheless, it’ll come out very spoiled, because I have just gotten a really cool Momo steering wheel, to compensate for having been away from it for so many days…’ On the one hand, -ísimo is used in all Spanish varieties, and it is considered a standard superlative form in Spanish grammars (Alcina & Blecua, 1991; Alarcos, 1994).2 On the other hand, -azo is a traditional augmentative suffix used with nouns in most varieties, and its function as an adjectival suffix is not standard in the dialects in which it is used, with no mention of such a usage appearing in descriptive grammars. This study is organized as follows: first, I address the nature of the variants under investigation and their historical development; then I present the data and methodology used in this study; next I introduce the factors analyzed in this paper; afterward I show the results and discuss them; and finally, I offer the conclusions to this article.
2. A historical overview of the two variants Traditionally, -ísimo has been considered a suffix signaling the absolute superlative or elative, which is used to express the quality of the adjective at its highest or most intense degree (Alarcos, 1994: 85). For instance, in (3) the quality of the adjective antipática is exaggerated by the suffix at issue. (3) Rosario será antipatiquísima pero es inocente, así de simple. (Peru 21 online news comments, 09/27/2013) ‘Rosario may be very unpleasant, but she is innocent; it’s that simple.’ However, this suffix cannot be used in comparative structures, in which case the relative superlative structure, i.e. ‘más/menos + positive adjective + de’ is used, as shown in (4).
. Interestingly, Bello (1981 [1847]: 228–229) considers superlatives in -ísimo as instances of the more general morphological category of augmentatives.
Morphological adjectival intensifier variation in Lima, Peru
(4) La esposa del ex presidente Alejandro Toledo, Eliane Karp, recibió el nada honroso título de ser el personaje más antipático del Perú. (Correo online newspaper article, 02/23/2011) ‘Ex-President Alejandro Toledo’s wife, Eliane Karp, was awarded the non-honorable title of being the most unpleasant character of Peru’. Following this rationale, a number of studies (Porto Dapena, 1973; Cerrón-Palomino, 2003; Montoro & Almela, 2008) have questioned the characterization of -ísimo as a superlative or a comparison degree suffix. Instead, there seems to be consensus in considering it an appreciative suffix, in particular, an intensifier. The original categorization of suffix -ísimo as a superlative can be traced back to its history: it was derived from the Latin superlative suffix -issimus, which functioned both as an absolute superlative and a relative superlative.3 This suffix disappeared in late Latin and the absolute superlative meaning was replaced by a number of syntactic forms, out of which ‘multum + positive adjective’ became widely used in medieval Spanish, giving birth to the current analytic alternative ‘muy + positive adjective’, present in every contemporary Spanish dialect. However, as Martinell (1989: 1256) points out, -ísimo was reincorporated into Spanish around the 15th Century, although its use was more prominent in the 16th Century. Nowadays, however, -ísimo is used in every Spanish variety, and it is not only restricted to written registers as it was originally. By the same token, the second variant in this study, -azo, traditionally categorized as an augmentative, has also undergone a process of recycling, though its path has been quite different. Originally characterizing adjectives, -azo traces its origin back to the Latin derivative suffix -aceus, attested in derivations like rosa (rose) > rosaceus (rose-like). As stated above, -azo is widely used as an appreciative suffix for nouns in most contemporary Spanish varieties, as shown in (5). (5) Neri Cardozo se dio un golpazo contra cartel de publicidad. (El Comercio online newspaper, 01/31/2011) ‘Neri Cardozo suffered a big blow against an advertising panel.’ It has been pointed out that the unorthodox use of -azo as an adjectival intensifier is more common in Hispanic America than in Spain (Falcinelli, 2007: 24). Moreover, Kany (1969: 73) found this use to be more extended in South America, in particular, in the rural areas of Río de la Plata and Chile.
. A more detailed account about the Latin suffix -issimus can be found in Cerrón-Palomino (2003: 270–271).
Álvaro Cerrón-Palomino
However, this variant is not exclusive of these South American regions. One of the earliest examples of this non-traditional use of -azo in Peruvian Spanish belongs to the 19th Century writer Clorinda Matto de Turner, in her 1889 novel Aves Sin Nido (Birds without a Nest), shown in (6).4 (6) ¡Qué traguito tan confortable, francamente, que es… buenazo! (Matto, 1948 [1889]:80) ‘What a refreshing drink, it is, frankly… really good!’ Nowadays, the use of -azo as an adjective suffix has extended considerably in Peruvian Spanish, concretely, in the Lima variety, which serves as a model to the rest of the varieties in the country. Surprisingly, the only Hispanic American sociolinguistic study that includes -azo as one of its variants is Peñalillo’s (2012) account of diverse types of forms that exhibit hyperbolic, metaphoric, reiterative, and comparative effects similar to that of a superlative adjective or adverb in the Spanish of Valdivia, Chile. Given this wide range of forms investigated, only a few findings are relevant to the present study: (1) -ísimo represents only 9% of the total superlative-like forms (for adjectives and adverbs alike) registered in 34 sociolinguistic interviews; (2) overall, women use more superlative forms than men; (3) men use more -ísimo forms than women; and (4) superlatives occur more with predicative adjectives (with verbs ser and estar ‘to be’) than with attributive adjectives. In light of this general lack of attention on the intensifier -azo, the research question motivating this study is the following: what factors, linguistic and extra-linguistic, are responsible for the expansion of -azo over -ísimo as an adjectival intensifier in the Spanish of Lima, Peru?
3. Data and methodology Unlike other morphological variables, the two variants analyzed in this paper have an extremely low rate of occurrence in sociolinguistic interviews, making it impossible to conduct an orthodox quantitative sociolinguistics analysis about them in Limeño Spanish. As a result, online data were used in this study. I prepared a list of 128 qualifying adjectives of common usage, including both lexical items belonging to the general Spanish lexicon and lexical items typical of the Spanish variety of Lima, Peru. The online data were extracted from two main types of sources: on the one hand, online corpora were used, mainly written data from magazines, journals and books
. Although she was from Cuzco, Matto de Turner moved to Lima and published her books in Lima.
Morphological adjectival intensifier variation in Lima, Peru
that were digitalized and included in the CREA and CORDE corpora.5 On the other hand, data from the most popular e-networks, such as Facebook, Twitter, and YouTube, and also Peruvian forums, blogs, online journals were used. It is important to point out that only data corresponding to the Limeño variety were used. Since CREA and CORDE have their own search engines, the searches in their databases were straightforward. In contrast, searching e-networks proved to be more challenging. Besides Facebook, Twitter and Youtube, I had access to a list of Peruvian forums, blogs, and online journals. However, none of these e-networks had search engines to track the occurrences of adjectives containing the suffixes under study. Therefore, I used the general Google search engine, following each link provided to identify the particular lexical occurrence in the text. As stated before, this study–although unorthodox in the source of the data employed–follows the variationist methodology (Labov, 1972). The data, after being coded, were entered in the multivariate statistical program Goldvarb X.
3.1 Envelope of variation: Exclusions First and foremost, only adjectives with the morphological intensifiers -azo and -ísimo were analyzed in this study. The reason for this decision was that the identity of the referential meaning can only be guaranteed for the two variants at issue, because syntactic intensifiers such as muy (very), bien (really), tan (so), extremadamente (extremely), demasiado (too) usually add a different connotation (Vidal Lamíquiz, 1991; Real Academia Española, 2010; Porto Dapena, 1985; Bosque & Demonte, 1999). However, not all instances of the two variants were included in the analysis, and the exceptions are presented below. 1. Adjectives acting as nouns, even if they have intensifier suffixes The standard use of the variant -azo is as a noun suffix, and its non-standard use is with adjective stems. In order to guarantee the adjectival nature of the tokens, I decided against including adjective stems affected by this morpheme when they were acting as heads of the noun phrase, as in (7) and (8), because in these cases, the -azo suffix is not an innovative form, but rather a standard noun intensifier, i.e. a different linguistic variable. (7) Algunos vivazos dicen que sólo castiga a las industrias, o sea justo lo que se debería estimular.(Voz Liberal del Perú Blog, 05/21/2013) ‘Some real wise-@sses say that it only hurts the industry, which is precisely what should be stimulated.’
. CREA is the Corpus de Referencia del Español Actual (Reference Corpus of Contemporary Spanish), and CORDE is the Corpus Diacrónico del Español (Spanish Diachronic Corpus). The data from the latter was restricted to the period from the two last decades of the 20th Century to the 21st Century.
Álvaro Cerrón-Palomino
(8) Yo sí recuerdo cuando el mentirosazo de Aguinaga decía que Fuji tenía cáncer de alto riesgo, y ahora sale que no, está limpio. (La República online newspaper, comments, 03/28/2013) ‘I do remember when that big liar of Aguinaga said that Fujimori had a high-risk cancer, and now it turns out that he didn’t, he’s clean.’ 2. Adverb-like constructions By the same token, cases where the function of the structure containing the suffix -azo was not clearly adjectival, but rather of an adverbial nature as in (9), were excluded from the analysis.6 (9) Pero normalazo pueden ir al Terrazas, aún tengo mis influencias por ahí si es que se les ocurre a los organizadores cobrar la entrada para el torneo, aunque no creo.(AT Perú forum, 08/12/2007) ‘But very typically they can go to the Terrazas [club]; I can still pull strings there if the organizers consider charging for the tickets to the tournament, although I doubt it.’ 3. Determiners with intensifier suffixes Finally, determiners with the suffix -ísimo were not included in the study because there is lack of variation in such contexts: they do not allow the -azo suffix (as in (10)–(11)). (10) ¡Tu Dios salve tu alma porque irás al mismísimo infierno! (Real Academia Española: CREA online. 〈http://www.rae.es〉 [09/24/2013]) ‘May your God save your soul, because you will go to the very same hell!’ (11) Perú tiene un ataque de primerísimo nivel. (Net Joven online newspaper article, 11/10/2011) ‘Peru has a [soccer] attack of the highest level. In sum, this study is restricted to qualifying adjectives with a clear modifying function, and any diverging contexts were excluded.
4. Factors analyzed in the present study As stated in Section 2, the only known sociolinguistic study on superlatives/intensifiers in Spanish (Peñalillo, 2012) is not only not restricted to adjectives, but it also comprises forms that are not sensu stricto intensifiers. . A study of - azo or -ísimo intensifying adverbs requires a different analysis. For instance, most modal adverbs allow the alternation in their shortened forms, such as rapidísimo/rapidazo (very fast), but only allow–ísimo in their full forms: rapidísimamente/*rapidazamente. Other adverbs only allow -azo: atrasazo/*atrasísimo (way back).
Morphological adjectival intensifier variation in Lima, Peru
As a consequence, I based most of the linguistic hypotheses for the variable studied in this paper on Ito & Tagliamonte (2003), Tagliamonte (2008) and Brown & Tagliamonte (2012), which are studies on syntactic English intensifiers in different English speech communities. Factors presented in 4.1.1–4.1.3 are adapted from the aforementioned studies; factor 4.1.4 is based on the qualitative observations of Bello (1981 [1847]) and Peñalillo (2012); and factor 4.1.5 is new. Regarding the extra-linguistic factors, I analyzed gender and the type of text the occurrences appeared in.
4.1 Linguistic factors Five linguistic factors of morphological, syntactic, semantic, and pragmatic nature were analyzed in this study.
4.1.1 Adjective function Ito & Tagliamonte (2003) and Tagliamonte (2008) found that predicative adjectives favored the use of the syntactic intensifier really over very, so and pretty, which, according to these authors, shows that it is a variant in a well-advanced stage of use. Peñalillo (2012) also found that in Spanish, predicative adjectives favor intensification more than attributive ones. In view of these facts, the hypothesis underlying this factor group is that if predicative contexts like (12) contribute to the use of variant -azo, this would imply that it is not an initial stage of an innovative form, but rather a somewhat well-established adjectival suffix. On the contrary, if it is favored by attributive contexts as in (13), this would suggest a relatively innovative use. (12) Miren la cara de ese tombo: está asustadazo. La bronca que me da. (YouTube, 10/2010) ‘Look at that cop’s face: he is totally scared. It makes me so angry.’ (13) Yo tenía un perro chuscazo; mejor dicho, tuve dos, padre e hijo, y a los dos los llamé “Pipo”. (Caín Subte blog, 05/01/2008) ‘I had a really mixed-breed dog; better put, I had two, father and son, and I named both “Pipo”.’
4.1.2 Semantic classification of the adjectives Brown & Tagliamonte (2012) had found that adjectives whose semantic content indicated human propensity favored the overall occurrence of intensifiers in Toronto, while adjectives conveying value and physical property meanings disfavored it. In the present study, no prediction was made about the semantic type of adjective that would promote the use of -azo, but rather the factor group was used in an exploratory way. To this end, I classified the adjectives as human propensity as in (14), value as in (15), and physical property as in (16).
Álvaro Cerrón-Palomino
(14) ¡Es cierto que Gaby a esta edad todavía estaba crudaza! Pero su evolución fue tal que los resultados se vieron tiempo después. (YouTube, December 2010) ‘It is true that Gaby was still really immature at that age! But her evolution was such that we saw the results a while after.’ (15) Quiero que mi viejita me prepare causa de atún, le sale brutalazo. (Facebook, 11/22/2012) ‘I want my mom to make me tuna causa; she prepares a really great one.’ (16) Yo me fui directito a la librería a comprar Un Mundo para Julius, pero como es un libro tan gruesazo me aburrí y lo dejé. (La tertulia es en mi casa blog, 04/01/2010) ‘I went straight to the bookstore to buy A World for Julius, but since it is such an incredibly thick book, I got bored and did not get it’.
4.1.3 Emotional value in the adjective Ito & Tagliamonte (2003) found that, when including ‘emotion’ as a semantic category, it favored the occurrence of the intensifier really more than very. However, Brown & Tagliamonte (2012) used emotionality as a separate factor group, and they discovered that the presence of emotionality in the utterance containing the adjective favored the overall occurrence of intensifiers, whereas the absence of emotionality disfavored it. In this study, I hypothesized that, since -azo is the variant of non-educated origin, utterances with an emotional content like (17) would contribute to its occurrence more than emotion-free ones like (18), in accordance with its colloquial origins. (17) Así que caballero nomás, salí, asustadazo, creí que me iban a matar, joven… (Amigos de Villa online stories, 2005) ‘So, just like that, I left really scared, I thought they were going to kill me, young man…’ (18) Con la billetera gruesaza podrá pagar varias arrugas y además contratarle más jugadores a Bam Bam. (Líbero online newspaper article, 02/01/2013) ‘With a really thick wallet he will be able to pay many debts, and also hire more players for Bam Bam [Wilmar Valencia, Alianza Lima F.C.’s coach].’
4.1.4 Positive versus negative evaluation of the property of the adjective Bello (1981 [1847]:225) has pointed out that augmentatives (including -azo) are frequently used to convey the idea of ugliness or hoarseness and frivolity or scorn, i.e. negative semantic values like gigantazo (too much of a giant). By the same token, Peñalillo (2012: 197) found some cases of augmentatives used to intensify adjectives with undesirable qualities, such as flojaza (really lazy). I operationalized the negative/ positive opposition to the property described by the adjective as follows: positive-like (19), negative-like (20) and neutral-like (21).
Morphological adjectival intensifier variation in Lima, Peru
(19) Está buenazo, mas vamos a ver con qué lo acompañarás. (Facebook, 10/24/ 2012) ‘It is really good, but now let’s see what you are going to eat it with.’ (20) Ahora Jackson japonesa es muy buena, y bueno B.C. Rich no pasa nada: son malazos, nada definidos y balanceados. (Facebook, 12/25/2012) ‘Now, Japanese Jackson [basses] are very good, and well, B.C. Rich is not doing well: they are really bad, not defined or balanced.’ (21) Ya van casi 4 años que no voy al Jockey; de hecho, lo encontraré cambiadazo. (Skyscraper City forum, 03/17/2012) ‘It’s been almost 4 years since I’ve been to the Jockey [Plaza Mall]; I’m sure I will find it extremely changed.’ The prediction for this factor group was that adjectives with negative semantic content would contribute to the presence of suffix -azo, in view of previous work.
4.1.5 Morphological origin of the adjective This factor group divides adjectives according to whether or not they were originated as a participle. This is the first time that this factor group has been proposed, and it was motivated by a prima facie perception of the distribution of -azo. I hypothesized that this variant will be favored by adjectives with a participial origin such as (22), as opposed to cases like (23), where the adjective is not participial. (22) Hay sumas que restan, y Toledo está quemadazo. (Trome online newspaper, 10/31/2013) ‘There are additions that subtract, and [ex-president Toledo] is totally burned out’. (23) Temazo tonerazo tanto en italiano como en español. (YouTube, 03/2011) ‘Really great party song, in Italian as well as in Spanish’.
4.2 Extra-linguistic factors There were only two extra-linguistic factor groups used in this study, given the nature of the data: the gender of the speaker/writer, and the type of the text which contained the superlative suffix occurrence.
4.2.1 Gender In most of the examples used, the biological gender of the author of the token could be retrieved. However, there were some cases in which this was not possible, mainly in forums where some members did not have complete/updated profiles. The texts of the gender-identifiable ones were coded as male or female, while the ones whose authors’ gender could not be retrieved were assigned a separate coding. The latter were excluded from the factor group in the multivariate analysis.
Álvaro Cerrón-Palomino
In general, the use of intensifiers has been associated, for the most part, with women, and it is assumed that women are the leading ones in the use of novel intensifiers (Tagliamonte, 2011: 321). However, since the use of -azo is not prestigious, it was expected that women would not favor its use, as opposed to men, since the former tend to use non-prestigious forms considerably less frequently than the latter (Labov, 2001: 264–267).
4.2.2 Text type This factor group was not divided according to the source of the data, i.e. online corpora versus online searches. Rather, the texts were classified as follows: those found in books, the ones found in newspapers, the ones obtained from magazines (digitalized or online), and the ones present in e-networks such as YouTube, Twitter, Facebook or various forums, including the comments section of online newspapers.7 Given that the use of intensifiers has been associated with colloquial styles and non-standard varieties (Tagliamonte, 2011: 321), the prediction for this factor group is that, on the one hand, the non-standard suffix -azo would be promoted by more orallike media, like e-networks. On the other hand, -ísimo, being the suffix with literary origins, should be favored by books, newspapers and magazines, despite the general scarcity of intensifiers in formal styles. As stated in Section 3, the data were coded and then entered in the statistical program Goldvarb X for a multivariate factor analysis. The results obtained are presented in the following section. 5. Results and discussion As Table 1 shows, the multivariate analysis indicates that e-networks, male speakers, a negative evaluation of the adjective’s property, adjectives with a predicative function, and adjectives of participial origin are, in that order, the factors favoring the occurrence of the innovative variant. The factors eliminated in the multivariate analysis (presented in square brackets) were the presence/absence of emotionality and the semantic classification of the adjectives. The factor group with the greatest probabilistic weight was that of the text type, which suggests that the two variants are stylistically stratified. The very scarce presence of -azo in books and newspapers, as well as its abundant occurrence in e-networks underscores its colloquial nature.
. Tokens from articles in online journals were coded as newspapers, whereas tokens from the readers’ comments sections of online journals were coded as e-networks.
Morphological adjectival intensifier variation in Lima, Peru
Table 1. Variable Rule Analysis of the factors contributing the occurrence of the intensifier suffix -azo in the Spanish of Lima, Peru Corrected mean
0.591
Log likelihood
493.960
Significance
0.008
Total N
1102 Factor weight
%
N
E-networks
.70
75.8
811
Newspapers
.35
40.6
64
Magazines
.33
40
25
Books
.03
4.1
171
Range
67
Text type
Gender Male
.59
62.8
731
Female
.28
51.5
268
Range
31
Evaluation of the adjective’s property Negative
.58
62.5
426
Positive
.44
69.5
555
Neutral
.35
40.5
121
Range
23
Syntactic function of the adjective Predicative
.57
69.5
711
Attributive
.38
41.9
391
Range
19
Original category of the adjective Participle
.58
75.1
358
Non participle
.46
52.3
744
Range
12
Semantic classification of the adjective Value
[.52]
53.6
386
Human Propensity
[.51]
71.4
518
Physical property
[.43]
40.9
198
Emotional
[.58]
74.8
242
Non emotional
[.49]
55.5
860
Emotional value of the adjective
Álvaro Cerrón-Palomino
Bello (1981 [1847]:225) had already stated that -azo, along with the rest of the Spanish morphological augmentatives, did not belong to formal registers, and the results of the present study corroborate this tendency. Along the same lines, the almost exclusive occurrence of the traditional variant -ísimo in books and printed newspapers (mostly in the online corpora) and its restricted appearance in e-networks stresses its formal character. Moreover, it is a well-known fact that colloquial Spanish abounds in exaggeration and hyperbole (García, 1999). Along these lines, 15.3% of the non-standard variant’s occurrences were emphasized by another term, like in (24). (24) Eres recontra hermosa y hablas bonito, qué suertudazo el que llegue a ser tu novio o enamorado.(Foros Perú forum, 09/14/2010) ‘You are extremely beautiful and well-spoken, what a very lucky guy, the one who becomes your boyfriend or fiancé.’ In contrast with the emphatic trend affecting -azo, only 0.5% of the standard form’s occurrences were intensified by another word, as in (25). (25) Estaba realmente bellísima, en su vaporoso vestido blanco. (Real Academia Española: CREA online. 〈http://www.rae.es〉 [09/20/2013]) ‘She was really very pretty in her steamy white dress.’ In addition, some of the adjectives intensified by -azo were used in their overtly colloquial meaning, like asadazo (originally, ‘really roasted’) in (26), and quemadazo (originally ‘really burnt’) in (27), which would never be employed in an even slightly formal register. (26) La Pulga está asadazo y jura que el Barcelona alcanzará al Real Madrid y no se rendirán para ganar la liga. (Deportes Perú online newspaper article, 01/30/2012) ‘The Flea (soccer player Lionel Messi) is really mad and swears that Barcelona will catch up with Real Madrid and they won’t give up winning the league.’ (27) Ese sí era trauma. Total, ese pata estaba quemadazo. (YouTube, 12/2007) ‘That was in fact a trauma. Anyway, that dude was really burnt out.’ Another revealing fact, related to -azo’s colloquial nature, is the contributing factor with the second strongest probability: that men favor its use considerably more than women. I had hypothesized in 4.2.1 that women should use the more educated variant -ísimo, due to their known preference for the more prestigious forms, and this study’s results corroborate this tendency. On the other hand, men’s overwhelming use of the non-standard variant could be a case of covert prestige, namely, the adherence, generally by working class men, to non-standard linguistic variants (Trudgill, 1972), although no information about the social class of these men was obtained due to the
Morphological adjectival intensifier variation in Lima, Peru
nature of the data collection. With regard to this last point, Tagliamonte (2011: 321) has also pointed out that “the use of particular intensifiers is said to signal-in group membership”. The results suggest that the males’ preference for -azo is a demonstration of masculinity in contrast with the standard variant. The third factor contributing to the use of -azo is that of adjectives with a negative semantic content, in accordance with what Bello (1981 [1847]) and Peñalillo (2012) have indicated. In this particular case, it seems that the colloquial nature of this variant leads speakers to use it more when they want to intensify adjectives of whose content they disapprove. For instance, in (28), the commentator uses the suffix to boost the property, attributed to a contestant she dislikes, of being angry for not being on top instead of being happy for not being eliminated from the TV contest. The use of the more refined variant -ísimo would have conveyed a weaker judgment about the contestant’s attitude. (28) No me cae la tal Ariadna: se cree la gran cosa. Y cuando pasó por la pasarela se notaba que estaba asadaza en vez de estar contenta por no quedar sentenciada. (YouTube, 12/2012) ‘I don’t like that Ariadna: she thinks she is the next great thing. And when she walked on the runway you could tell she was really mad instead of being happy for not being eliminated [from the TV contest].’ By the same token, in (29), the commenter uses the -azo suffix twice, in order to reinforce the acrimonious adjectives pasado (drunk) and estón (stoned) used in reference to a professional soccer player. The same effect would not have been achieved by using the literary suffix -ísimo, where the meaning would have been less severe. (29) Ese Vargas está pasadazo, está recontra estonazo, se metió de la mala; que boten esa huevada y nunca más lo convoquen. (YouTube, 12/2012) ‘That Vargas is really drunk, he is so very stoned, he smoked the bad one; they should kick this trash out [of the national soccer team] and they should never call him back [to the team] again.’ In general, it is well-known that in informal/colloquial contexts, irritation can cause speakers to produce non-standard forms in addition to swearing. It seems that adjectives intensified by -azo are the preferred variant when speakers decide to express a negative feeling in a colloquial situation. The fourth and fifth factors favoring -azo are, respectively, predicative structures and adjectives of participial origin. An explanation for the latter is related to the type of participle that most favors the non-standard variant. In fact, all of the participles that contribute to -azo are derived from verbs with an experiencer argument, in structures like (30), that Varela (1989: 229) calls middle constructions. No instances of passive participles with the innovative intensifier were attested in the data, unlike passive participles with the standard intensifier, as in (31).
Álvaro Cerrón-Palomino
(30) El pata estaba asustadazo y yo solo quería demostrarle que no tenía nada que ocultar. (YouTube, 06/2012) ‘The dude was really scared and I just wanted to show him that he had nothing to hide.’ (31) Les cuento que la semana pasada estuve en el cóctel que realizó la marca Moda & Cía. en su tienda de Larcomar y donde muchas prendas estuvieron rebajadísimas.(El Closet de Giuliana Blog, 2013/07/29) ‘Let me tell you that last week I was at a cocktail [reception] that the brand Moda & Cía organized in its Larcomar [Mall] store, where many clothes were steeply discounted.’ It seems clear that it is not the mere category of participles that favors the occurrence of -azo, but rather participles with an experiencer argument, which, according to Varela (1989: 224), possess a delimited meaning, i.e. a perfective aspect. Along the same lines, Sánchez, Marín & Evert (2012) suggest that the estar + participle structure in modern Spanish has been extended to describe the result state of a process involving a particular entity, whereas ser + participle has become restricted to eventive passive structures. This development is probably related to the perfective nature of estar, which expresses the result of a process, as opposed to the imperfective aspect of ser, which expresses a durative process (Roca Pons, 1958: 268). According to Sánchez, Marín & Evert (2012), the context propitious for the extension of estar is the one involving participles of verbs with an experiencer. Coincidentally, this type of participles favors -azo over -ísimo in this study’s data, as seen in (32)–(33). (32) Estaba interesadazo en tu equipo, pero leí que este celular tiene problemas para conectarse a redes sociales. (ForosPeru.net forum, 09/05/2012) ‘I was really interested in your equipment, but I read that this cell phone has problems when connecting to social networks.’ (33) Si Erick Delgado está confiadazo en hacerla linda en la final, no es por las puras.(Depor.pe online journal, 12/09/2012) ‘If Erick Delgado is really confident in succeeding at the finale, it’s not in vain.’ In order to investigate whether participles intensified by -azo are favored by stative/ resultative structures or by eventive/resultative structures, I analyzed the subset of predicative contexts where the two suffixes intensify participial and non-participial adjectives, examining separately instances with estar and instances with ser. Table 2 corresponds to the occurences of the two variants by participial/non- participial origin with the verb estar. The first striking fact is that 65.9% of the adjectives used with estar are of participial origin, whereas 34.1% are not. As we can see, in both variants there is a strong participial effect. However, the difference between the percentages for -azo (69.5% versus 20.5%) is greater than the one for–ísimo (57.1% versus 42.9). It seems clear, then, that the structure estar + participial adjective as in (34)–(35) is a context strongly favoring the presence of -azo. This would suggest,
Morphological adjectival intensifier variation in Lima, Peru
Table 2. Distribution of the variants with the verb estar according to the origin of the adjective Non-participial -azo -ísimo Total
Participial
Total
N
46
105
151
%
30.5%
69.5%
100%
N
27
36
63
%
42.9%
57.1%
100%
N
73
141
214
%
34.1%
65.9%
100%
following Sánchez, Marín & Evert (2012), that the suffix -azo is slowly becoming specialized in intensifying participles in stative/resultative structures, with a perfective aspect. (34) Ahí estaré de todas maneras, ¡solo pásenme un correito y de todas estoy apuntadaza!(Solo una Alejandra blog, comments section, 08/12/2013) ‘I will be there, at any cost, just send me an email, and for sure I am definitely scheduled [to go].’ (35) ¡Oye ya pues, man, coopera y mándame la foto! Estoy angustiadazo por esa foto. (Fotolog network, 11/17/2006) ‘Come on, man, cooperate and send me the picture! I am desperate for that picture.’ In contrast, by looking at Table 3, one can see that only 14.2% of the adjectives used with ser are of participial origin, as opposed to 85.8% of adjectives that are not participles. Although both variants are favored by non-participial adjectives, the effect is stronger for -ísimo (98.4% versus 1.6%), with an almost categorical preference, than for -azo (80.4% versus 19.6%). Thus, the construction ser + non-participial adjective is a context that favors the occurrence of -ísimo. Table 3. Distribution of the variants with the verb ser according to the origin of the adjective
-azo -ísimo Total
Non-participial
Participial
Total
N
115
28
143
%
80.4%
19.6%
100%
N
60
1
61
%
98.4%
1.6%
N
175
29
204
%
85.8%
14.2%
100%
Álvaro Cerrón-Palomino
In sum, the structure estar + participial adjective favors -azo, whereas the structure ser + non-participial adjective favors -ísimo. The fact that estar + participle structures favor -azo suggests that this variant is becoming specialized in intensifying stative (experiencer) participles, with a perfective aspect; in contrast, -ísimo has a broader distribution, including eventive (passive/agent) participles, with an imperfective aspect, as well as non-participles. I return now to the fourth factor favoring the use of -azo: structures including predicative adjectives. As can be seen in Table 1, there is a total of 711 (64.5 %) occurrences of intensifiers with predicative adjectives, and 391 (35.5%) with attributive adjectives. This suggests that, overall, adjectival intensification is favored by predicative adjectives, which goes along with the findings in Brown & Tagliamonte (2012) for English and Peñalillo (2012) for Chilean Spanish. According to Ito & Tagliamonte (2003), when an intensifier occurs abundantly with predicative adjectives, this implies that it is at an advanced stage of diffusion, and not an initial one.8 The fact that there are written records of -azo as early as the 19th Century in Peruvian Spanish supports this idea. However, it seems clear that this variant, although relatively established in Limeño Spanish, is for the most part still restricted to contexts that are overtly colloquial, such as e-networks, but it is almost non-existent in contexts with even a slightly more formal tone, like a sociolinguistic interview.
6. Conclusions In accordance with the findings in Ito & Tagliamonte (2003), Tagliamonte (2008), and Brown & Tagliamonte (2012) for different dialects of English, this paper has shown that intensifiers in Spanish, are also varied and recyclable. On the one hand, an intensifier lost early on and re-introduced in the 15th Century like -ísimo is still robust although favored in formal styles; on the other hand, an adjectival derivative suffix, which was re-interpreted as a noun augmentative like -azo has made its way back into adjectival morphology, although with a different function (that of an intensifier). In addition, the results of this study support Tagliamonte’s (2011) statement about the importance of intensifier variation studies in sociolinguistics because of the interaction between both extra-linguistic and linguistic factors underlying such alternation. Accordingly, the strongest factors predicting the use of -azo are social: texts belonging to colloquial styles and male speakers. These results confirm trends widely
. As mentioned in Section 4, Ito & Tagliamonte (2003) found this to be true for the English syntactic intensifiers very and really.
Morphological adjectival intensifier variation in Lima, Peru
attested in the variationist literature: on the one hand, innovative forms appear in styles where less attention is paid to language use; on the other hand, females mostly favor variants with overt prestige, whereas men favor variants with covert prestige. The third strongest predictor of -azo is a semantic one: adjectives with a negative value. This factor seems to pattern well with the two social factors mentioned above. In fact, a colloquial/informal style better fits the use of adjectives with undesirable qualities, and the combination of somewhat acrimonious adjectives and informal suffixes fits male covert prestige tendencies. Showing less probabilistic strength there are two morphosyntactic factors that contribute to the occurrence of -azo: predicative structures and participial adjectives. The occurrence of -azo in predicative structures suggests, following Tagliamonte (2003, 2008), and Brown & Tagliamonte (2012), that this variant is already well-established in the variety’s system, despite its almost exclusively colloquial use. In addition, there seems to be a close relationship between the two morphosyntactic factors. In particular, participial adjectives in combination with a predicate including the verb estar favor the use of the non-standard variant. In accordance with the findings of Sánchez, Marín & Evert (2012), this suggests that the innovative intensifier is on its way to specializing as a stative/resultative suffix, whereas the standard variant has a broader distribution. Finally, the fact that -azo is not found intensifying participles with a passive theme agent, but rather only with participles whose verbs take an experiencer argument as a subject implies that this suffix is becoming the preferred choice in predicative structures with a perfective aspect. In conclusion, the emergence of -azo as a competitor does not appear to endanger the survival of the traditional intensifier -ísimo in the Spanish of Lima, Peru. On the one hand, -azo is a colloquial variant whose functions seem to be specializing in stative/resultative constructions including participles with an experiencer argument, i.e. with a perfective aspect; on the other hand, -ísimo is a formal variant whose linguistic functions are broader. The sharp contrasts between the two variants suggest that they could coexist for centuries to come.
References Alarcos, E. (1994). Gramática de la lengua española. Madrid: Espasa-Calpe. DOI: 10.3917/ling.462.0145 Alcina, J., & Blecua, J.M. (1991). Gramática española. Barcelona: Ariel. DOI: 10.2307/472904 Bello, A. (1981 [1847]). Gramática de la lengua castellana destinada al uso de los Americanos. Edición Crítica de Ramón Trujillo. Tenerife: Instituto Universitario de Lingüística Andrés Bello/Cabildo Insular de Tenerife. DOI: 10.1515/zrph.2005.708 Bosque, I., & Demonte, V. (1999). Gramática descriptiva de la lengua española 3. Madrid: Espasa. DOI: 10.14409/texturas.v1i4.2823
Álvaro Cerrón-Palomino Brown, L.A., & Tagliamonte, S. (2012). A really interesting story: The influence of narrative in linguistic change. University of Pennsylvania Working Papers in Linguistics 18, Selected Papers from NWAV 49, 2, 1–10. Cerrón-Palomino, Á. (2003). Sobre los grados del adjetivo calificativo español. Lexis, 27(2), 261–282. Falcinelli, A. (2007). Valores y funciones del sufijo -azo en el español actual y sus equivalencias en italiano. Rivista di Lingue e culture moderne, 1, 23–54. García Zapata, C. (1999). Aspectos pragmáticos de la exageración en el habla coloquial antioqueña. Colombia Lingüística y Literatura, 1(1), 9–20. Ito, R., & Tagliamonte, S. (2003). Well weird, right dodgy, very strange, really cool: Layering and recycling in English intensifiers. Language in Society, 32(2), 257–279. DOI: 10.1017/s0047404503322055 Kany, C. (1969). Sintaxis hispanoamericana. Madrid: Gredos. Labov, W. (1972). Sociolinguistic Patterns. Philadelphia, PA: University of Pennsylvania Press. DOI: 10.1017/s0047404500004528 Labov, W. (2001). Principles of linguistic change, social factors. Malden, MA: Blackwell. Lamíquiz Ibáñez, V. (1991). La cuantificación lingüística y los cuantificadores. Madrid: UNED. Martinell, E. (1989). Estilística de la gradación en los adjetivos. Actas del X Congreso de la Asociación Internacional de Hispanistas, 4, 1253–1262. Universidad de Barcelona. Matto de Turner, C. (1948 [1889]). Aves sin nido. Cuzco: Universidad Nacional del Cuzco. Montoro del Arco, E.T., & Almela Pérez, R. 2008. Grados de comparación: Errores tradicionales y nuevo marco teórico. Estudios Románicos, 16/17, 727–736. Peñalillo, M.I. (2012). Estudio sociolingüístico sobre el uso del superlativo en la región de Valparaíso, Chile. Unpublished Doctoral Dissertation. Universidad de Valladolid. Porto Dapena, J-Á. (1973). A propósito de los grados del adjetivo: Aportación al estudio del sistema de cuantificación en el adjetivo español. Thesaurus, 28, 344–357. Porto Dapena, J-Á. (1985). La cuantificación del adjetivo en español actual desde el punto de vista de la expresión. In A. Blecua, J.M. Blecua, & F. Rico (Eds.), Philologica Hispaniensia in Honorem M. Alvar 2, (pp. 541–555). Madrid: Gredos. Real Academia Española. (2010). Nueva gramática de la lengua española. Buenos Aires: Editorial Planeta. DOI: 10.1353/hpn.2013.0037 Roca Pons, J. (1958). Estudios sobre perífrasis verbales del español. Revista de Filología Española, 67, 1–403. DOI: 10.2307/471210 Sánchez, C., Marín, R., & Evert, S. (2012). Measuring lexical extension: The case of Spanish estar + past participle. Poster presented at Linguistic Evidence 2012. Empirical, Theoretical and Computational Perspectives, Tübingen, February 9–11. Tagliamonte, S. (2008). So different and pretty cool! Recycling intensifiers in Toronto, Canada. English Language and Linguistics, 12, 361–394. DOI: 10.1017/s1360674308002669 Tagliamonte, S. (2011). Variationist Sociolinguistics: Change, Observation, Interpretation. Malden, MA: Wiley-Blackwell. DOI: 10.7764/onomazein.28.6 Trudgill, P. (1972). Sex, covert prestige and linguistic change in urban British English. Language in Society, 1, 179–195. DOI: 10.1017/s0047404500000488 Varela, S. (1989). Verbal and adjectival participles in Spanish. In C. Laeufer & T. A. Morgan (Eds.), Theoretical analyses in Romance linguistics: Selected papers from the Linguistic Symposium on Romance Languages XIX, (pp. 219–234). Amsterdam: John Benjamins. DOI: 10.1075/cilt.74.16var
An experimental approach to hypercorrection in Dominican Spanish Barbara E. Bullock & Almeida Jacqueline Toribio The University of Texas
A stigmatized trait of popular Dominican Spanish is the insertion of preconsonantal or final [s] into a lexical item (e.g. Ja[s]queline). The predominant explanation among phonologists for this uniquely Dominican phenomenon is that the uneducated are ‘lost-s’ speakers who do not know which words contain lexical /s/ so they insert [s] into any coda, subject to theoretical constraints (Núñez Cedeño, 1988a; Harris, 2002; Bradley, 2006). However, a sociolinguistic analysis of rural Dominicans revealed that the majority of ‘s-ful’ tokens realized in spontaneous speech were correct productions (Bullock et al., 2014). The present analysis reports the results of a follow-up, controlled study designed to assess whether speakers could perceive and correctly reproduce coda [s] equally well across different phonological contexts, or whether internal and external factors (gender, literacy, age) affect how a speaker lexically codes and produces novel ‘s-ful’ tokens. Keywords: Spanish; Dominican Republic; intrusion; insertion; hypercorrection; s-deletion
1. Hypercorrection and language change In vernacular Dominican Spanish, postnuclear [s] is so frequently elided that speakers who produce it at high rates are regarded as inauthentic or as attempting to hablar fisno (speak hoity-toity) (Núñez Cedeño, 1988a; Alba, 2004, 2009). Alba (2004) refers to a speaker’s use of elevated rates of etymological /s/ as hipercorrección. This he distinguishes from ultracorrección, the insertion of a postnuclear [s] into a lexical item in which it was not present etymologically, as in José[s] for José or ma[s]ta for mata ‘tree’. Both types of hypercorrection are subject to social evaluation, but only the intrusive [s] is stigmatized as a risible stereotype of popular, often rural, Dominican Spanish. The two types of “correction” identified for Dominican Spanish by Alba reflect the taxonomy articulated by Janda & Auger (1992). For them, quantitative hypercorrection (Alba’s hipercorrección) defines a speaker’s overuse of a linguistic variable in his or her
doi 10.1075/ihll.4.13bul © 2015 John Benjamins Publishing Company
Barbara E. Bullock & Almeida Jacqueline Toribio
attempt to emulate the speech of others. Qualitative hypercorrection (Alba’s ultracorrección) refers to a speaker’s use of the variable in a linguistic context in which it is not historically warranted; in the case at hand, this is exemplified by the intrusive [s] in vernacular Dominican Spanish. Although Labov (1966, 1972) has argued that the study of hypercorrection provides essential information about speakers’ attitudes toward linguistic variables and about the direction of linguistic change, the major focus of variationist work has been, and continues to be, on quantitative hypercorrection. Janda & Auger (1992: 195) aptly remark that qualitative hypercorrection “has been virtually ignored by variationist scholars.” Indeed, this is true of the lion’s share of analyses of coda [s] in Spanish, where researchers have pursued Labov’s paradigm, weighing the effects of the various social and/or stylistic factors that were most predictive of the deletion/retention of etymological coda /s/, generally finding that the more formal the setting, the more speakers adjusted their rate of [s] usage upward. In other words, speakers quantitatively hypercorrect to match what they perceive to be a normative style (see File-Muriel, 2007 for a comprehensive overview of this research). We do not know if cases of qualitative hypercorrection – tokens of an intrusive [s] – appear in these corpora, perhaps because they fall outside the envelope of variation examined in these studies. In the Dominican context, where one can find both types of hypercorrection, there have been comparatively few variationist studies of etymological /s/ (see López Morales, 1990 and Bullock et al., 2014 for exceptions). Instead, linguists have tended to ignore the issue of whether or not popular Dominican Spanish speakers can adjust their rates of etymological /s/ production and have focused their analytical lens exclusively on the phenomenon of the intrusive [s], but from a theoretical perspective. One influential, and widely adopted, theory starts from the premise that illiterate speakers have lost the postnuclear [s] of normative Spanish from their lexical representations so that when they attempt to emulate the speech of others, they randomly insert [s] into a coda (Terrell, 1979; Harris, 1983, 2002; Núñez Cedeño, 1986, 1988a; Núñez Cedeño & Morales Front, 1999; Vaux, 2001, 2002; Vaux & Nevins, 2008; B radley, 2006). Recent quantitative sociolinguistic work, discussed in detail below, finds no empirical support for this theory (Bullock & Toribio, 2009; Bullock et al., 2014). At the same time, the corpus of natural speech examined by Bullock et al. (2014) yielded a very small number of intrusive [s] productions, reflecting the observations of Janda & Auger (1992) and of Yaeger-Dror (1992) that qualitative hypercorrection, although often salient, is also rare. The small token count prohibits a full exploration of the relationship, if one exists, between etymological /s/ and intrusive [s] production in vernacular Dominican speech. The present study returns to Dominican [s] hypercorrection, in its most general sense. Through an elicited repetition task, we induce rural Dominican speakers to produce elevated rates of ‘s-ful’ novel words in order to assess whether the [s] tokens
An experimental approach to hypercorrection in Dominican Spanish
they produce are most often correct (akin to quantitative hypercorrection) or incorrect (akin to qualitative hypercorrection). In particular, the study is designed to test whether incorrect [s] placements are found more often among semiliterate speakers than among literate speakers, as some theoreticians would predict. Before turning to the study and analysis of its results, we overview the background literature on Dominican [s] in the following sections, highlighting the theoretical connections drawn between [s] production and literacy in the Dominican context. We then discuss the challenges of working with rural, uneducated participants to justify the methods that we use here.
2. Previous work on Dominican [s] As noted in Bullock & Toribio (2009) and Bullock et al. (2014), the intrusive [s] for which popular Dominican Spanish is well known has been fundamentally mischaracterized in the research literature. Working largely from the secondary source data collected originally by Terrell (1979) and later supplemented through examples provided by Núñez Cedeño (1988a), linguists have argued that the intrusive [s], traditionally called hypercorrect [s] (Bradley, 2006), is a form of rule inversion (Venneman, 1972) in which an original rule of deletion (in this case of coda or word-final /s/) has become a rule of insertion. Furthermore, the literature would have readers believe that [s] intrusion is ubiquitous in Dominican Spanish, as the following quote contends (Lipski, 2008: 136): Even among educated speakers, rates of loss of word- and phrase-final /s/ are so high as to be nearly categorical. This has given rise to an incipient phonological restructuring and an exceptionally high level of hypercorrection, exemplified by the term hablar fisno < fino (to speak fancy), with hypercorrect inserted [s] (Bradley, 2006; Núñez Cedeño, 1986, 1988b; Terrell, 1982, 1986).
Harris (2002: 97) assumed that the level of etymological /s/ deletion was so high “in certain varieties of Caribbean Spanish” that he deemed the speakers of these varieties “lost-s speakers”. He attributed the use of an intrusive [s] to these putative “lost-s” speakers, and remarked that, among these speakers, coda [s] insertions are “essentially random”. Harris’ notion that the intrusive [s] could appear anywhere in a word was, perhaps, inspired by Núñez Cedeno’s (1988a) rule-based, coda-insertion account of Dominican hypercorrection, in which he posited that the input form abogado ‘lawyer’, could have any one of the following possible realizations: asbogado, abosgado, abogasdo, abogados. More recently, theoreticians have argued about whether rule-based or optimality-theoretical approaches can best account for the Dominican hypercorrect data (Vaux, 2001, 2002; Vaux & Nevins, 2008; Bradley, 2006).
Barbara E. Bullock & Almeida Jacqueline Toribio
The problem with these proposals is that they are not supported by the empirical data. Sociolinguistic analyses of popular Dominican Spanish have demonstrated that etymological /s/ is not categorically deleted but subject to social and stylistic conditioning. Even Terrell (1986), whose early work inspired Harris’ influential “lost-s” hypothesis, commented on the gender difference in overt [s] production, noting that women maintain considerably more [s] than men. Similarly, López Morales (1990) contests the assumption that Dominican Spanish has undergone phonological restructuring by demonstrating via a quantitative analysis that uneducated speakers are able to adjust their rates of postnuclear [s] production according to the level of formality of the interview. With regard to the linguistic patterning of intrusive [s], Morgan (1998) reports the results of a case study of 200 tokens of intrusive [s] that were uttered by a male tour guide in Santo Domingo. Morgan’s results show that far from being randomly distributed, intrusive [s] largely occurs in a restricted set of environments. Specifically, inserted [s] appears overwhelmingly before a voiceless stop consonant, occasionally word-medially but, more often, in a word-final environment before a voiceless stop. Morgan’s observations were subsequently confirmed by Bullock & Toribio (2010), who analyzed 103 tokens of intrusive [s] produced by Dominicans in an interview setting. In more recent work, Bullock et al. (2014) conducted a quantitative analysis of [s] production in Dominican Spanish with 40 participants, adults and children, in rural Cibao, a region well-known for its stigmatized speech forms. For etymological /s/, their findings mirrored those of other analyses of Spanish varieties that display coda [s] weakening. Even after the contribution of the speaker and the lexical item were taken into account as random terms, the factors that most favor /s/ retention include: the linguistic context (s is favored before voiceless stops and nasals, word-internally, before vowel-initial words, and prepausally), literacy (coda /s/ is produced more by literate speakers than by semiliterate speakers), and sex (females produce more coda [s] than males). Crucially, the study also considered intrusive [s] production among these same speakers and revealed that this form of qualitative hypercorrection was extremely rare. In the Bullock et al. study, there were only 82 tokens of intrusive [s] produced in a corpus of nearly 3,500 observations, spread out among 20 different speakers.1 Unfortunately, the limited distribution of these data points precluded a quantitative analysis of [s] intrusion, and only tenuous conclusions about qualitative hypercorrection could be drawn. Specifically, the data indicates that intrusive [s] appears to be
. Note that 4% of the productions of semiliterate speakers included an intrusive [s] compared to 1% for the literate speakers.
An experimental approach to hypercorrection in Dominican Spanish
primarily attracted to following voiceless stops, both within a word and, more often, between words, and secondarily, to prepausal position. Critically, however, the analysis demonstrated that there was no significant correlation between a speaker’s rate of etymological /s/ usage and his/her rate of intrusive [s], as would have been predicted by the “lost-s” hypothesis that implies that intrusive [s] should be confined to those who do not produce etymological /s/. To date, the research focus on Dominican [s] has been exclusively devoted to production. From this we know that some speakers, semiliterate males in particular, rarely produce coda [s] in their spontaneous speech. But the fact that they do not produce [s] spontaneously does not entail that they do not hear it in the speech of others or that it is categorically absent from their lexical representation. The following study describes an elicited repetition study designed to provide insights into how rural Dominicans might encode word-internal or word-final postnuclear [s] in words that they hear produced by others.2 The present research seeks to address the following questions and hypotheses: Question 1: Do literate speakers produce more correct novel “s-ful” tokens than semiliterate speakers in an elicitation task? Hypothesis 1: Yes; based on the results of the naturalistic interview data, we hypothesize that literate speakers will produce more correct tokens than semiliterates in a controlled task. Question 2: Do female speakers produce more correct novel “s-ful” tokens than males? Hypothesis 2: Yes; following the naturalist results, we expect females to produce more correct tokens than males in a controlled task. Question 3: Do semiliterate speakers produce more inserted [s] tokens than literate speakers? Hypothesis 3: Yes; following the naturalistic results, we expect semiliterate speakers to produce more intrusive [s] than literate speakers. Question 4: Are there linguistic factors that induce correct [s] placement? Hypothesis 4: Yes; following the naturalistic data and the phonotactics of Spanish, correctly placed [s] will occur most often before a voiceless stop.
. We limit the present to study to the word level, recognizing that this may only partially replicate the contexts in which intrusive [s] appears in natural speech. The descriptive work of Morgan (1998) and Bullock & Toribio (2010) suggests that the intrusive [s] occurs most frequently between words before a following voiceless stop. However, it also occurs word internally. A consequence of limiting our research to the word level is that, if our design is successful in eliciting realistic speaker behavior, the experiment may underestimate the occurrence of intrusive [s].
Barbara E. Bullock & Almeida Jacqueline Toribio
Question 5: Are there linguistic factors that induce an [s] to be moved from its original position? Hypothesis: Yes; following the naturalistic data, “moved” [s] will be attracted most to the position before a voiceless stop.
3. An excursus on working with uneducated participants The present contribution reports the results of a controlled study intended to elicit elevated rates of overt postnuclear [s] from rural Cibaeño speakers who are similar to those who participated in the sociolinguistic interview study discussed above (Bullock et al., 2014). A primary motivation for this study is to provide an empirical contribution to the literature on linguistic hypercorrection. However, a secondary goal, and one that we believe should be accorded pride of place in discussions of experimental methodology, is to demonstrate effective ways in which linguists can overcome the challenges of working with speakers who do not read and who have little education. It bears mentioning that the methodology that we report on here may not be as meticulously constructed as is desirable for a laboratory study, but it represents our first successful attempt in assessing how uneducated speakers potentially lexically encode the postnuclear [s] that they hear produced by others. Our earlier attempts included tasks that were, perhaps to the linguist’s mind, much simpler and much more controlled, including a phoneme detection task and a lexical decision task. However, despite training, our participants were unable to complete the types of identification tasks that are commonly used in the psycholinguistics lab. Therefore, we decided to construct a task that would have local meaning. The methodology that we employ in this study is an outgrowth of our many years of experience with and observation of language behavior in the rural Cibao. Specifically, we have long noted a tendency among Dominican speakers to epenthesize an intrusive [s] in proper names. Our own names form sites of [s] intrusion such that we are frequently referred to as Bárbara[s] and Ja[s]cqueline, and not in jest. More surprising still is the fact that some speakers commonly use an intrusive [s] in their own names when they introduce themselves: Eri[s]ka, José[s], Alfredo[s], Braulio[s], Gu[s] tiérrez, etc. Additionally, Dominicans are fanciful in the names they bestow on their children, e.g. Raibelis, Leonelbis, Yanquel, Cruzdilandi, Aneuri, Leogris, Brailin, Wandel, Minosca, and Yudelqui, choosing names that are syllabically complex and quite frequently portmanteau constructions, e.g. Yanelson < Yianni + Nelson, Elcaleomaris < Elga + Leonardo + María. Occasionally when there are multiple children in a family, each child might possess a name that is an alliterative variant of another, for example Waster, Willi, Wendi, Wanda or Liche, Lichei, Licheini, Licheinmi. These names can be so exotic that we have encountered participants on more than one occasion who are
An experimental approach to hypercorrection in Dominican Spanish
not actually sure how to pronounce their children’s names. We exploit the phonological creativity of Dominican names in the construction of our experimental stimuli.
4. Methodology For the present investigation, we created what we call a “name game” experiment where a target [s] was embedded in 28 novel female proper names. We did this because, as discussed above, a task in which speakers are asked to repeat novel names is more plausible to our semiliterate participants than one in which they are asked to repeat nonce or real words out of context.
4.1 Stimuli The stimuli (28 items), included in the Appendix, were all three-syllable novel words, with the position of [s] balanced across words to appear in final position or wordinternally before a voiceless stop, a voiced approximant, a fricative, a nasal, or a liquid in stressed and unstressed syllables. There are also 28 distractor items of three-syllable length, each of which contained at least one coda other than [s]. In preparing the materials, the second author, a Dominican native to the region of field study, recorded each target item six times, from which the best token was selected. These tokens were embedded into a pre-recorded carrier phrase that included a declarative sentence, followed by the stimuli produced with a narrow focus intonation, and then a question: Mi prima tiene un nombre muy especial: [name]. ¿Cómo se llama mi prima?” ‘My cousin has a very special name: [name]. What’s my cousin’s name?’ All prompts (carrier phrase + target) were edited to be of the same length, and the intensity of the prompts was normalized in Praat (Boersma & Weenink, 2013) using the script created by Yi Xu (2002). The auditory stimuli were randomized in SuperLab Pro.
4.2 Participants There were 51 participants in the name game experiment: 27 children (15F and 12M, ages 7–15) and 24 adults (11F and 13M, ages 21–64), literate and semi-literate, from the same community as the above referenced sociolinguistic study of [s] (Bullock et al., 2014). We determined the literacy level of the adults by asking them about their ability to read and write and their terminal level of schooling. Those who had completed fewer than four years of school were classified as semiliterate. In our experience, the adults are very forthcoming about their literacy skills. As for the children, they too are forthcoming about their literacy, but occasionally they report being able to write but not read. In these cases, we administered a short reading passage; invariably these children could only write their names, but could not read. Because some of our participants are proud
Barbara E. Bullock & Almeida Jacqueline Toribio
of the fact that they can write their names, we prefer to refer to them as semiliterate rather than illiterate. All of our participants report normal hearing.
4.3 Procedure Participants were seated in front of a laptop, wearing headphones and a head-mounted Shure WH20 cardioid microphone. Participants were instructed to listen to the prompt and respond with the phrase “Tu prima se llama [name]” ‘Your cousin’s name is [name]’, repeating the name they heard. This procedure follows the paradigm of Guion (2003), which minimizes mimicry and requires speakers to attempt a lexical coding. After an initial training period, the subjects completed the self-paced task in the presence of the first author, who recorded their responses onto a Marantz PDM 660 digital recorder, at a sampling rate of 44kHz with 16 bit quantization. The experiment itself took approximately 8 minutes to complete.
4.4 Coding The recordings were downloaded onto a laptop and relabeled for coding. We broadly transcribed each production. Then we coded each token for whether the postnuclear [s] from the prompt was repeated correctly, whether it was deleted, whether it was moved from its original position, whether it was both produced in its original position and reduplicated, or whether it was resyllabified into a syllable onset. We also coded the phonological context (preceding and following) of each realization or deletion. Items were coded as ‘correct’ if the speaker maintained the [s] in the same syllable and before the same consonant. If [s] was maintained in the same syllable but the following context was changed, the response was coded as incorrect, and the [s] was coded as appearing before the actual consonant that the participant produced. Examples of the coding procedure are given in Table 1. In the case of multiple realizations of [s], the token was coded twice: once because the original [s] was maintained and once because an intrusive [s] appeared in the production. The same is true for moved [s] tokens: these were coded once for the deleted [s] and a second time for the moved [s] production. In a few, rare cases, two sibilants appeared in a word, as in Yosnéki > Yoné[s]ki[s]. In this case the production was coded three times: twice as moved [s] and once as deleted. Resyllabifications occurred when the [s] preceded a liquid, as in Jilísra > Jilísa. In these cases, the item was coded as incorrect, and [s] was coded as resyllabified. We should note that participants faithfully reproduced the stress profile of each of the target words. Observations were excluded in the case where a participant failed to respond to the prompt, or in cases in which the participant failed to produce anything similar to the prompt. The data for two of the 51 original study participants, one adult male and one adult female, were discarded from the final analysis for this latter reason.
An experimental approach to hypercorrection in Dominican Spanish
Table 1. Sample coding Input
Example Production
Correct
Classification
Metonís
Metonís
Correct
Original s
#
Me[s]toní[-]
Incorrect Incorrect
Moved s Deleted s
t #
Metoní[-]
Incorrect
Deleted s
#
Me[s]toní[s]
Incorrect Correct
Moved s Original
t #
Me[s]poni[-]
Incorrect Incorrect
Moved s Deleted s
p #
Jilísra
Correct
Original s
r
Jilí[-]ra[s]
Incorrect Incorrect
Deleted Moved
r #
Jilí[-]ra
Incorrect
Deleted
r
Jilí[s]ra[s]
Correct Incorrect
Original s Moved s
r #
Jilísa
Incorrect
Resyllabified
a
Jilisra
Following Environment
4.5 Results The experiment yielded 1,406 observations, of which 55 (4%) were excluded. The total counts are presented in Table 2. Table 2. Tokens analyzed Production
N
Ratio
Coda [s] produced in its original position
502
.369
Coda [s] moved from its original position
200*
.147
No coda [s] produced in the word
563
.414
Coda [s] resyllabified to onset
96
.070
Total observations
1361
* This number includes the tokens with a second coda [s] as well as an [s] in original position (n = 47)
As shown, participants deleted [s] at fairly high rates: 41% of their productions contained no [s] although they would have heard an [s] in the input form. In addition, nearly 15% of the observations contained moved [s], that is, an [s] that was present in the input that the participant heard but reproduced elsewhere in the word.
Barbara E. Bullock & Almeida Jacqueline Toribio
4.6 Analysis Mixed effect regression models in R (R development Core Team, 2008) were fitted to the data, in order to examine the factors that are most predictive of both correct [s] production and moved [s] production, the latter interpreted here as the experimental counterpart of the intrusive [s]. The independent variables included the fixed effects of Literacy of the participant (Literate or Not), the Age group of the participant (Child or Adult), the Sex of the participant (M or F), the Following Environment (voiceless stop (T), voiced approximant or stop (D), liquid (L), nasal (N), fricative (F), or pause (#)), and Stress (whether the [s] occurs in a stressed syllable or not). The target word and the individual participants were entered as random terms. Our decision to enter the individual study participants as a random factor was based on our personal observation and knowledge that some speakers produce more [s] (both correct and hypercorrect) than others. We follow Johnson (2009) and Drager & Hay (2012) in employing a mixed statistical model in order to avoid false effects. The dependent variable for the first analysis was whether the participant’s production was Correct or Incorrect. Table 3 presents the results of the best fitting regression analysis in the Rbrul format (Johnson, 2009). The best model for the data with Participant and Target as random effects included the Following Environment and Literacy as (highly) significant fixed main effects. No other independent variables contributed significantly to the variance and there were no significant interactions among the fixed effects.3 Factor weights above .5 are interpreted as favoring correct [s] production. This particular model uses centered factor weights, which is standard in GoldVarb analyses. As shown, literate subjects highly favor correct [s] production, and semiliterates disfavor it. With regard to the phonological environment, voiceless stops highly favor the retention of a correct s, followed by nasals, and marginally by the voiced approximants/stops. Note that the prepausal position disfavors [s] retention. The dependent variable for the second analysis was whether the participant moved the [s] in the reproduction or not. Here, we are interested in the factors that contribute to the probability that an [s] will be inserted into the wrong position. The independent variables remain the same. In Table 4, the results of the best fitting model with Participant and Target as random factors return the Following Environment and Stress as significant main effects. Note that Stress in this model is only marginally significant. None of the other independent variables were significant and no interaction between them was significant.
. A reviewer suggested that we model the data selecting only fixed effects. Thus, we created additional models without target word or individual participant as random intercepts. The best fixed effects model returns Literacy and Following Environment as highly significant main effects as well. However, the model fits the data poorly, relative to the model we report here.
An experimental approach to hypercorrection in Dominican Spanish
Table 3. Best Step-Up Model for Correct [s] Is With Participant [random] and Target [random] and Following Environment (4.93e-17) + Literacy (0.000854) [p-values building from null model] Deviance
1078.617
df
9
Grand mean
–1.016
Literacy
factor
logodds
tokens
y/y+n
centered factor weight
Y
0.952
606
0.553
0.721
N
–0.952
556
0.302
0.279
factor
logodds
tokens
y/y+n
centered factor weight
T
2.449
325
0.683
0.92
N
0.319
251
0.422
0.579
D
0.060
102
0.461
0.515
#
–0.658
184
0.250
0.341
L
–0.716
182
0.308
0.328
F
–1.454
118
0.220
0.189
Following Environment
Participant random SD
1.79
Target random SD
1.011
Table 4. Best Step-Up Model for Moved [s] Is With Participant [random] and Target [random] and Following Environment (2.11e-46) + Stress (0.0446) [p-values building from null model] Deviance df
643.99 9
Grand mean
–3.027
Following Environment
factor
logodds
tokens
y/y+n
centered factor weight
T
3.790
194
0.469
0.978
Stress
#
3.082
197
0.299
0.956
N
0.625
184
0.212
0.651
D
–0.761
61
0.098
0.318
F
–2.095
95
.032
0.11
L
–4.641
128
0.016
0.01
factor
logodds
tokens
y/y+n
centered factor weight
Y
0.302
295
0.339
0.575
N
–0.302
564
0.177
0.425
Participant random SD
0.85
Target random SD
2.104
Barbara E. Bullock & Almeida Jacqueline Toribio
Three levels of the Following Environment proved to favor moved [s]: voiceless stops and pause are highly favorable following environments for moved [s], and nasal stops to a lesser degree. Stressed syllables marginally favor moved [s] production. Unlike the case of correct [s] production, no language-external factors proved to be significant predictors of when an [s] would be inserted into the wrong position.4
5. Discussion This study complements our previous variationist analyses of [s] production (both etymological and intrusive) in the natural speech of rural Dominicans with a controlled elicitation technique. In other words, it is an attempt to experimentally induce both quantitative and qualitative hypercorrection, i.e. to induce elevated rates of [s] production among speakers who often delete /s/. The data from the experiment are intended to mimic productions of etymological /s/, what we have called “correct [s]”, as well as intrusive [s], which we term “moved [s]”. These parallels between natural speech and elicited repetitions are not exact since speakers perform differently under experimental conditions than they do under more natural interview conditions. In addition, we cannot know if our rural participants interpreted the experimental task as a formal one requiring them to attempt to adjust their style upward. Our primary aim here was to provide them with input that contains novel s-ful tokens to examine how they might encode and reproduce the segment. Our analysis of correct [s] production in this experimental task is partially consistent with variationist results regarding the effect of literacy on [s] production. Literate speakers produce significantly higher rates of etymological /s/ in natural speech than do their semiliterate counterparts (Bullock et al., 2014). The same is true of this experimental data: literate speakers produced significantly higher rates of correct coda [s]. However, in contrast to the results of studies of natural speech, there was no effect of sex in this controlled elicitation study. In all variationist studies of Dominican speech, males have been shown to produce significantly diminished rates of [s] relative to females (Terrell, 1979; López Morales, 1990; Bullock et al., 2014). In essence, while literacy remains a significant predictor of correct [s] production, the effect of sex disap-
. Here again, we created additional models without target word or individual participant as random intercepts at the recommendation of a reviewer who is interested in revealing subtle effects. In the present case, a model with only fixed effects (i.e. without the Participant or Target as random intercepts) reveals that literate speakers are significantly more likely than semiliterate to move s in this task and adults are more likely than children to do so. However, we believe these effects to be potentially spurious, reflecting a Type 1 error.
An experimental approach to hypercorrection in Dominican Spanish
pears in a controlled task. We do not have an explanation for this finding, but it could be due to a variety of cultural and/or socio-indexical factors. For instance, a speaker’s gendered identity may not be at issue in a repetition task. Also consistent with the results of analyses of natural speech data, voiceless stops act as attractors for correctly produced postnuclear [s]. Our participants’ elevated rates of correct [s] in this environment mirror the phonotactics of Spanish word-internal sC clusters reported in File Muriel (2007): 90.6% of /s/ occurs before voiceless stops in the Spanish lexicon, and this is the environment most favorable to correct [s] reproduction among our participants. Curiously, the word-final, prepausal position disfavors [s] reproduction in this task. Thus, the question arises as to whether word-final [s] is difficult for Dominicans to perceive. We tested this possibility by examining whether a final [s] in the stimulus was reproduced in any syllable-final position in the repetition less frequently than a word-internal [s] prompted by the stimuli. For instance, we calculated whether the final [s] in a word like Metonís would be more or less likely to be overtly realized, correctly or incorrectly, in our participants’ utterances than a word-internal [s], as in a word like Kasmíta. The results of a chi-square test revealed that there was no significant difference between these two types of words (X [s]quare = .4045 df = 1, ns (p = .525)), indicating that Dominicans hear final [s] as well as they hear word-internal [s]. However, they tend to avoid pronouncing an [s] in word-final position. A potential explanation for our participants’ apparent aversion to word-final [s] realization is that they may have interpreted the prepausal [s] produced in the prompt as a mistake on the part of the speaker and, thus, filtered it out of that position in their production or restored it elsewhere. Alternatively, it is possible that our participants do not perceive a prepausal [s] as part of the word. Instead, for them, it could mark a phrasal boundary or have socio-indexical meaning. In fact, Bullock & Toribio (2010: 69) note that some speakers frequently use an intrusive [s] at the end of phrasal boundaries, irrespective of the grammatical category of the phrase final word. The results for the moved [s], our experimental counterpart to the intrusive [s] of natural speech, demonstrate that literate and semiliterate speakers alike will produce a moved [s] at exactly the same rates. This is unlike the pattern observed for the correctly reproduced [s], our experimental counterpart to etymological /s/. This difference is depicted graphically in Figures 1 and 2. Figure 1 shows the patterns of distribution of correct [s] according to the following phonological context. The size of the circles in the graph represents the number of observations in each environment. The top line displays the results for the literate speakers (Y) and the bottom line, the results for the semiliterate speakers (N). It should be noted from Figure 1 that the patterns of distribution according to the following context are identical for the two groups of participants. However, the two groups differ with respect to the rate of [s] production: the literate speakers produce
Barbara E. Bullock & Almeida Jacqueline Toribio N Y 1.0
Correct word
0.8
0.6
0.4
0.2
0.0 #
d
f
l
n
t
Following
# = Pause; d = voiced approximant/stop; f = fricative; l = liquid; n = nasal; t = voiceless stop Figure 1. Correct [s] production x literacy and phonological context
significantly higher rates of [s] in every environment. Both groups produce more [s] before voiceless stop consonants and very few in prepausal position or before fricatives or laterals. Figure 2 displays the results of the observations of moved [s] by phonological context. Here, it is clear that the differences observed between literacy levels have collapsed. All participants produce moved [s] at the same rate and to the same degree in each environment. Again, they produce the sibilant most often before a voiceless stop. However, unlike the correct [s], which is dispreferred in prepausal position, the moved [s] favors this context. In sum, the social and the linguistic factors that predict the realization of moved [s] differ from those that predict the realization of correct [s]. The distributional results for the two types of [s] produced in the experiment contest the rule-inversion view of Dominican qualitative hypercorrection. Rule-inversion accounts of intrusive segments hold that they are most frequently inserted where they are also most likely to be deleted. If this were true of the intrusive [s] in Dominican Spanish, we would expect that it would occur frequently in prepausal position, where it is often deleted in both laboratory and natural speech, but not pre-consonantally before a voiceless stop. It is clear from these results that the voiceless stop acts as a strong attractor for both types of [s].
An experimental approach to hypercorrection in Dominican Spanish N Y 1.0
Moved S
0.8
0.6
0.4
0.2
0.0 #
d
f
l
n
t
Following
Figure 2. Moved [s] production x literacy and phonological context
The hypotheses that we formulated in response to the research questions posed in this study were based on what we knew to be or expected to be true of rural Dominicans’ realization of [s] in natural speech. Returning to our original research questions, we find that only some of these hypotheses have been confirmed by the results of this controlled study. Specifically, our data confirmed the following hypotheses: literate speakers produce more correct [s] tokens than semiliterate speakers (Hypothesis 1), mirroring how these groups behave in natural speech, and there are specific linguistic factors that constrain the placement of correct and moved [s] (Hypotheses 4, 5). However, our data do not support Hypothesis 2: while we expected female speakers to produce more correct novel s-ful tokens than males, as they do in natural speech, this was not the case in this experimental task. More surprisingly, our data do not support Hypothesis 3: semiliterate and literate speakers were observed to produce moved [s] at identical rates in this controlled task. In sum, literacy, but not gender, remains a significant predictor of correct [s] realization. However, no social factors were found to be predictive of moved [s] realization. The fact that literate and semiliterate speakers did not differ quantitatively or qualitatively in the production of moved [s], which we take to be akin to the intrusive
Barbara E. Bullock & Almeida Jacqueline Toribio
[s] of natural speech, is a compelling finding. In natural speech, the rates of intrusive [s] production are too low to determine the linguistic internal and social factors that are most predictive of its use. Through this experimental task, we have been able to induce robust numbers of occurrences of incorrect placement in novel words, which has allowed us to test whether this form of hypercorrection would be limited to semiliterate speakers as has been proposed and assumed in the theoretical literature. In fact, it appears not to be conditioned by literacy at all. In this respect, its production differs markedly from that of the correct [s] in novel words, which we take to be the experimental counterpart of the etymological /s/. One implication of the present research is that moved [s] and correct [s] may be entirely different entities, as they are subject to different linguistic and social conditioning. In the natural speech context, they have always been presumed to differ in terms of the social dimension in that producing elevated rates of etymological /s/ marks a speaker as uppity and/or effeminate while producing an intrusive [s] stigmatizes a speaker as uneducated. This study demonstrates that, in reality, it is only the production of etymological /s/ that is socially conditioned. All speakers, irrespective of their literacy level, produced the same rates of [s] intrusion. This suggests that the occurrence of an intrusive [s] is not at all motivated by a speaker’s lack of knowledge about lexical /s/ distribution. In other words, the “lost-s” hypothesis is unfounded. But, what is the function of the inserted [s] in popular Dominican Spanish? Who uses it and under what circumstances? These questions await further research.
References Alba, O. (2004). Cómo hablamos los Dominicanos. Santo Domingo, Dominican Republic: Ramón Jimenés. Alba, O. (2009). La identidad lingüística de los Dominicanos. Santo Domingo, Dominican Republic: Ediciones Librería La Trinitaria. Boersma, P., & Weenink, D. (2013). Praat. Doing phonetics by computer (version 5.3.51). Bullock, B.E., & Toribio, A.J. (2009). Reconsidering Dominican Spanish: Data from the rural Cibao. Revista Internacional de Lingüística Iberoamericana, 14, 49–73. Bullock, B.E., & Toribio, A.J. (2010). Correcting the record on Dominican [s]-hypercorrection. In S. Colina, A. Olarrea, & A. Carvalho (Eds.), Romance linguistics 2009 (pp. 15–24). Amsterdam: John Benjamins. DOI: 10.1075/cilt.315.01bul Bullock, B.E., Toribio, A.J., & Amengual, M. (2014). The status of [s] in Dominican Spanish. Lingua, 143, 20–35. DOI: 10.1016/j.lingua.2014.01.009 Bradley, T. (2006). Spanish rhotics and Dominican hypercorrect s. Probus,18, 1–33 DOI: 10.1515/probus.2006.001 Drager, K., & Hay, J. (2012). Exploiting random intercept: Two case studies in sociophonetics. Language Variation and Change, 2, 59–78. DOI: 10.1017/s0954394512000014 File-Muriel, R. (2007). The role of lexical frequency and phonetic context in the weakening of syllable-final lexical /s/ in the Spanish of Barranquilla, Colombia. Unpublished Doctoral Dissertation. Indiana University.
An experimental approach to hypercorrection in Dominican Spanish
Guion, S. (2003). The vowel systems of Quichua [s]panish bilinguals. Phonetica, 60, 98–128 DOI: 10.1159/000071449 Harris, J. (1983). Syllable structure and stress in Spanish. Cambridge, MA: The MIT Press. DOI: 10.1017/s027226310000526x Harris, J. (2002). Flaps, trills and syllable structure in Spanish. In A. Csirmaz, Z. Li, A. Nevins, O. Vaysman, & M. Wagner (Eds.), MIT Working Papers in Linguistics 42, (pp. 31–108). Cambridge, MA: The MIT Press. Janda, R.D., & Auger, J. (1992). Quantitative evidence, qualitative hypercorrection, sociolinguistic variables — And French speakers’ ‘eadaches with English h/Ø. Language & Communication, 12, 195–236. DOI: 10.1016/0271–5309(92)90015–2 Johnson, D.E. (2009). Getting off the GoldVarb standard: Introducing Rbrul for mixed-effects variable rule analysis. Language and Linguistic Compass, 3(1), 359–383. DOI: 10.1111/j.1749-818x.2008.00108.x Labov, W. (1966). The social stratification of English in New York City. Washington, DC: Center for Applied Linguistics. DOI: 10.1017/cbo9780511618208.018 Labov, W. (1972). Language in the inner city: Studies in the Black English vernacular. Philadelphia PA: University of Pennsylvania Press. DOI: 10.1177/089124167600400410 Lipski, J. (2008). Varieties of Spanish in the United States. Washington, DC: Georgetown University Press. DOI: 10.1017/s0047404510001053 López Morales, H. (1990). En torno a la /s/ final dominicana: cuestiones teóricas. Voz y Letra, 1, 129–137. Morgan, T. (1998). The linguistic parameters of /s/-insertion in Dominican Spanish: A case study of qualitative hypercorrection. In J. Gutiérrez-Rexach & J. del Valle (Eds.), Perspectives on Spanish linguistics (pp. 79–96). Columbus, OH: Ohio State University. Ñúñez Cedeño, R. (1986). La /s/ ultracorrectiva en dominicano y la estructura silábica. In J.M. de Alba (Eds.), Actas del II Congreso Internacional sobre el Español de América, (pp. 337–47). México: Universidad Nacional Autónoma de México. Núñez Cedeño, R. (1988a). Structure-preserving properties of an epenthetic rule in Spanish. In D. Birdsong & J.P. Montreuil (Eds.), Advances in Romance linguistics, (pp. 319–335). Dordrecht: Foris. Núñez Cedeño, R. (1988b). La discontinuidad fonética como criterio dialectológico. Thesaurus, 43, 1–17. Núñez Cedeño, R. & Morales-Front, A. (1999). Fonología generativa contemporánea de la lengua española. Washington, DC: Georgetown University Press. R Development Core Team. (2008). R: A language and environment for statistical c omputing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-0701, URL 〈http://www.R-project.org〉 Terrell, T. (1979). Final /s/ in Cuban Spanish. Hispania, 62, 599–612. DOI: 10.2307/340142 Terrell, T. (1982). Relexificación en el español dominicano: Implicaciones para la educación. In O. Alba (Ed.), El español del Caribe, (pp.301–318). Santiago de los Caballeros: Universidad Católica Madre y Maestra. Terrell, T. (1986). La desaparición de /s/ posnuclear a nivel léxico en el habla dominicana. In R. Núñez Cedeño, I. Páez Urdaneta, & J. Guitart (Eds.), Estudios sobre la fonología del Caribe, (pp. 117–134). Caracas: Ediciones la casa de Bello. Vaux, B. (2001). Consonant epenthesis and hypercorrection. Unpublished manuscript. Harvard University. Vaux, B. (2002). Consonant epenthesis and the problem of unnatural phonology. Unpublished manuscript. Harvard University.
Barbara E. Bullock & Almeida Jacqueline Toribio Vaux, B., & Nevins, A. (2008). Rules, constraints and phonological phenomena. Oxford: Oxford University Press. DOI: 10.1017/s002222670800563x Vennemann, T. (1972). Rule inversion. Lingua, 29, 290–242. DOI: 10.1016/0024–3841(72)90025–3 Yaeger-Dror, M. (1992). Introduction. Language & Communication, 12, 181–192. DOI: 10.1016/0271–5309(92)90014-z Yi Xu. (2002). 〈http://www.phon.ucl.ac.uk/home/yi/praat/_normalize_amplitude.praat〉
Appendix “Name game” stimuli * All novel words appear with a diacritic over the stressed syllable, whether or not Spanish orthographic conventions would warrant it. Metonís Lipukás Pelústi Omíska Fénosda Bodásna Mafísmi Rumésli Jilísra
Yadósfi Fistóba Muskéti Lisdóma Yosnéki Kasmíta Chaslégui Nasríma Disfómi
Daléras Notéguis Jánesti Vémiska Gánesda Nópasna Dátesmi Cámusli Yánusra Lóbesfi
Dialect identification and listener attributes Do you hear la tonada? Jennifer Lang-Rigal
James Madison University This study investigates the perception of vowel-lengthening, a feature characteristic of the Spanish spoken in Córdoba, Argentina: the tonada cordobesa. Uniquely, the lengthening occurs on the pre-tonic syllable (Fontanella de Weinberg, 1971; Yorio, 1973; Lang, 2010), and is believed to be accompanied by a pitch peak (Fontanella de Weinberg, 1971). The goals of this experiment are to determine if duration alone (i.e. without intonational changes) is a strong enough cue for identifying speakers from Córdoba, and what listener features affect this perception. A matched-guise methodology is employed in which natural and manipulated (pre-tonic vowel duration) tokens are presented to Argentine listeners in a dialect identification task. Results show that longer pre-tonic vowel durations are associated with a Córdoba speaker origin, regardless of the speaker’s true regional origin or other linguistic cues. Keywords: Argentina; prosody perception; dialect identification; vowel lengthening
1. Introduction Although dialects are often described as having different intonations, only a limited set of studies have systematically tested dialect categorization among listener populations by focusing on prosodic or intonational features specific to each regional variety. Of these studies, some find that intonation alone is sufficient for determining the dialect of a given speaker and test listeners on filtered speech in which phonetic information is removed leaving only the intonational contour (Gooskens, 1997; van Bezooijen & Gooskens, 1999; Peters et al., 2002; Leyden & Heuven, 2006, inter alia). Other studies suggest that listeners require more than the intonational contour in order to identify a speaker’s dialect, and their findings indicate that having more stressed syllables, longer utterances, and stronger break indices are necessary for categorization (Soukoup, 2009; Feizollahi & Soukup, 2011). While the same results are not expected to determine the perception of the dialect spoken in Córdoba, Argentina, the rhythmic p atterns,
doi 10.1075/ihll.4.14lan © 2015 John Benjamins Publishing Company
Jennifer Lang-Rigal
and in particular the lengthening of the vowel in pre-tonic position, is expected to influence perception of this dialect. Listeners perceive this lengthening as a stress-shift (Malmberg, 1950), and testing this feature experimentally and in isolation is an innovation that allows for a more controlled investigation of dialect perception. In the present study we apply speech perception methodology through use of a dialect identification task to examine the influence of pre-tonic vowel lengthening on the ability of Argentine listeners to identify a speaker’s Córdoba origin. Pre-tonic vowel lengthening is observed to be characteristic of the tonada cordobesa, the Spanish spoken in Córdoba, Argentina by linguists and laymen alike1 (Fontanella de Weinberg, 1971; Yorio, 1973). For example, for the word ‘tonada’ most varieties of Spanish would have increased duration, pitch, and perhaps even intensity on the tonic syllable /na/ relative to the other syllables (Hualde, 2002). Whereas, in the tonada cordobesa, these correlates of stress might apply to the pre-tonic syllable /to/ of ‘tonada’, resulting in a realization [toːˈna.ða]. Despite the widespread popular attention this dialect has received and the uniqueness of pre-tonic vowel lengthening across languages, perceptions of this dialect feature have not been examined. The goals of the present study are: to determine whether the feature of vowel duration alone is enough for a speaker to be perceived as being from Córdoba, and to investigate differences in perception between listeners from Córdoba (in-group) and listeners from Buenos Aires (out-group). These questions will be addressed using a matched-guise perception test.
2. Intonation in Argentinean Spanish 2.1 Prosodic features of the tonada cordobesa The variety of Spanish spoken in Córdoba, Argentina has not escaped popular attention, although it has rarely been an object of serious study and remains poorly understood. The nicknames used to denote this regional variety, la tonada cordobesa (‘the Cordoban tune’) and el cantito cordobés (‘the Cordoban lilt’), suggest that it is distinguished from other dialects at least in part by its intonation. Early linguists also have pointed out the notable rhythm characteristic of this dialect and recognize the lengthening before the stressed syllable. Malmberg (1950) suggests that this lengthened vowel is accompanied by a high tone and the displacement of the accent expiratoire, the latter being the essential characteristic of this phenomenon. Vidal de Battini (1964) calls it “un acento musical” (‘a musical accent’) and makes the observation that the vowel lengthening, combined with a tone height change, is so evident that
. See for example the ‘grammar rules of the tonada cordobesa’ on the Córdoba tourism webpage from Jaworski, 〈http://www.vallepunilla.com.ar/gramatica.php〉, and additional examples in Section 2.1 of this chapter.
Dialect identification and listener attributes
it often gives the impression of producing a geminate vowel. However, only two linguists have devoted their attention to the tonada cordobesa beyond simple commentary. F ontanella de Weinberg (1971) and Yorio (1973) describe the phenomenon as a pattern of lengthening of the syllable immediately preceding the last accent of the intonational phrase, shown in (1) and (2) with stress marked by an acute diacritic and lengthening marked by a colon. Importantly, lengthening is believed to be accompanied by tonal movement, or at least tonal contrast occurring on the lengthened vowel (Fontanella de Weinberg, 1971). Vowel lengthening is also found to occur, although less commonly, in other phrasal positions, which may convey emphasis (Fontanella de Weinberg, 1971). Yorio’s findings imply that lengthening may vary according to speech style (i.e. reading and conversational style). The pre-tonic vowel lengthening occurs more frequently both in pretonic position within a breath-group and in conversational style speech (Yorio, 1973). (1) No la coːnózco. ‘I don’t know her/it.’ (2) Quedáte a cenár con noːsótros. ‘Stay and have dinner with us.’ An important limitation to these studies, however, is that they relied on impressionistic analyses, which may not accurately capture the interplay of duration and pitch, nor other prosodic features that may be relevant to either the production or perception of this linguistic feature. A more recent pilot study provided additional insight into the prosodic characteristics of pre-tonic vowel lengthening (Lang, 2010). Lang’s study examined duration, intensity, and average pitch (F0) for the pre-tonic and tonic vowels in phrase-final position of 47 speakers in both spontaneous and read speech. Neither intensity nor duration were significantly different in t-tests comparing the pre-tonic and tonic vowels for each utterance. Lang’s finding for duration is unusual for Argentine Spanish and many other varieties, which typically exhibit a longer duration and increased intensity in tonic vowels (Hualde, 2002). However, when averages for Fundamental Frequency were compared between pre-tonic and tonic vowels of each token, pitch was found to be significantly higher in the tonic syllable relative to the pre-tonic (p < .01). Furthermore, pre-tonic vowel lengthening was found to vary between speaker groups divided by gender, age, and social class, and by speaking context. Unlike previous work, this pilot study is based on data from acoustic analyses.
2.2 Speech perception and intonation research on Argentinean Spanish As mentioned above, current data on the prosody characteristics of the tonada cordobesa has been limited to impressionistic observations, with no available systematic empirical or controlled experimental study of its peculiar intonational curves, or
Jennifer Lang-Rigal
of listeners’ ability to accurately discern it. The vast majority of linguistic studies in Argentina have focused on Buenos Aires, or porteño, Spanish. This is not unexpected, since Buenos Aires is not only the nation’s capital, but also the political, economic and cultural center, a major port of entry, and most populated city in Argentina. Furthermore, more generally, the research on Spanish intonation remains a recent development in the field of linguistics, with the most significant portion of this research being focused on the Castilian variety in Spain (cf. the many works of Face, Prieto). And, while a recent increase of attention has been given to Argentine Spanish intonation (Colantoni & Gurlekian, 2002, 2004; Pešková et al., 2012; Kaisse, 2001; Labastía, 2006; Toledo, 1989, 1997, 2000; Rodriguez, 2008; Fontanella de Weinberg, 1980; Enbe & Tobin, 2011; Gabriel, 2010; Gabriel et al., 2010, 2011, 2013; Estebas-Vilaplana, 2010), most studies have focused on Buenos Aires and porteño Spanish, which has become synonymous with ‘Argentine Spanish.’ In contrast to the increasingly abundant studies on Buenos Aires Spanish, other dialect zones in Argentina are much less frequently the object of linguistic research. Thus, while the capital city of Córdoba is second to Buenos Aires in size and rivals it in historical significance, only a handful of studies have been devoted to its dialect, primarily the works of Colantoni (2005, 2011), Castellani (1998), Prevedello (1991), Cacciavillani (1988), Catinelli (1985), as well as Fontanella de Weinberg (1971) and Yorio (1973), who are the only two to investigate intonation. This work has been descriptive, listing the features of Cordoban speech as they deviate from Latin American or Castillian standard Spanish.
2.3 Dialect categorization and language attitudes Much linguistic variation is due to the tonal and durational contrasts, such as suprasegmental features of intonation or prosody, some of which have been tested by use of experimental methods of speech perception. Although dialects are often remarked to have different intonational patterns, only a limited set of studies has systematically tested dialect categorization and/or attitudes among listener populations focusing on the prosodic or intonation features specific to each regional variety. Of these studies, some claim that intonation alone is sufficient for a listener to determine the dialect of a given speaker, testing listeners on filtered speech in which phonetic information is removed, and only the intonational contour remains (Bezooijen, 1988; Gooskens, 1997, for Glasgow dialect and Received Pronunciation (English); Peters et al., 2002 for Northern Standard German, Berlin Urban, and Hamburg Urban varieties of German). Other studies suggest that listeners require more than the intonational contour in order to identify a speaker’s dialect in a perception task. Differing strengths and frequencies of stress and tone breaks (corresponding to silences between intonational phrases) are also believed to index the use
Dialect identification and listener attributes
of dialect versus the standard variety for Austrian listeners, while the intonational contour alone (Hann pass filtered) was not enough to differentiate between these two varieties (Feizollahi & Soukup, 2011; Soukoup, 2009 for Austrian dialects). Specifically, F eizollahi & Soukup (2011) found that the samples of the standard variety (Austrian German) consistently showed more stressed syllables and stronger break indices, and that perceptions were biased towards the standard variety when utterances were longer and displayed a greater pitch range, while the shorter sentences were biased towards perception of the dialect (Middle Bavarian-Austrian dialect). An experimental approach to language variation in which suprasegmental information is altered can reveal what cues are important in both the auditory (sound) perception and the social (meaning) perception of a given speaker. Language experience and attitudes can greatly influence the perception of linguistic variation, complicating the issue of dialect categorization. For example, linguistic experience with and exposure to the dialects being tested influenced listeners’ performance on an identification task of German dialects (Peters et al., 2002) and American English dialects (Clopper & Pisoni, 2004). The present study employs experimental methodology with the objective of bridging some gaps in the research literature. This study investigates language variation from the standpoint of the listener, with attention given to a specific prosodic feature from a regional variety of Spanish. In sum, this paper seeks to demonstrate how the saliency of a single variant, namely, pre-tonic vowel lengthening, may carry shared social meaning among Argentine Spanish speakers.
3. Methodology The aim of this paper is threefold: to ascertain whether listeners can identify the Cordoban dialect; to determine whether the feature of ‘pre-tonic vowel duration’ alone is sufficient for identifying a speaker from Córdoba; and to investigate the influence of listener background on dialect identification. To this end, this study uses a matched guise technique. Specifically, a perception test, created from recordings previously collected by the author, was administered to Argentine participants originating from Córdoba, Buenos Aires and other provinces of Argentina.
3.1 Speakers Recordings from a total of 40 speakers were used to develop this perception test, all of which had been interviewed by the author in 2010 and 2009 in three Argentinean cities (or urban areas): 1. Córdoba (provincial capital), 2. San Miguel and Yerba Buena, Tucumán, and 3. Buenos Aires (national capital). All selected speakers claimed to be lifelong residents of their respective urban area (see Table 1 for a listing of speaker
Jennifer Lang-Rigal
origin). As the present study’s focus is on the perception of the tonada cordobesa, the inclusion of speakers from Tucumán and Buenos Aires provided experimental controls from two distinct dialect zones, the Northwest and the Littoral zone, respectively.2 All recordings present naturalistic, spontaneous speech though the method of elicitation varied for the Córdoba speakers. The Córdoba recordings come from sociolinguistic interviews in which participants responded to open-ended questions about their life. For Tucumán and Buenos Aires speakers, spontaneous speech was elicited by means of a role-play task, presenting hypothetical scenarios to participants and asking them to respond as if they were involved in the scenario (La Encuesta Porteña, Gabriel et al. 2010).3 The demographic profiles of the speakers from different regions also vary. All Córdoba speakers included in this study share the following characteristics: (1) They had lived their entire lives in Córdoba capital; (2) They had parents who were both from the province of Córdoba; (3) They had not spent more than 3 months outside of Córdoba; and (4) They were not fluent in any language other than Spanish. Finally, in allowing for the greatest variation within this dialect sampling, this experiment included a balanced sample of speakers according to sex, age, and social class (as estimated by their responses to a socio-demographic questionnaire). In contrast, there were fewer speakers from Tucumán and Buenos Aires and they had similar profiles in terms of social class and age. These speakers are listed in Table 1. The token types shown in Table 1 are described in the following section. Table 1. Speaker and Token distribution SPEAKER ORIGIN CÓRDOBA
SEX
TOKENS
Male
Female
Natural
Manipulated
Younger (19–32)
2
2
8
8
Middle-aged(33–47)
2
2
8
8
Older(48–64)
2
2
8
8
6
6
Total # CO speakers
12
24
24
BUENOS AIRES
2
2
8
8
TUCUMAN
2
2
8
8
Total # BA+TU speakers
8
16
16
Total all speakers
20
40
40
. Based on the five dialect zones of Argentina (Vidal de Battini, 1964). . The encuesta porteña, and other versions are available from Atlas interactivo de la entonación del español, 〈http://prosodia.upf.edu/atlasentonacion/metodologia/encuestas/encuestaargentina.doc〉
Dialect identification and listener attributes
3.2 Stimuli Short excerpts, including phrases, partial phrases and entire utterances, all ranging from 4–9 syllables and averaging 1.0 sec total duration were used as stimuli for the perception test. Since this study is interested in measuring perceptions of Córdoba speakers and their speech patterns, the majority of the excerpts (two-thirds, n = 24) are from 12 different Córdoba speakers balanced by sex, age group and social class. Of the remaining 16 stimuli, half come from Buenos Aires speakers, and half from Tucumán speakers, balanced by sex (total 8 speakers). Stimuli are divided into three main groups by speaker, which also correlates with the function of the token within the experiment. Additionally, there are two different stimuli for each speaker, resulting in a total of 40 unique stimuli (see Table 1). Spontaneously produced speech was considered necessary for this study, as the author, in the piloting of this experiment and in Lang (2010) found lengthening of the pre-tonic vowel to occur to a lesser degree or not at all in controlled speech, a behavior that was also reported by Yorio (1973), who compared spontaneous and reading styles of speech. The selection of stimuli from the naturalistic recordings considered many linguistic factors. First, the content was controlled for references to proper nouns, regionally associated slang words or expressions, or any other information that would identify the speaker’s socio-demographic profile. These included but were not limited to expressions not common to standard Spanish, use of the augmentative suffix -azo, specific regional references, and all cases in which the alveopalatal fricative /ʒ/ (i.e. words with ‘y’ or ‘ll’), and the assibilated (or trilled) /r/ could occur (i.e. words with ‘rr’). Furthermore, tokens necessarily included a multisyllabic word with a syllable in pre-tonic position; this word is considered here as the target word, in which the experimental manipulation was made (details of the manipulation are discussed later in this section). Thus, a word like ‘casa’ would not qualify as a viable target word, since the tonic (stressed) syllable /ka/ is at the beginning of the word and thus there is no syllable before it; whereas a word like ‘tomate’ could be included, as it has in pre-tonic position the syllable /to/ before the tonic syllable /ma/. Only monopthong vowels were included in the target words. The target word occurred in different intonational phrase positions within the token, with the purpose of testing pre-tonic vowel lengthening in medial as well as final phrase positions (the latter is believed to be the principal environment in which pre-tonic vowel lengthening occurs (Fontanella de Weinberg, 1971; Yorio, 1973)). Using an example of a token from the test stimuli, “siempre un poquito”, the target word is ‘poquito’ which has a syllable in pre-stress position (unlike ‘siempre’, which is stress-initial) and the target word ‘poquito’ occurs at the end of the intonational phrase. In addition, all tokens were controlled for non-dialectal linguistic features, including number of syllables, intonational contour, and syllable weight and position in words with pre-tonic vowel lengthening. An example of one of the tokens can be seen in Figures 2 and 3. Sound quality was also
Jennifer Lang-Rigal
considered; only the highest quality and most intelligible excerpts were included as tokens. After applying these criteria, tokens were chosen by random selection in order to reduce bias while controlling for other linguistic and quality factors which could interfere with interpreting the results. Tokens were then submitted to Praat (Boersma & Weenik, 2010) and all recordings were converted to 44100 KHz and to stereo sound. The process for creating the second pair part of the matched guise set-up, which involves manipulating the pre-tonic vowel duration of the target word in each token. Each of the 40 tokens was duplicated, the original copy was left as is, creating the natural token group, while the second copy was submitted to Praat for manipulation. For the manipulated tokens, first the pre-tonic vowel of the target word was segmented to determine the natural durations. Within the manipulation window in Praat, points were placed at the vowel onset and offset, to allow for the manipulation of just the vowel segment, leaving the duration of the rest of the stimulus (as well as all other linguistic and acoustic features) unchanged. The vowel segment was then altered to create either a longer or shorter duration, depending on the speaker’s region of origin. For Córdoba speakers, the natural pre-tonic vowel duration was on average double that of the Buenos Aires and Tucumán tokens, so these vowels were shortened to 50% of their original duration. This was done by lowering the selected segment to the 0.5 marker in the manipulation window in Praat. For the Buenos Aires and Tucumán speakers, the pre-tonic vowel duration was lengthened to become twice as long as the original by raising the segment to 2.0 in the manipulation window.4 The natural durations proved to be greatly varied, with the shortest (35ms) and the longest (194ms) durations coming from the Córdoba speakers. However, the duration averages revealed a pre-tonic vowel duration for the Córdoba speakers (n = 12; duration = 109ms) that was nearly double that of the durations from the Buenos Aires speakers (n = 4; duration = 59ms) and notably longer than that of the Tucumán speakers (n = 4; duration = 72ms). Hence, the reasoning for using proportions to create the manipulated stimuli was to emulate natural differences, account for the variation between tokens and speakers, and ensure that the manipulation was exaggerated enough to be perceptible in the experiment. The final tokens, both natural and manipulated, were presented to five native Spanish speakers (four of them Argentine) to test for naturalness of stimuli and ease of the task before gathering data. None of the piloting participants or actual participants noticed that tokens had been manipulated and believed all of them to be natural.
. This type of manipulation did not affect the vowel transitions (i.e. no segments were “cut-off ”) and did not result in any perceivable differences from the original, other than duration itself.
Dialect identification and listener attributes
Recall that each token consists of a multisyllabic word (i.e. the target word) with a syllable in pre-tonic position naturally occurring within a partial, or entire intonational phrase (declarative neutral). While the target word usually occurs at the end of the token, their position within the greater intonational utterance may be either medial or final. The examples shown in Figures 1 and 2 represent one of the 40 matched pairs of manipulated and non-manipulated tokens used in the Dialect Identification task. This speaker from Córdoba had an already long pre-tonic vowel (Figure 1). Like all of the tokens from Córdoba speakers, the natural token was manipulated to create a duplicate of the natural token in which the pre-tonic vowel duration is only half of that of the original (Figure 2). For the Buenos Aires and Tucumán speakers, the same method is applied and only the vowel is doubled in duration. The proportion of the manipulation, either doubling or halving the original pre-tonic vowel duration, was performed for each token irrespective of the original vowel length, which may have been naturally long or short relative to other vowel durations within the same word. Instead, the manipulation is executed based solely on the speaker’s region of origin: for Córdoba speakers it is halved and for Buenos Aires and Tucumán speakers it is doubled from its natural duration.
a
s
.024 .100
i-e
.072
n
l a
k
.064.012 .029 .084
o
s
.120
.122
i
n
a
.028.046.031
Asi en la cocina. 0
Time (s)
0.9216
Figure 1. Pitch, waveform, and spectrogram for male, Córdoba speaker (token id_05), “Así en la cocina” with highlight on natural vowel duration of pre-tonic vowel /o/.
Jennifer Lang-Rigal
a
s
i-e
.024
.100
.072
n
l a
k
.064.012 .029 .084
o
s
.060
.122
i
n
a
.028.046.031
Asi en la cocina. 0
Time (s)
0.8609
Figure 2. Pitch, waveform, and spectrogram for male, Córdoba speaker (token id_06), “Así en la cocina” with highlight on manipulated (shortened) vowel duration of pre-tonic vowel /o/
3.3 Procedure Stimuli in the perception task were presented through a website (www.lablinguistica.com). Participants were recruited through the personal networks of the author, social websites (e.g. Facebook) and an announcement through the university listserv. The experiment was advertised as a test to determine your ability to identify an Argentine speaker, and in particular a Cordobés, by his accent. The manipulations to the stimuli were not advertised in the description or instructions, only that the task would be challenging and required an ideal environment for listening and concentration. This was followed by specific directions for completing the dialect identification task, which was the first of three tasks performed as part of the larger experiment. The task took participants 10–15 minutes to complete. The written directions in Spanish asked participants to listen carefully to each voice, and make their best guess as to the speaker’s origin based on their first impression. A multiplechoice response between three regions (Córdoba, Buenos Aires, Tucumán), forced them to choose one of the regions along a three-point certainty scale: definitivamente ‘definitely’; probablemente ‘probably’; or quizás ‘perhaps’ (see Figure 3). The purpose of this scale is to give a sense of flexibility to the participant for being forced into guessing the speaker’s identity, which proved to be a difficult task in this study.
Dialect identification and listener attributes
Participants were allowed to listen to the stimuli as many times as they wanted, but were instructed to make a quick decision and move on to the next stimulus. Participants were not allowed to modify answers once they were submitted. The order of the stimuli was pseudo-randomized so that each participant did not hear any of the matched “pairs” of stimuli within four tokens of each other. The experiment also includes a second perception test of language attitudes (not reported as part of the present study), and a questionnaire soliciting personal information regarding residential and language experience.
Figure 3. Screenshot of Dialect Identification task (www.lablinguistica.com)
3.4 Listeners Over 100 participants were recruited for this study, but only those who both completed all parts of the survey and claimed Argentine identity were included in the data analysis, for a total of 63 listeners representing 10 different provinces of Argentina. All participants were adults (aged 18–66; mean age = 36 years) who claimed to have no known hearing problems. Participants are further broken down into groups based on their self-identified regional affiliation, which directly reflects the listener’s sense of identity and origin, and is likely to be more revealing than the perception of their identity based on their place(s) of residence. This was elicited with the question: “Si alguien de otra provincia de Argentina te preguntara de dónde sos, qué le dirías?” (‘If someone from another province of Argentina were to ask you where you were from,
Jennifer Lang-Rigal
what would you say?’). These answers were compared to their responses to a similar origin question,5 and since nearly all participants (61 of 63) responded identically to both questions, they are treated as synonymous for the participants for designating their regional origin. The listeners were also grouped according to their exposure to Córdoba speakers. This category is determined by the responses to three yes/no questions: (1) Do you have family in Córdoba?; (2) Do you have friends in Córdoba?; and (3) Have you ever visited Córdoba (province)?. The combined responses lend a scalar grouping of low, medium or high experience/affinity to Córdoba speakers with the objective of gauging the observations from previous studies that the more influential aspect of dialect categorization is level and variety of exposure to the dialects being tested (see Peters et al., 2002). If the participant answered “yes” to all three questions, he or she received a high score of affinity, two “yes” answers receives a medium score of affinity, and one or zero receives a low score.
3.5 Analysis Upon hearing each auditory stimuli, listeners responded to the question “De que región viene [el hablante]?” (‘What region is he/she [the speaker] from?’). Results from the dialect identification task were tested to determine the accuracy of the listener’s response, i.e. whether they were able to select the correct province of the speaker in both the natural and manipulated versions of a given speaker’s stimuli. Listener responses were coded as 1 for a correct response and 0 for an incorrect response, and accuracy was analyzed to reveal three different effects of the manipulation. First, the responses for all listeners were grouped to test identification accuracy based on the speaker region (Córdoba, Buenos Aires, Tucumán) and identification accuracy based on token type (natural and manipulated). These were tested with a Repeated Measures 2-way ANOVA. Second, the incorrect responses were analyzed to investigate the factors influencing a misidentified speaker. The greatest number of misidentifications occurred with the manipulated stimuli and so the analysis of incorrect responses focuses on these stimuli only. Finally, one-way ANOVAs were performed to determine the effect of two listener-specific variables on dialect identification accuracy: region of origin and experience with the Córdoba dialect.
. “Dónde te criaste? Si has vivido en más de una ciudad, escribe el lugar más importante para vos (con el cual te sentís más identificado).” (‘Where did you grow up? If you have lived in more than one city, write the most important to you (with which you identify the most).’).
Dialect identification and listener attributes
4. Results 4.1 Dialect identification Several ANOVAs tested for the significance of the following factors: token type (natural versus manipulated); speaker region (Córdoba versus Buenos Aires versus Tucumán); listener origin (Córdoba versus Buenos Aires versus Tucumán versus Santa Fe versus Other); and listener affiliation with Córdoba (low versus medium versus high). A repeated measures 2-way ANOVA found token type to be significant in the accuracy of responses: the natural tokens resulted in significantly higher accuracy (59%) relative to the manipulated tokens (35% accuracy; (F(1,62) = 223.7, p < .01). When the two token types are considered together, accuracy differed significantly according to speaker region (F(2,62) = 76.4, p < .01). The average accuracy for the Córdoba speaker stimuli is 57.3% for the natural tokens, and 27.9% for the manipulated tokens with a shortened pre-tonic vowel, a significant difference when tested with a one-way ANOVA (F(1,47) = 17, p < .01) Figure (4). The accuracy was highest in responses to the Buenos Aires speaker stimuli, averaging 80.6% correct identification across all listeners; this is significantly different from the identification of the manipulated tokens from the same speakers, in which an average accuracy of 51.8% was attained (in a post-hoc one-way ANOVA: F(1,15) = 86.7, p < .01). For the
Correct indentification of speaker region (%) 100 90
80.6
80 70 60
57.3
51.8
50 39.3
40
29.6
28.0
30 20 10 0 Córdoba: natural
Córdoba: Buenos Aires: Buenos Aires: manipulated natural manipulated
Tucumán natural
Tucumán manipulated
Figure 4. Correct speaker identification averages of all listeners, grouped by Speaker Region and Token Type
Jennifer Lang-Rigal
Tucumán stimuli, the averages for both natural and manipulated tokens were low when all listener groups were combined: 39.3% accuracy for the natural tokens and 29.6% for manipulated tokens. The interaction of token type and speaker region is also significant (F(2,62) = 19.7, p < .01). Post-hoc (LSD) tests found significant differences between all regions for natural tokens and significant differences for manipulated tokens between Córdoba and Buenos Aires speakers (p < .01), as well as Buenos Aires and Tucumán speakers (p < .01).
4.2 Dialect identification between different listener groups 4.2.1 Listener origin When listeners are divided by groups according to their province of origin the accuracy of listener identification shows some significant differences between certain groupings, and very little difference between others. There were four groups created for participants from Córdoba (CO), Buenos Aires (BA), Tucumán (TU), and Santa Fe (SF) with eight participants or more in each group, and a fifth group was added for Argentines originating from other provinces to provide some insight into how a mixed-origin group performs. In general, identification accuracy improves slightly when the listener group origin matches that of the speaker group. Turning first to results for the Córdoba speaker stimuli (CO), for the natural tokens, the mean accuracy was 57.5% and there were no significant differences by listener region of origin (F(4,62) = 1.277, p > .05). For the manipulated tokens there was a main effect for listener region (F(4,62) = 4.523, p < .01). The highest accuracy was for the Córdoba listeners (40%), followed by Other (27%), Buenos Aires (26%), Santa Fe (22%), and Tucumán (21%). A post-hoc (LSD) test showed significant differences in accuracy between Córdoba listeners and all groups except Other (p < .01, see Figure 5). CO Natural
100 90 80 70 60 50 40 30 20 10 0 BA
SF
TU
CO Manipulated
100 90 80 70 60 50 40 30 20 10 0 Other
CO
TU
SF
BA
Other
CO
Figure 5. Identification accuracy means for Córdoba stimuli (natural and manipulated). Listeners are grouped by region of origin and ordered left to right, from lowest to highest accuracy
Dialect identification and listener attributes
With respect to the responses to the Buenos Aires speaker stimuli, the natural tokens received the highest accuracy averages for all listener groups (80%), with a significant main effect for listener region (F(4,62) = 5.365, p = .001). Buenos Aires listeners scored significantly higher (89.3%) than Córdoba (70%) and Tucumán (68%) listeners in a post-hoc LCD test (p < .01). Buenos Aires listeners also out-performed listeners from Santa Fe (86%) and Other (83%), but only in a one-tailed test (p < .05; Figure 6). For the manipulated Buenos Aires tokens (overall mean accuracy = 51.7%) there were no significant differences between any listener groups (F(4.62) = 0.986, p > .05). BA Natural
BA Manipulated
100 90 80 70 60 50 40 30 20 10 0
100 90 80 70 60 50 40 30 20 10 0 TU
CO
Other
SF
BA
TU
CO
BA
SF
Other
Figure 6. Identification accuracy means for Buenos Aires stimuli (natural and manipulated). Listeners are grouped by region of origin and ordered left to right, from lowest to highest accuracy
Finally, the responses to the Tucumán speakers’ stimuli received the lowest accuracy averages for all groups (42%), except for the Tucumán listeners. For the natural tokens, listener groups performed significantly differently (F(4,62) = 6.726, p .05; Figure (7).
4.2.2 Listener experience Analyses of variance were re-run on accuracy rates for Córdoba stimuli using a different method for grouping listeners: they were organized according to their affinity to Córdoba speakers (based on having friends, family, and/or experience traveling to
Jennifer Lang-Rigal TU Manipulated
TU Natural 100 90 80 70 60 50 40 30 20 10 0
100 90 80 70 60 50 40 30 20 10 0 BA
SF
CO
Other
TU
BA
SF
CO
TU
Other
Figure 7. Identification accuracy means for Tucumán stimuli (natural and manipulated). Listeners are grouped by region of origin and ordered left to right, from lowest to highest accuracy
Córdoba) into three groups of low, medium, and high affinity. Results showed no main effects for affinity to Córdoba speakers for either the natural or manipulated Córdoba speaker tokens. The low affinity group did show a lower accuracy average (53.7%) for the natural Córdoba tokens than the medium affinity (59.1%) and high affinity (58.5%) groups; however, since these differences are not significant we can make no conclusions about this listener factor.
4.3 Dialect misidentification In this section we consider the responses to the manipulated tokens only to assess how the listeners identified the speakers, which as reported in the previous sections, tended to be inaccurate in comparison to the natural tokens. Not all manipulated tokens resulted in a misidentification, however the majority of these tokens were not accurately identified, and the extent to which they were misidentified varied depending on the speaker’s region of origin (Figure 8). Slightly over 70% of the (shortened) manipulated tokens from Córdoba speakers were misidentified. They were correctly identified at a nearly equal rate between the three dialect regions: Córdoba (27.5%), Buenos Aires (32.8%) and Tucumán (39.7%). The majority of the manipulated (lengthened) tokens from the Tucumán speakers were misidentified as speakers from Córdoba (60.2%), and only a small percent, 8.6%, of the lengthened Tucumán tokens were identified as coming from Buenos Aires speakers. The manipulated (lengthened) tokens from Buenos Aires speakers received the highest percentage of correct identifications, at 51.8%, with 48.2% of all tokens being misidentified – 31.1% of which were erroneously perceived as spoken by Córdoba speakers (Figure 8).
Dialect identification and listener attributes Cordoba - shortened CO 27.5%
BA 32.8%
TU 39.7%
Tucuman - lengthened TU 31.1% BA 8.6%
CO 60.2%
Buenos Aires - lengthened
BA 51.8%
CO 31.1% TU 17%
Figure 8. Response percentages (CO = Córdoba, BA = Buenos Aires, TU = Tucumán) for 3 types of manipulated tokens: (1) Córdoba speakers with shortened pre-tonic vowels; (2) Tucumán speakers with lengthened pre-tonic vowels; and (3) Buenos Aires speakers with lengthened pre-tonic vowels
5. Discussion 5.1 Pre-tonic vowel length and Córdoba-ness The most important implication of the dialect identification results in this study is the difference in accurate identification for tokens from natural speech and those from manipulated speech. Recall that the manipulated tokens differed from the natural tokens in only one respect – the vowel duration of the pre-tonic vowel of the target word. The results for the stimuli from Córdoba speakers demonstrate that reducing the pre-tonic vowel duration by half significantly influences the accuracy in identifying a speaker’s province of origin: the shortened (manipulated) versions of the tokens result in a very low accuracy rate (60%) nearly equally divided between the two other options: Buenos Aires and Tucumán. This result very strongly suggests that longer pre-tonic vowel duration is associated with speakers from Córdoba. There is a similar effect in the results for the dialect identification accuracy of the stimuli from the other speaker groups. The lengthening of the pre-tonic vowel duration in the manipulated versions of tokens for the Buenos Aires and Tucumán speakers also appears to have contributed to inaccurate identification. Nearly half of the manipulated Buenos Aires speaker stimuli were misidentified, and of those most (68%) were believed to be from Córdoba. The results for the lengthened Tucumán tokens are even more notable: they were misidentified at the highest rate with most being identified as Cordoban: 60.2% of the total responses, or 87.5% of the misidentified responses. These results suggest that pre-tonic vowel duration serves as a cue for identifying a speaker as being from Córdoba, even when pitch or other segmental and linguistic features are unchanged. There are always doubts when comparing the perception
Jennifer Lang-Rigal
of natural speech to (computer) manipulated speech, as is done in this experiment. However, since the duration manipulation does not disturb other segmental features or result in unnatural sounding speech, the effects of the manipulation were minimal. Furthermore, in the piloting of this experiment, no listeners were able to identify which stimuli were manipulated and which were not. In fact, at no point in time did any participants show awareness that some of the stimuli had undergone manipulation, although many commented that the task itself was difficult. In previous work on speech perception, it has been found that duration differences of less than 30 milliseconds are difficult for humans to distinguish (Toledo, 1989, 2000). Thus those tokens where an already short pre-tonic vowel duration was manipulated to create an also very short ( nasal, it was more likely that the trill was produced as a voiced multiple vibrant as opposed to a voiceless approximant (Lewis, 2004:119). Recasens (1991) and Recasens & Pallarès (1999) discovered a small effect of surrounding vowels, where back vowels seemed to facilitate apical closure, and thus, more trill-like variants. From an aerodynamic perspective, Solé (2002) found that reduced oropharyngeal and subglottal pressure, imperfect articulatory positioning, and increased tongue-tip tension and mass brought about more fricative and approximant variants, thus showing that trills require more constrained conditions in their production (Solé, 2002:682). Furthermore, some analyses have found that prosodic factors such as syllable stress, word position, and word length significantly affect trill production with regard to normative versus non-normative variants, number of occlusions, and segment duration. Normative variants were favored word-initially and in words with four or more syllables in Venezuelan Spanish (Díaz-Campos, 2008). In Jerezano Andalusian Spanish, there were more occlusions for medial trills in unstressed syllables (Henriksen & Willis, 2010). Additionally, Willis (2006, 2007) discovered longer segments in word-initial position in Dominican Spanish, and Bradley & Willis (2012) found longer segments in stressed syllables in Veracruz Mexican Spanish. Extralinguistic factors have also been a focus in variable trill production. Lastra & Martín Butragueño (2006), Díaz-Campos (2008), and Henriksen & Willis (2010) found that younger speakers always differed from older speakers in trill production. In Díaz-Campos (2008), younger speakers favored more innovative variants (approximants and single taps), and in Lastra & Martín Butragueño (2006) and Henriksen & Willis (2010), younger speakers favored more standard-like realizations. With regard to gender differences, innovative variants were more attested in the speech of women in L astra & Martín Butragueño (2006), whereas innovative variants were found more in the speech of males in Bradley and Willis (2012), Díaz-Campos (2008) and Henriksen & Willis (2010). Lastly, Lastra & Martín Butragueño (2006) and Díaz- Campos (2008) also found significant effects of social class. While the former revealed that assibilation was favored among speakers of the lower and middle classes, the latter observed that the lower and higher classes favored approximants and single taps. It appears that frequency of use could be a driving force behind the variation found in trill production, thus necessitating a focus on usage-based models of phonology (e.g. exemplar models). These models hypothesize that increased frequency of a form leads to phonological reduction. Different reduced variants are recorded in the lexicon, forming various exemplar clusters, which become stronger and easily activated due to the predictability and recency of the words in which they appear (Langacker,
Delano S. Lamy
2000; Bybee, 2001; Ernestus, 2014). Therefore, it could be that non-standard variants of the trill, which appear to be reduced forms, are the result of substantive and temporal reductions in speech, particularly in high frequency words (see Mowrey & Pagliuca, 1995). This perspective of phonology differs from abstractionist models in which there is one lexical representation of a word where phonetic details are absent and emerge in production through the application of rules (see Ernestus (2014) for a more detailed discussion on this comparison). It is important to note that, in usage-based models, reduction is controlled (Pierrehumbert, 2001) specifically by extralinguistic factors. That is, speakers choose to produce (non)reduced variants according to their gender, age, socioeconomic class, etc. (Ernestus, 2014), and language variation and change ensues (Bybee, 2000; Bybee & Torres-Cacoullos, 2008). This variation is evident in the case of trill production, as will be shown in the present analysis. We see that trill production has been a well-documented variable in the literature; however, there are very few analyses of this sound in Panamanian Spanish. To the best of my knowledge, there is only one descriptive/impressionistic account of rhotics in this variety (see Alvarado de Ricord, 1971) and three quantitative variationist studies (see Broce & Torres Cacoullos, 2002; Cedergren, Rousseau & Sankoff, 1986; Cedergren & Sankoff, 1974). All of these analyses focus on the alveolar tap in coda position and its alternation with the alveolar lateral. To date, no study has given a detailed account of trill production in Panamanian Spanish. The present analysis seeks to fill this gap by examining trill production within a variationist sociolinguistic framework. Additionally, I included three variables related to trill production: (1) normative/non- normative alternation (N/NN), (2) mean number of occlusions (MO), and (3) mean duration (MD). Although all these variables have been discussed in previous literature, the latter two have not been considered in variable rule analyses. Furthermore, the present analysis explores usage-based hypotheses, particularly, that high lexical frequency favors reduction. Section 3 presents the methodology used in the analysis.
3. Methodology 3.1 Data elicitation and acoustic analysis The data for the present study were elicited through sociolinguistic interviews conducted with 11 monolingual speakers of Panama City Spanish (henceforth PCS). The speakers represented three age groups (14–19, 26–37, and 53+), two socioeconomic classes (lower and middle),1 and included both male and female participants (see Table 1).
. Socioeconomic class was determined according to the area of origin and personal knowledge that the author has about the participants. There are certain areas in Panama where different socioeconomic classes are represented. This is the case for Las Acacias, where we see
A sociophonetic analysis of trill production in Panamanian Spanish
Table 1. Participants Speaker Sex Age Area
Class
Education
Jennifer
F
19 Las Acacias
lower
university (currently)
55
Lesly
F
18 Pedregal
lower
high school (currently)
54
Manuel
M
18 Las Acacias
middle high school (currently)
62
Leo
M
53 Las Cumbres middle high school with post-secondary courses
87
Gladys
F
30 Tocumen
middle university (currently)
49
Magaly
F
37 Balboa
middle high school with post-secondary courses
60
Magda
F
53 Las Cumbres middle university
75
Richard M
50 El Dorado
middle university
71
Joel
M
14 Las Acacias
lower
high school (currently)
24
Sara
F
62 San Miguelito lower
high school
21
Valeria
F
26 Las Acacias
high school not completed; post-secondary courses
50
lower
Tokens
In order to ensure natural, spontaneous speech, interview modules were adopted from Tagliamonte (2006) and tailored to fit culturally with the speakers’ lives in Panama. All interviews lasted between 45 minutes and 1.5 hours. The recordings were done using a Sony Digital Voice Recorder ICD-SX750 and a Shure Microphone SM10A. 608 tokens of the trill were extracted and submitted to acoustic analysis using Praat (Boersma & Weenink, 2012). Following Bradley & Willis (2012), the acoustic cues considered were presence of occlusions, waveform amplitude reduction, transition in formant structure and reduced intensity.
3.2 Statistical analysis and coding A variable rule analysis was carried out using Rbrul (Johnson, 2009). As previously mentioned, trill production was analyzed based on three variables: (1) normative/ non-normative alternation, (2) mean number of occlusions, and (3) mean duration. The categorization of normative/non-normative was based on the number of occlusions. A variant with 2 or more occlusions was labeled as normative (or trilllike), and a variant with fewer than 2 occlusions was labeled as non-normative (or non-trill-like). Normative/non-normative alternation was treated as a binary variable, and thus the data were submitted to a logistic regression analysis. In this type of analysis, all
that Manuel is from a different socioeconomic level than Jennifer and Valerie. Therefore, other factors were taken into account, such as occupation of the participant or the participants’ parents, characteristics of their home and belongings, etc.
Delano S. Lamy
factor groups are considered at once in order to ascertain which ones significantly condition trill production. Furthermore, the analysis determines which factors within the factor groups favor or disfavor the variants in question. Rbrul provides both logodds and factor weights for the logistic regression. The latter will be used in the presentation of the results in keeping with previous variationist work. A factor weight of higher than .5 has a favoring effect, whereas a factor weight below .5 has a disfavoring effect. Mean occlusions and mean duration are considered numeric, or continuous variables, and therefore must be submitted to linear regression analyses. This type of analysis estimates the effects of factors on mean values (Johnson, 2009). The results are reported as coefficients, whereby a positive coefficient shows favoring towards greater mean values, and a negative coefficient shows favoring towards lower mean values. The independent linguistic factor groups included in the analyses were the following: (1) preceding segment type (vowel versus consonant versus zero); (2) syllable stress (tonic versus atonic); (3) number of syllables; (4) position in the word (initial versus medial); (5) grammatical category; and (6) word frequency. The independent extralinguistic factor groups included were the following: (7) speaker age; (8) speaker sex; (9) socioeconomic class; and 10) level of education (high school versus university). Factor groups 3, 6, and 7 could have been included in the multivariate analyses as continuous predictors; however, for purposes of the present study, number of syllables consisted of three groups (a: 1–2, b: 3, and c: 4+), age, as mentioned above, consisted of three groups (a: 14–19, b: 26–37, c: 53+),2 and frequency consisted of two groups (lower frequency words and higher frequency words). Local frequency was used in an attempt to capture the true experience speakers have with words (see Erker & Guy, 2012). That is, frequency counts were taken from the corpus from which the data were extracted (13 speakers; 13,681 words) because it represents the immediate experience with the discourse. This is important when taking into account the effects of frequency within a usage-based model of language because, as mentioned above, frequency of occurrence determines not only predictability, but also recency. The most recent words are those being produced in the moment. Therefore, calculating frequency based on a corpus consisting of the participants’ speech is a more accurate account of their experience than using a frequency corpus of General Spanish. In the present analysis, words that made up at least 1% of the corpus were
. Age was not included as a continuous variable because there are clear separations between each age group. There is a difference of 7 units between groups A and B, and a difference of 26 units between groups B and C. Also, as the author is familiar with all of these participants, the categories are not, in my opinion, arbitrary. I believe that a fine-grained predictor such as the mean value of age is not relevant to this pool of participants.
A sociophonetic analysis of trill production in Panamanian Spanish
considered higher frequency words. Also, frequency was only included in the model looking at normative/non-normative alternation. The results of the acoustic analysis and the three statistical regressions are presented in Section 4.
4. Results 4.1 Acoustic analysis The acoustic analysis reveals eight trill variants in PCS based on the four acoustic cues mentioned in the methodology (Section 3). Figure 1 shows an alveolar trill apparent in the spectrogram by three occlusions with very little presence of formants. In Figure 2, a single voiced tap is shown, indicated by an occlusion that contains formant structure. Figure 3 presents a trill produced as an approximant with periodic waves and formant structure apparent in the spectrogram. In Figure 4, the trill is produced as a sibilant-like sound with aperiodic waves and concentration of frication in the higher frequencies of the spectrogram. Figure 5 shows a post-approximantized tap where a short occlusion is followed by a segment of formant structure and periodic waves. A pre-assibilated tap is presented in Figure 6 where a segment of frication is visible in the spectrogram followed by an occlusion. Figure 7 shows the opposite sequence in a postassibilated tap. Finally, in Figure 8, a perceptual tap is apparent where there is a very small reduction in the waveform and clear formant structure, thus not distinguishing this variant from the surrounding vowels.
Frequency
5000
0 c 0
a
rr
e Time (s)
Figure 1. An alveolar trill with 3 occlusions in PCS: carrera ‘career’
r
a 0.747
Delano S. Lamy
Frequency
5000
0 el
a
0
rr
oh 0.355
Time (s)
Figure 2. A single voiced tap in PCS: arroz ‘rice’
Frequency
5000
0 bueno 0
r Time (s)
Figure 3. An approximantized trill in PCS: realmente ‘really’
ea
men(te) 0.598
A sociophonetic analysis of trill production in Panamanian Spanish
Frequency
5000
0 de
r
0
e
pente
Time (s)
0.744
Figure 4. An assibilated trill in PCS: de repente ‘suddenly’
Frequency (Hz)
5000
0 de 0
r
e
p
Time (s)
Figure 5. A post-approximantized tap in PCS: de repente ‘suddenly’
e
n
t(e) 0.507
Delano S. Lamy
Frequency (Hz)
5000
0 ca
rr
e
0
r
a 0.294
Time (s)
Figure 6. A pre-assibilated tap in PCS: carrera ‘career’
Frequency
5000
0 he 0
r
e Time (s)
c(ibido) 0.327
Figure 7. A post-assibilated tap in PCS: he recibido: ‘I have received’
According to this analysis, PCS shows a high amount of variation in trill production as has been observed in the related literature. This high variability has been explained on the basis of coarticulatory effects, aerodynamic instability, and sociolinguistic factors. Furthermore, the underlying mechanism for this variation could be a frequency effect, where high lexical frequency favors non-normative trills and low lexical frequency, normative trills. These claims will be further explored in Section 5.
A sociophonetic analysis of trill production in Panamanian Spanish
Frequency
5000
0 de
r
0
e
p
ente 0.562
Time (s)
Figure 8. A perceptual tap in PCS: de repente ‘suddenly’
Percentage
For the purposes of the present analysis, the eight variants were grouped together according to normative and non-normative production.3 Those variants exhibiting 2 – 3 occlusions were considered normative, and those with fewer than 2 occlusions were labeled non-normative. In Figure 9 it is observed that the normative alveolar trill is the most frequent variant in the sample (57%), which is not in line with claims that the alveolar trill is nearly non-existent in Spanish (Hammond, 1999). Additionally in Figure 10, we observe that the single most frequent variant in PCS is the trill with 2 occlusions (43.6%), followed by the single tap (32.7%). 100 90 80 70 60 50 40 30 20 10 0
57 43
Normative
Non-normative Variants
Figure 9. Percentage of normative versus non-normative in PCS
. Other researchers have used labels such as “trill-like” and “non-trill-like” for this categorization (see Díaz-Campos, 2008); however, the criterion is the same.
Delano S. Lamy 100 90 80 Percentage
70 60 50
43.6
40
32.7
30 20 10 0
13.5
10.2 0
1 2 Number of occlusions
3+
Figure 10. Number of occlusions of the trill variants in PCS
Furthermore, Table 2 presents a comparison between normative and non- normative variants in terms of duration and mean number of occlusions. As is expected due to the criterion for grouping the variants, a Mann-Whitney’s U-test revealed that normative variants have a significantly greater mean number of occlusions (2.29) than non-normative variants (0.77). Likewise, the average duration of the normative trills is significantly greater (78.84 ms) than that of the non-normative ones (60.40 ms). These findings would be expected because, according to Blecua (2001), the more components the segments are made of, the longer they will be. This pattern seems to hold true in the case of PCS trills. Table 2. Overall mean duration and mean occlusions of normative and non-normative variants Variants
Duration (SD)
Occlusions (SD)
Normative
78.84 (26.41)
2.29 (0.60)
Non-normative
60.40 (25.90)
0.77 (0.43)
U = 26537.5, p < 2.2e-16
U = 232, p < 2.2e-16
4.2 Variable rule analysis As I explained in Section 3, in order to determine which linguistic and extralinguistic factors play a significant role in trill production among PCS speakers, a multivariate analysis was carried out for normative/non-normative alternation, mean number of occlusions, and mean duration. The application value in the case of normative/ non-normative alternation was the normative production. The weight values tell us whether the factors favor or disfavor the use of the normative variant. With regard to
A sociophonetic analysis of trill production in Panamanian Spanish
mean occlusions and mean duration, the analysis estimates the effect that the factors have on the mean values. Positive coefficients tell us that greater means are favored by a particular factor. Table 3 presents the linguistic and extralinguistic factors that significantly condition variable trill production in PCS. We notice that the three statistical models select almost the same significant factor groups. With regard to the linguistic factors, preceding segment type, syllable stress, and number of syllables are selected in all of the analyses. The linear regression of mean duration is the only analysis that selected grammatical category as a significant factor group. Turning to the extralinguistic factors, we observe that both speaker age and speaker sex are selected as significant factor groups in all but one model, mean occlusions. Said analysis only selects speaker age. Level of education is selected in all three models. Despite the small differences, it can be concluded just based on the factor groups that normative/non-normative alternation, mean occlusions, and mean duration are related characteristics of the trill. In Section 4.2.1 and 4.2.2, I further explore the results of the constraint hierarchies of the factors within the factor groups mentioned. Table 3. Conditioning factors selected as significant for variable trill production in PCS Normative/Non-normative alternation
Mean occlusions
Mean duration
Preceding segment type
Preceding segment type
Preceding segment type
Syllable stress
Syllable stress
Syllable stress
Number of syllables
Number of syllables
Number of syllables
Linguistic factors
Frequency Grammatical category Extralinguistic factors Speaker age
Speaker age
Speaker sex Level of education
Speaker age Speaker sex
Level of education
Level of education
* Normative/Non-normative: deviance = 760.027, intercept = 0.052, Total N = 608 Mean occlusions: deviance = 449.894, intercept = 1.488, Total N = 608 Mean duration: deviance = 374441.2, intercept = 81.998, Total N = 599 (perceptual taps excluded)
4.2.1 Linguistic factors Table 4 presents the results for preceding segment type, in which we see that normative variants and a greater mean number of occlusions are favored when there is a preceding vowel or consonant, and are disfavored when nothing precedes the trill, that is, in absolute word-initial position. The pattern is different for mean duration, where a greater mean duration is favored with preceding consonants or in absolute
Delano S. Lamy
word-initial position. This particular finding leads us to believe that in terms of this factor group, mean duration is not as related to mean number of occlusions. Table 4. Variable trill production according to preceding segment N/NN
MO
MD
Preceding segment
FW
N
Coef
N
Coef
N
Vowel
0.64
508
0.298
508
–4.861
499
Consonant
0.62
81
0.198
81
4.427
81
Zero
0.26
19
0.496
19
0.434
19
Table 5 presents the results for syllable stress. We see that normative variants, greater mean occlusions, and greater mean duration are favored in stressed syllables, and are disfavored in unstressed syllables. This pattern seems to be in line with the previous literature concerned with the effect of syllable stress (Bradley & Willis, 2012; but see Henriksen & Willis, 2010). Table 5. Variable trill production according to syllable stress N/NN
MO
Syllable stress
FW
N
Tonic
0.57
Atonic
0.43
MD
Coef
N
Coef
N
179
0.159
429
–0.159
179
4.884
177
429
–4.884
422
Turning to number of syllables, in Table 6 we see that normative variants are favored with words that have one or two syllables, and slightly favored in words with three syllables. Furthermore, greater mean occlusions and greater mean duration are favored in words with one or two syllables. These means decrease as the words become longer, thus still following the order of effect observed in terms of normative/non-normative alternation. Díaz-Campos (2008) found the opposite effect in Venezuelan Spanish. Table 6. Variable trill production according to number of syllables N/NN
MO
Number of syllables
FW
N
1–2
0.61
3
0.55
4+
0.34
MD
Coef
N
Coef
N
146
0.219
242
–0.018
146
8.831
144
242
–0.640
239
220
–0.201
220
–8.192
216
A sociophonetic analysis of trill production in Panamanian Spanish
Factor weights
Finally, in terms of frequency, Figure 11 shows that lower frequency words favor normative variants, whereas higher frequency words favor non-normative variants. This pattern reveals that reduced forms are indeed related to higher frequency of use (see Bybee, 2001). In Section 4.2.2, I present the extralinguistic factors. 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0.604 0.396
low
high Frequency
Figure 11. Normative/non-normative alternation according to frequency
4.2.2 Extralinguistic factors Of the four extralinguistic factors included in the statistical models, three were relevant in variable trill production – speaker age, speaker sex and level of education. Table 7 presents the results for speaker age, where speakers between the ages of 26 and 37 favor normative variants and higher mean occlusions. Speakers between the ages of 14 and 19, and those 53 and older favor non-normative variants and lower mean occlusions. In terms of mean duration, the middle group still differs from the two extremes:. they favor longer segments, while the younger and older groups favor shorter segments. However, the older group has a positive coefficient, thus revealing that they slightly favor a greater mean duration. Table 7. Variable trill production according to speaker age N/NN
MO
MD
Speaker age
FW
N
Coef
N
Coef
N
14–19
0.40
159
–0.138
195
–6.700
192
26–37
0.60
254
0.224
159
6.409
158
53+
0.49
195
–0.085
254
0.291
249
With regard to speaker sex, Table 8 shows that men favor normative variants and a greater mean duration, whereas women favor non-normative variants and a lower mean duration. As mentioned above, speaker sex is not a significant factor group in terms of mean number of occlusions. Since it has been observed in sociolinguistic literature that different speech communities may treat variables in different ways, it is
Delano S. Lamy
not surprising that the patterns found for speaker age and sex in the present analysis differ from the studies previously mentioned. Table 8. Variable trill production according to speaker sex N/NN
MO
MD
Speaker sex
FW
N
Coef
N
Coef
N
Male
0.58
244
–
–
4.783
241
Female
0.42
364
–
–
–4.783
358
Finally, Table 9 presents the results for level of education where we observe that speakers with a university education favor normative variants and higher mean occlusions, and those with only a high school education favor non-normative variants and lower mean occlusions. Interestingly, in terms of mean duration, there is an opposite effect, where speakers with a university education favor a lower mean duration, and those with only a high school education favor a higher mean duration. These findings will be further discussed along with the linguistic factor results in Section 5. Table 9. Variable trill production according to level of education N/NN
MO
MD
Deviance
767.329
458.528
382618.8
Intercept
–0.025
1.499
80.394
Total N
608
608
599
Level of education
FW
N
Coef
N
Coef
N
University
0.56
250
0.083
250
–2.877
352
High School
0.44
358
–0.083
358
2.877
247
5. Discussion In this section, I discuss the results presented in Section 4 in light of previous findings, and compare the patterns found in PCS to those of other varieties. The acoustic analysis revealed eight variants of the trill, of which the normative alveolar trill was the most frequent. This finding goes against Hammond’s (1999) claim that the alveolar trill is nearly non-existent in Spanish. In his analysis of several Spanish and Latin American varieties, he found that the alveolar trill made up only 0.998% of the sample (p. 142). In the present analysis, the distribution of the alveolar trill is significantly higher (57%). Additionally, in Bradley & Willis’s (2012) study of Veracruz Mexican
A sociophonetic analysis of trill production in Panamanian Spanish
Spanish, although not the most frequent, the alveolar trill still made up 49% of the sample (p. 62). It is important to note that Hammond’s (1999) criterion for a trill is more restricted; a trill was a variant containing at least three occlusions. However, in the present analysis and in Bradley & Willis (2012), a variant is considered a normative trill when produced with at least two occlusions. Nonetheless, we notice that, with a percentage of 13.5, the group of three occlusions or more is still more numerous than what Hammond (1999) found. Therefore, based on the present study, I cannot claim that the alveolar trill is non-existent in Spanish. When comparing normative to non-normative production, I found that there is a significant difference with regard to segment duration and mean number of occlusions. Blecua (2001) showed in her analysis of Peninsular Spanish that there is a correlation between the number of components of a trill and its duration. That is, the more components the segment is made of, the longer it will be (Blecua, 2001:22). This seems to be the pattern observed in PCS. The non-normative variants have either zero or one occlusion, thus the segment of those variants is shorter than those with three or more occlusions. The three multivariate analyses show that both linguistic and extralinguistic factors condition trill production. I analyzed the effect of the factors on three dependent variables: normative/non-normative alternation, mean number of occlusions, and mean duration, and found that some factor groups are significant in all three analyses, namely, preceding segment type, syllable stress, number of syllables, speaker age, speaker sex and level of education. One explanation for this finding is that these dependent variables are related in some way. The normative/non-normative category is based on the number of occlusions (see Methodology), and there is also a correlation between number of occlusions and segment duration. Therefore, we should expect to find similar results with regard to factor effects. The relationship between these three variables is further supported when observing the order of effect within the factor groups. In each factor group, the constraint hierarchies are the same, with the exception of preceding segment type and level of education in mean duration. Focusing on the effect of the linguistic predictors, we see that prosodic factors, such as syllable stress and number of syllables, significantly constrain trill production. Normative variants, greater mean occlusions, and greater mean duration are favored in stressed syllables, similar to varieties such as Veracruz Mexican Spanish (Bradley & Willis, 2012). However, in Jerezano Andalusian Spanish, the mean number of occlusions was greater in unstressed syllables (Henriksen & Willis, 2010:121). It would be expected that stressed syllables favor more trill-like variants due to a higher degree of oropharyngeal pressure when producing the phone in this context. However, it seems that aerodynamic events could be affected by other factors, such as lexical frequency, which will be discussed below. Furthermore, we see that words containing 1–3 syllables favor normative variants, greater mean occlusions and greater mean duration.
Delano S. Lamy
An opposite effect was found by Díaz-Campos (2008) in Venezuelan Spanish. He explained that lexical diffusion and frequency effects could be the reason for this distribution, which is why we find a different distribution in PCS. That is, there is a gradual change occurring with regard to trill production, and PCS and Venezuelan Spanish are at different stages in said change. As will be discussed, one way to investigate change is by looking at frequency of use. Phonetic context also seems to affect trill production, as is seen in the results for preceding segment type. Only two previous analyses of trills found effects of surrounding sounds. Lewis (2004) found that the degree of constriction of the previous consonant brought about different realizations of the trill in Spanish. Recasens (1991) and Recasens & Pallerès (1999) found that surrounding vowels slightly affected trill production in Catalan. In the present analysis, preceding segment type was a significant factor group, and within that group, vowels and consonants favored normative variants and greater mean occlusions. In order to determine what was causing this effect, I looked at the distribution of surrounding sounds in terms of place of articulation of the preceding and following segment, that is, whether the sound is produced more towards the posterior of the buccal cavity as opposed to the front. Tables 10 and 11 present these distributions. Table 10. Distribution of normative variant according to place of preceding segment Place of preceding segment
N
%
glottal
22
73
back
110
69
central
210
66
alveolar
59
53
front
188
44
zero
19
21
Table 11. Distribution of normative variant according to following vowel Place of following vowel
N
%
central
82
73
back
83
69
front
443
52
What we see is that the percentages of normative variants are higher with posterior sounds (i.e. /h/ and /o, u/), which is in line with Recasens’s (1991:274) explanation
A sociophonetic analysis of trill production in Panamanian Spanish
for Catalan. That is, when the predorsum and the mediodorsum of the tongue are lowered and the postdorsum is retracted to the rear wall of the upper pharynx, apical closure is facilitated and thus, multiple vibrations are easier to produce. In other words, tongue dorsum raising, which occurs in front/high vowels (/e, i/), is incompatible with apical closure (Recasens, 1991:274). Moving on to the extralinguistic factors, we have noted that speaker age significantly affects trill production, in that speakers between the ages of 26 and 37 favor normative variants, greater mean occlusions, and greater mean duration, and speakers between 14 and 19 and 53 and older favor non-normative variants and lower mean occlusions. That is to say, the middle group differs from the two extremes. Previous findings report that younger speakers always differ from older speakers, whether they favor or disfavor normative variants (Díaz-Campos, 2008; Henriksen & Willis, 2010; Lastra & Martín Butragueño, 2006). The present analysis differs from the previous ones in this respect due to the effect of the linguistic market commonly found in the sociolinguistic literature. That is, speakers find a need to conform to more standard speech due to their economic activity (Bourdieu, 1991; Sankoff & Laberge, 1978:239). This effect seems to be the case in PCS, at least with regard to the present pool of participants. The middle group consists of speakers who have regular employment or attend college, whereas the younger group is made up of adolescents and young adults without employment or just starting college. The older group consists of speakers who are retired. Therefore, the participants in the extreme groups have less pressure to produce standard speech. The results of speaker sex show that men favor normative variants and greater mean duration, whereas women favor non-normative variants and lower mean duration. The patterns of this social factor also differ from previous findings (Bradley & Willis, 2012; Díaz-Campos, 2008; Henriksen & Willis, 2010; but see Lastra & Martín Butragueño, 2006). Although women are normally found to have more conservative speech than men, so much so that this has been termed the “sociolinguistic gender pattern” (Fasold, 1990:92), it is not uncommon for women to favor innovative variants (Fontanella de Weinberg, 1979; Rissel, 1989). This pattern usually occurs in situations of change from below and when the innovative variant is not stigmatized (Labov, 1990). Therefore, it could be the case in Panama that non-normative realizations of the trill do not carry any negative weight, and so women may use them as an indicator of identity. It was also observed in the present analysis that level of education had a significant effect on trill production. In an attempt to avoid interaction in the statistical analyses, speaker age and social class were excluded when analyzing the effect of this factor group. The patterns show that speakers with a university education favored normative variants and higher mean occlusions, whereas those with only a high school education favored non-normative variants and lower mean occlusions. However, in terms of
Delano S. Lamy
mean duration, speakers with a university education favored a lower mean duration, and those with only a high school education favored a higher mean duration. The reason for this opposite effect is not clear; however, I suspect this difference emerges because level of education is not a straightforward factor among these participants. That is, although some speakers did not receive a university education, they still completed post-secondary courses (simply referred to as “cursos” in Panama) that are either required by their employers or thought to be necessary to obtain employment. This is the case for Magaly and Leo. We see that Valeria did not complete high school but did complete “cursos” for work. Thus, it appears that age is a better predictor for trill production from which we can extrapolate effects of education and the workforce as discussed above. Finally, I turn to the results for frequency effects. As frequency can significantly interact with other independent factors (see Erker & Guy, 2012), it was first analyzed alone to determine its effect on normative/non-normative alternation. The results revealed that higher frequency words favored non-normative variants. Subsequently, the factor was included with the other linguistic predictors in the multivariate analysis. The program rendered the same results; higher frequency words favored non- normative trills (see Figure 11). Since non-normative trills are reduced realizations, it appears that variable trill production can generally be explained within a usagebased model of language. That is, higher frequency of use brings about more reduced exemplars of a linguistic form (Bybee, 2001). Frequency of occurrence is a measure of predictability of a word in discourse, where highly frequent words are more predictable in speech, and thus, are produced more quickly. This faster rate of production brings about more acoustically reduced forms of a sound (Ernestus, 2014). Furthermore, because of their high frequency of occurrence, reduced forms become more entrenched and active in memory, which promotes diffusion (Langacker, 2000). Frequency effects show gradiency, and thus phonological reduction will appear at first in certain words, and then diffuse slowly through the lexicon. The present sample shows that non-normative variants appear more in adverbs, such as de repente ‘suddenly/ maybe’ and realmente ‘really’, which are very frequent (see Section 3). They appear less in infrequent nouns such as chatarra ‘junk’ and infrequent verbs and verbal forms, such as aferrar ‘to hold on to’ or querramos ‘want- sbjv-1pl’. Further analyses with different types of frequency counts and a larger corpus are needed in order to shed more light on the effects of frequency on trill production.4
. Although not a focus of the main analysis, the effect of speech rate was explored in simple linear regressions. The results revealed that a faster speech rate does indeed favor reduced forms (normative/non-normative: +1, -0.21; mean occlusions: +1, -0.087; mean duration:
A sociophonetic analysis of trill production in Panamanian Spanish
6. Conclusions The aim of the present study was to describe variable trill production in Panama City Spanish within a variationist sociolinguistic framework. The acoustic analysis revealed several variants, which were grouped according to normative and non-normative realizations. Normative alveolar trills were the most frequent among the 11 participants of the study. Furthermore, they differed significantly from non-normative trills in terms of mean number of occlusions and mean duration. The study also included multivariate analyses with three different variables: normative/non-normative alternation, mean number of occlusions, and mean duration. The statistical regressions showed that trill production was significantly constrained by linguistic predictors, such as preceding segment type, syllable stress, number of syllables, and grammatical category. Extralinguistic predictors, such as speaker age and sex also played an important role in trill production. Additionally, this study explored claims made about frequency effects in usage-based models of language, and more specifically, of phonology. In said models, it is hypothesized that higher frequency of use brings about phonological reduction. This claim held true in the analysis of trill production in PCS; non-normative variants were favored in higher frequency words. Further analyses on variable trill production should take into account frequency effects and other factors involved in language use, such as speech rate, as these could have important implications for language theory and sociophonetic analyses in the Spanish-speaking world.
References Alvarado de Ricord, E. (1971). El español de Panamá: Estudio fonético y fonológico. Panamá: Editorial Universitaria. Blecua, B. (2001). Las vibrantes del español: Manifestaciones acústicas y procesos fonéticos. Unpublished Doctoral Dissertation. Universidad Autónoma de Barcelona. Boersma, P., & Weenink, D. (2012). Praat: Doing phonetics by computer (Version 5.3.04). Retrieved from 〈http://www.praat.org/〉 Bourdieu, P. (1991). Language and symbolic power. Ed. by J. B. Thompson. Trans. by G. Raymond and M. Adamson. Cambridge, MA: Harvard University Press. DOI: 10.1017/s004740450001705x Bradley, T.G. & Willis, E.W. (2012). Rhotic variation and contrast in Veracruz Mexican Spanish. Estudios de Fonética Experimental, 21, 43–74. Broce, M. & Torres Cacoullos, R. (2002). ‘Dialectología urbana’ rural: La estratificación social de (r) y (l) en Coclé, Panamá. Hispania, 85(2), 342–354. DOI: 10.2307/4141096
+1, –4.905), thus revealing the importance of usage-based factors in the study of acoustic phonetics.
Delano S. Lamy Bybee, J. (2000). The phonology of the lexicon: evidence from lexical diffusion. In M. Barlow & S. Kemmer (eds.), Usage based models of language (pp. 65–85). Stanford, CA: CSLI. Bybee, J. (2001). Phonology and language use. Cambridge: Cambridge University Press. DOI: 10.1017/s0952675703214445 Bybee, J., & Torres Cacoullos, R. (2008). Phonological and grammatical variation in exemplar models. Studies in Hispanic and Lusophone Linguistics, 1, 399–413. Cedergren, H.J., Rousseau, P., & Sankoff, D. (1986). La variabilidad de /r/ implosiva en el español de Panamá y los modelos de ordenación de reglas. In R. Núñez Cedeño, I. Páez Urdaneta, & L. Guitart (Eds.), Estudios sobre la fonología del español del Caribe (pp. 13–20). Caracas: Ediciones La Casa de Bello. Cedergren, H.J., & Sankoff, D. (1974). Variable rules: Performance as a statistical reflection of competence. Language, 50, 333–355. DOI: 10.2307/412441 Colantoni, L. (2006a). Increasing periodicity to reduce similarity: An acoustic account of deassibilation in rhotics. In M. Díaz-Campos (Ed.), Selected Proceedings of the 2nd Conference on Laboratory Approaches to Spanish Phonetics and Phonology (pp. 22–34). Somerville, MA: Cascadilla Proceedings Project. Colantoni, L. (2006b). Micro and macro sound variation and change in Argentine Spanish. In N. Sagarra & A.J. Toribio (Eds.), Selected Proceedings of the 9th Hispanic Linguistic Symposium (pp. 91–102). Somerville, MA: Cascadilla Proceedings Project. Díaz-Campos, M. (2008). Variable production of the trill in spontaneous speech: Sociolinguistic implications. In L. Colantoni & J. Steele (Eds.), Selected Proceedings of the 3rd Conference on Laboratory Approaches to Spanish Phonology (pp. 47–58). Somerville, MA: Cascadilla Proceedings Project. Erker, D. & Guy, G.R. (2012). The role of lexical frequency in syntactic variability: Variable subject personal pronoun expression in Spanish. Language, 88(3), 526–557. DOI: 10.1353/lan.2012.0050 Ernestus, M. (2014). Acoustic reduction and the roles of abstractions and exemplars in speech processing. Lingua, 142, 27–41. DOI: 10.1016/j.lingua.2012.12.006 Fasold, R. (1990). The sociolinguistics of language. Oxford: Blackwell. DOI: 10.1017/s0272263100010330 Fontanella de Weinberg, M.B. (1979). Dinámica social de un cambio lingüístico. La reestructuración de las palatales en el español bonaerense. Mexico City: Universidad Nacional Autónoma de México. DOI: 10.1017/s004740450000943x Hammond, R. (1999). On the non-occurrence of the phone [r] in the Spanish sound system. In J. Gutiérrez-Rexach & F. Martínez-Gil (Eds.), Advances in Hispanic linguistics, (pp. 135–151). Somerville, MA: Cascadilla Press. Henriksen, N. & Willis, E. (2010). Acoustic characterization of phonemic trill production in Jerezano Andalusian Spanish. In M. Ortega-Llebaria (Ed.), Selected Proceedings of the 4th Conference on Laboratory Approaches to Spanish Phonology (pp. 115–127). Somerville, MA: Cascadilla Proceedings Project. Hualde, J.I. (2005). The sounds of Spanish. Cambridge: Cambridge University Press. Johnson, D.E. (2009). Getting off the GoldVarb standard: Introducing Rbrul for mixed-effects variable rule analysis. Language and Linguistics Compass, 3, 359–383. DOI: 10.1111/j.1749-818x.2008.00108.x Labov, W. (1990). The intersection of sex and social class in the course of linguistic change. Language Variation and Change, 2, 205–254. DOI: 10.1017/s0954394500000338
A sociophonetic analysis of trill production in Panamanian Spanish
Langacker, R.W. (2000). A dynamic usage-based model. In M. Barlow & S. Kemmer (Eds.), Usage-based models of language (pp. 1–63). Stanford, CA: CSLI. Lastra, Y.P., & Martín Butragueño, P. (2006). Un posible cambio en curso: El caso de las vibrantes en la ciudad de México. In A.M. Cestero Mancera (Ed.), Estudios sociolingüísticos del español de España y América. Madrid: Arco Libros. Lewis, A. (2004). Coarticulatory effects on Spanish trill production. In A. Agwuele, W. Warren, & S–H. Park (Eds.), Proceedings of the 2003 Texas Linguistics Society Conference (pp. 116–127). Somerville, MA: Cascadilla Proceedings Project. Martínez Celdrán, E. (1997). El mecanismo de producción de la vibrante apical múltiple. Estudios de Fonética Experimental, 8, 85–97. Mowrey, R., & Pagliuca, W. (1995). The reductive character of articulatory evolution. Rivista de Linguistica, 7(1), 37–124. Pierrehumbert, J.B. (2001). Exemplar dynamics: Word frequency, lenition and contrast. In J. Bybee, & P. Hopper (Eds.), Frequency and the emergence of linguistic structure (pp. 137–157). Amsterdam: John Benjamins. DOI: 10.1075/tsl.45.08pie Recasens, D. (1991). On the production characteristics of apicoalveolar taps and trills. Journal of Phonetics, 19, 267–280. Recasens, D. & Pallarès, M.D. (1999). A study of /ɾ/ and /r/ in the light of the “DAC” coarticulation model. Journal of Phonetics, 27, 143–169. DOI: 10.1006/jpho.1999.0092 Rissel, D. (1989). Sex, attitudes, and the assibilation of /r/ among young people in San Luis de Potosi, Mexico. Language Variation and Change, 1, 269–83. DOI: 10.1017/s0954394500000181 Sankoff, D. & Laberge, S. (1978). The linguistic market and the statistical explanation of variability. In D. Sankoff (Ed.), Linguistic variation: Models and methods (pp. 239–250). New York, NY: Academic Press. DOI: 10.1017/s0047404500008472 Solé, M–J. (2002). Aerodynamic characteristics of trills and phonological patterning. Journal of Phonetics, 30, 655–688. DOI: 10.1006/jpho.2002.0179 Tagliamonte, S. (2006). Analysing sociolinguistic variation. Cambridge: Cambridge University Press. DOI: 10.1017/cbo9780511801624 Vásquez Carranza, L.M. (2006). On the phonetic realization and distribution of Costa Rican rhotics. Filología y Lingüística, 32(2), 291–306. Willis, E.W. (2006). Trill variation in Dominican Spanish: An acoustic examination and comparative analysis. In N. Sagarra & A.J. Toribio (eds.), Selected Proceedings of the 9th Hispanic Linguistics Symposium (pp.121–131). Somerville, MA: Cascadilla Proceedings Project. Willis, E.W. (2007). An acoustic study of the “pre-aspirated trill” in Narrative Cibaeño Dominican Spanish. Journal of the International Phonetic Association, 37, 33–49. DOI: 10.1017/s0025100306002799 Willis, E.W., & Bradley, T.G. (2008). Contrast maintenance of taps and trills in Dominican Spanish: Data and analysis. In L. Colantoni & J. Steele (eds.), Selected Proceedings of the 3rd Conference on Laboratory Approaches to Spanish Phonology (pp. 87–100). S omerville, MA: Cascadilla Proceedings Project.
Index A accomplishment xii, xv, 49–50, 55–56, 62–64, 67, 205, 213, 215, 220, 221–222, 224, 226 accusative xiv, 150–151, 153–155, 158–159, 162, 164 achievements 54, 56, 205, 213, 215–216, 220, 222 acoustics 309, 313 activities 54–55, 195, 205, 215, 220, 224 adjectival intensifier xv, xvi, 233, 235–236 agreement xiv, 8, 12–13, 25, 33, 35, 47, 93, 98, 148–149, 152, 166–167, 169–170, 172, 176–178, 186–187 alveolar trill 298, 313–314, 319, 323, 328–329 analytic passives xii, 89, 99 Argentina xvi, 165, 197, 199, 269–270, 272–273, 279, 287–288, 291–292 aspect xii, xiv, xv, 49–51, 54–56, 67, 68, 86–88, 93, 98, 106, 154–155, 159, 163–164, 187, 191–193, 195, 205, 208, 209–217, 218, 221–223, 224, 225–229, 246–249, 280, 286 assimilation xiii, xiv, 127–131, 133, 137–142 attrition xv, 211–213, 216–217, 219, 224–227 auditory enhancement 113 B bare nominal xiv, 169 see also bare NP xiv, 169–170, 173–174, 177–179, 181, 184–185, 187 biclausal xii, 50–53, 67 binding xi, 3, 19, 25–26, 39, 43–44, 71 C CHILDES xii, xiv, xv, 87, 147–148, 153, 166, 198, 210
clitic xiv, 32, 44, 100, 147–156, 158–166, 217 clitic climbing 149, 160, 163, 217 clitic cluster xiv, 147–150, 152, 155–156, 159–161, 165 coda xvi, 251, 251–255, 257, 259, 262, 316 complement clauses xii, 69–71, 73, 75–76, 81, 83–86 compositional aspect 215 concord vii, xiv, 169–170, 185–187 Control xi, 25–26, 28–30, 33–35, 37–39, 43, 45, 46, 87 conversational implicatures xii, 49–50, 53, 62–64, 66–67 Córdoba xvi, xvii, 269–270, 272–278, 280–289, 292 crosslinguistic influence vii, xv, 211–213, 217–218, 225–228 D dative xiv, 147, 150–151, 153–155, 158, 162, 164–165 Determiner Phrase 23, 189 see also DP xi, xiv, 16, 26–27, 29, 31–32, 38, 39, 40–42, 94–95, 154, 156, 164, 169–174, 178–180, 181–187 dialect identification xvi, xvii, 269–270, 273, 277–282, 285–287, 289 distributed morphology xiii, 26, 36, 89–90, 111 ditransitive xiv, 154 Dominican Republic 251 doxastic xii, 71, 74, 76–78, 86 E e-networks xvi, 233, 233, 237, 242–244, 248 elsewhere morpheme xiii, 89, 103–104 enclitics 149, 160 etymological [s] xvi
F f0 xiii, 113–124, 271, 308 featural reassembly 212 features 3, 13, 25, 32–36, 38–39, 42, 67, 71, 90–91, 148, 150–151, 158, 164, 170, 172, 177–178, 194, 205, 209, 212, 233, 269–272, 275–276, 285–289, 291, 294 focus xi, 5, 9, 12, 23, 26, 32–33, 37–38, 41, 47, 69–70, 74, 148, 188, 191, 198, 252, 255, 257, 274, 315–316 Force Phrase xi, 12, 16, 18, 20 G gender feature xv, 171, 177–178, 187, 189 gender marking xiv, 169–170, 179, 181–183, 185–186 gender morphology 170, 176, 178 gestural blending 127, 129, 138–141 grammatical aspect 214–215, 217 grammatical gender xiv, 171, 176, 188 H Heritage Spanish xv, 211–212, 217 hipercorrección 251 hoarse voice 293, 308 hypercorrection xvi, 251–254, 256, 262, 264, 266 see also qualitative hypercorrection xvi, 252–254, 262, 264 quantitative hypercorrection xvi, 251–253 I imperfect xv, 85, 196, 201–203, 211–218, 220–226, 315 implicature 62–66
Index inchoative xiv, 96, 98, 102, 154–155, 163–164 incomplete acquisition xv, 212–213, 216–217, 226 infinitive xi, 25–26, 28–35, 37, 38–41, 43, 46, 71, 73, 149, 160 initial stage 86, 152, 239 intensifier xv, xvi, 154, 233, 235–240, 243, 245, 248–249 interface xi, xiii, 14, 22, 35–36, 38, 39, 43–44, 95–96, 109–111, 291 Intervention Effects xi, 3–6, 9–16, 18–22 intonational contour 269, 272–273, 275 intrusion 251, 253–254, 256, 266 intrusive [s] xvi, 251–256, 258, 260, 262–266 J Just Noticeable Differences 301 see also JND 301, 310 L L1 acquisition xiv, xv, 147, 191–192, 196–197, 208, 226 L1 attrition 211–213, 216–217, 219, 224–227 L2 acquisition vii, xiv, 169–172, 177, 187, 227–228 language dominance 211, 213, 217–218, 221, 225 late insertion 26, 35–36, 42, 91, 93 learnability 148–149 lexical aspect 214 Linear Discriminant Analysis 300 see also LDA 300–301 logistic deviance 113, 120 M minimalism 3, 111 modal speaker 297 modal voice 293–296, 299, 302, 307–309 monoclausal xii, 50–55, 57, 61, 65, 67, 194 N nanosyntax xiii, 109 nominative xi, 25, 28–29, 40
non-modal speaker 297 null subjects xi, 25, 35, 37, 43 O occlusions xvii, xviii, 313–319, 323–331, 333 onset f0 xiii, 113–124 overt pronouns xi, 39 overt subjects xi, 25–26, 28–29, 31–34, 38–41 overt/covert alternation xi, 26, 33, 35–36, 43 P Panama xvii, xviii, 313–314, 316–317, 331–333 participial adjective 246–248 Person-Case Constraint 150, 162 see also PCC 150–151 phonation type 293–294, 298, 308 pluractionality 49–50, 61 pre-tonic vowel xvii, 269–271, 273, 275–278, 281, 285–287, 289 present perfect xv, 160, 191–192, 199, 203–204, 206 preterit xv, 191, 193–203, 208–209, 211–218, 220–226 pro xi, 25–28, 32–35, 39, 43, 45, 257 pro-drop xi, 25–26, 33–35, 39, 43 proclitics 149, 160 prosodic boundaries xiii, 127, 129–130, 141 prosody perception 269 Q q-particle 3, 12–13, 16–22 R R-expressions xi, 29–31, 39–42 reflexive xiv, 100, 150–152, 154, 164 resultative xvi, 233, 246–247, 249 Romance languages 25, 33, 67, 76, 142, 189, 227, 250, 290 root allomorphy xiii, 89–90, 92, 108 Russian xiv, 169–170, 173–179, 181, 185–187, 189
S s-deletion 251 semantic lexical restrictions xii sequence of tenses xii, 69, 74, 77–78 sibilant voicing xiii, 127–128, 137 sociophonetics 310–311, 313 spectral xiii, 113–114, 117–119, 122, 124, 293–295, 298, 300, 307–308 spectral processing 113 states xiii, 55, 195, 205, 211, 213, 215, 217–220, 222, 267, 293 stative xvi, 54–56, 80, 195, 200, 213, 215–216, 221, 224–226, 233, 246–249 stop consonant xiii, 113–115, 117, 254 see also stop consonant voicing xiii, 113–115, 117 stress-shift 270 Superiority Effects xi, 3–13, 19 suppletion xii, xiii, 89–98, 101, 103, 108–109, 111 syllable stress xviii, 315, 318, 325–326, 329, 333 T telic 154, 193, 213, 215, 218 omission xiv, 152, 152–154, 158–159, 162–163 temporal xii, xiii, xv, 55, 57–58, 62, 69–74, 82–86, 113–114, 117–119, 122, 124, 129–130, 158, 160, 192–195, 197–201, 203, 205–207, 214–215, 223, 316 temporal adverbials 198, 200–201, 205–206 temporal dependence xii, 69–73, 82, 84–86 tense xv, 28, 43, 69–76, 79–87, 89, 92–93, 98, 176, 191–192, 194, 196–199, 202–203, 205, 208–217, 220, 222, 224–226, 228, 294, 307–308 tonada cordobesa 269–271, 274, 289 trill xvii, xviii, 298, 313–333 U ultracorrección 251–252
Index unmarked form xv, 215, 225 V valuation xi, 34, 36–38, 43 variable trill production xvii, xviii, 313, 315, 325–328, 332–333 variation viii, xi, xv, 3, 14, 22–23, 87, 89, 96, 133, 135, 138, 142, 149, 164, 176, 199, 210, 216, 231, 233, 237–238, 248, 252, 266, 272–274, 276, 290, 293–294, 296, 304–305, 310–311, 313, 315–316, 322, 334
variationist xvii, 201, 233, 237, 249–250, 252, 262, 313, 316, 318, 333 voice onset time xiii, 113 see also VOT xiii, 113–124 voice quality xvii, 293–296, 298–300, 302, 307–310 voiceless xiii, xvi, 114–118, 123–124, 127–129, 131, 134–136, 139, 143, 254–257, 260, 262–264, 294, 314–315 voiceless stop 128, 135, 254–257, 260, 264 voicing xiii, 113–118, 120, 122–124, 127–131, 133–142, 299
volition xii, 71, 76–77, 79–80, 86 vowel lengthening xvi, xvii, 269–271, 273, 275, 287, 289 W waveform amplitude reduction xvii, 317 wh-phrase xi, 3–5, 8-9, 10–17, 19–21, 22 wh-variable 3, 17, 22 word marker xv, 177–178, 186–187 word order 23, 41, 47, 148, 170, 174, 178–179, 184–185, 187, 189