PIRE undergraduate fellow Delaney Wilson presented at the Annual Meeting of the Psychonomic Society.
Studies examining voice onset time (VOT) in language-switching found that Spanish-English bilinguals have longer VOTs when naming isolated pictures, (Goldrick et al., 2014; Olson, 2013) indicating that switching impacts phonetic output. These studies all tested habitual code-switchers. To understand how this manifests in non-habitual code-switchers, we employed a sentence creation task focusing on word-initial /p/ and /t/ phonemes. Bilinguals produced sentences that switched from Spanish to English, from English to Spanish, or unilingual Spanish/English sentences. Critical target words appeared before or after the switch. English VOTs were longer than Spanish VOTs, in unilingual and code-switched sentences. VOTs in code-switched sentences and unilingual sentences were not statistically different, indicating that bilinguals maintain phonetic distinctions in code-switched sentences. This differs from the Goldrick at al. and Olson studies, and from spontaneous speech analyses (Arvaniti et al 2015; Fricke et al., 2015), suggesting that mechanisms of phonological convergence manifest differently in non-habitual code-switchers.
PIRE undergraduate fellow Delaney Wilson presented at the Current Approaches to Spanish and Portuguese Second Language Phonology Meeting.
Code-switching, alternating between multiple languages in an utterance, is a hallmark of bilingualism and shows that bilinguals keep both languages active. Code-switching is a popular topic, but most studies focus on switching between unrelated words/pictures rather than intrasentential switching, i.e. “the boy ate la pera en la cocina”1 Previous studies have shown that Spanish-English bilinguals have significantly longer voice onset times (VOTs) while naming unrelated pictures during switched trials2 and that their language-switched speech was “more accented”3 indicating that switching induces a difference in phonetic output. Since both of these studies focused on language switching during single picture naming, we studied whether the differences in VOT length are also present in more naturalistic code-switched sentences.VOT is the length of time between the release of a stop consonant and the beginning of the voicing of the following vowel. Languages differ in the types of VOT they use to implement contrasts, i.e. English is a long-lag language, meaning that the VOTs of voiceless stops are longer than the VOTs of Spanish. The interaction of the two phonologies potentially leads to a transfer of VOT patterns in code-switched speech. The present study is an acoustic analysis of voiceless stop VOTs (focusing on word-initial /p/ and /t/) in switched and non-switched sentences. The outcomes provide insight into how a meaningful sentence context may affect the phonetic realization of code-switched words. Kootstra et al. (2012)'s experimental technique4 was adapted to elicit code-switched sentences. Pictures contained two images (the subject and object in the target sentence), each circled in a separate color for a codeswitched sentence (see Figure 1). The colors indicated in which language each should be referenced. There were four language conditions: a complete English/Spanish sentence (unilingual conditions) or a switched sentence from English to Spanish or vice versa. Target words could appear before or after the language switch. Single-language sentences were presented in separate blocks and were not color-coded. Twenty-four Spanish-English bilinguals were tested at the University of Granada. 11 participants were analyzed based on achieved scores for the MELICET and DELE two tests that measure the level of proficiency in English and Spanish respectively. Forty-eight tokens were selected (2 languages*2 consonants*12 words). A total of 6,336 tokens (48 target words*2 conditions (unilingual, code-switch)*2 positions (before, after switch)*3 repetitions*11 participants) were collected. Currently, 3,168 tokens (all /t/ initial words) have been analyzed. Preliminary results for /t/ show there is a significant difference between English and Spanish VOTs in the unilingual conditions. As expected, English VOTs are longer than Spanish VOTs. Additionally, there is a significant difference in the VOT of switched versus non-switched sentences in Spanish (see Figure 2). While this suggests that there is no tangible anticipatory or carryover effect on VOT during code-switched sentences in naturalistic speech, it does show that bilinguals are capable of producing a distinction between the differences in VOT length within languages. Final analysis will also include /p/ initial words to ascertain if there is a similar effect for a different consonant.
PIRE fellow Anne Beatty-Martinez presents at the International Symposium on Bilingual Processing in Adults and Children (ISBPAC).
Linguistic prediction allows individuals to link what comprehenders expect to hear with what speakers actually say. The Production-Distribution-Comprehension model (PDC; Gennari & MacDonald, 2009) states that listeners’ and readers’ sensitivity to distributional patterns in production can constrain the comprehension system. The PDC model inherently predicts variability among speakers who codeswitch and speakers who don’t codeswitch. We test this hypothesis at the neural level by examining the relationship between codeswitching production patterns and comprehension difficulty. Because exposure to codeswitching is predicted to impact comprehension, we recruited Spanish-English codeswitchers in the US and non-codeswitchers in Spain. We focus on the production and comprehension of mixed noun phrases (mixed NPs; e.g., elM forkM). Evidence from naturalistic corpora suggests that bilinguals exhibit an overall preference for the masculine determiner (el), regardless of the noun’s gender in Spanish (Otheguy & Lapidus, 2003). In contrast, switches involving the feminine determiner (la) occur less frequently and are restricted to English nouns that are feminine in Spanish. To illustrate, el-codeswitches such as (elM forkM) and (elM spoonF) are extremely common. To a lesser extent, la-codeswitches involving feminine nouns (laF spoonF) have also been attested in bilingual speech, while those involving masculine nouns (*laF forkM) have not. We predicted that neurophysiological responses would reflect this asymmetry in codeswitchers but not in non-codeswitchers. Production was examined in a corpus of unscripted, task-oriented dialogues between two bilingual speakers (Anderson et al. 1991); comprehension was examined in sentential contexts using event-related potentials. As shown in Table 1, preliminary results for production show that codeswitchers (N=11) follow this pattern, while non-codeswitchers (N=11) do not. That is, a gender asymmetry in mixed NPs was found only in the bilingual speech of our participants tested in the US. For bilinguals who codeswitch, switches consistent with attested distributional patterns were easier to process. This is indicated by the N400 response for less frequent compared to more frequent switches. Crucially, no such effects were observed in non-codeswitchers. Our data suggest that individuals are highly sensitive to the constraints of their interactional context. Further research will look into how preferred and frequent switches compare to non-switched stimuli in these two groups of bilinguals. In this way, converging evidence from codeswitching sheds light on how production choices speakers make can predict comprehension performance.
PIRE fellow Anne Beatty-Martinez presents at the 57th Meeting of the Psychonomics Society.
Bilinguals are capable of efficiently negotiating task demands, especially when the task requires engagement of cognitive control. For example, bilinguals have been shown to outperform monolinguals on the AX-Continuous Performance Task (Morales et al., 2015), which pits proactive monitoring against reactive/inhibitory control (Braver et al., 2001; Braver, 2012). While some have hypothesized that language switching experience can enhance proactive control in bilinguals (Zhang et al., 2015), others have proposed that switching will have an impact on reactive control processes (Green & Abutalebi, 2013). We conducted an aggregate analysis of the AX-CPT in monolinguals, bilinguals, and L2 learners (N=818) to investigate how language experience may modulate cognitive control processes. While bilinguals overall exhibited slower RTs relative to L2 learners and monolinguals, only bilinguals immersed in the L2 performed as or more accurately across conditions. We discuss these results with respect to language immersion and dominance, as well as codeswitching experience.
PIRE fellow Manuel Pulido-Azpiroz presents at the 57th Meeting of the Psychonomics Society.
Studies exploring the role of the first language (L1) during the processing of second language (L2) collocations (e.g., set the table), show faster RTs in collocational over unrelated word pairs, and more efficient processing of L1-L2 congruent collocations (word-by-word equivalents) relative to incongruent ones (Wolter & Gyllstad 2011). No studies to date have investigated L1 interference in the processing of collocations. Using ERPs, our goal is to explore this question by capitalizing on increased L1 activation induced by the presence of a cognate noun in 50% of collocations. Cognates are predicted to enhance the effects of L1 on L2, resulting in increased facilitation via congruency, and in increased interference due to conflict in incongruent collocations. Faciliation/Interference should elicit modulated N400s. We recorded ERPs while participants performed a lexical decision task containing 656 Verb+Noun sequences (82 congruent, 82 incongruent collocations). Preliminary results (N=11) reveal modulated N400s based on congruency.
Kinsey Bice, a 2016 PIRE fellows, has recently presented a poster at the International Meeting of the Psychonomic Society.
Language contexts vary dramatically around the world, leading to diverse linguistic experiences for monolinguals, language learners, and proficient bilinguals. While only bilingual speakers experience the effects of switching between a second language (L2) and the native language (L1), all speakers, including monolinguals, encounter situations in which an L2 (known or unknown) is heard and semantically supported by context through gestures, the surrounding environment, or translation. The present study examined how comprehension in the L2 (in bilinguals) or the intention to comprehend the L2 (in monolinguals) affects the ability to subsequently speak in another language. Participants performed a picture-naming task while event-related potentials (ERPs) were recorded. They named pictures from the first block in one language, viewed pictures while listening to the names in another language during the second block, and returned to the first language for naming the pictures in the third block. Participants varied in language knowledge as well as context: Mandarin-English bilinguals in the USA, monolingual English participants in the USA, Cantonese- English simultaneous bilinguals in Hong Kong, and monolingual English participants in Hong Kong. Initial ERP analyses reveal a modulating effect of language experience, such that bilinguals benefit from repetitions of both heard and spoken items, monolinguals in the USA who listened to pictures named in an unknown language are not facilitated in subsequent naming, and monolinguals living in Hong Kong experience interference for naming items that were previously heard in Cantonese. These results suggest that language experience and language context change how comprehension affects production.
Grant Berry, a 2016 PIRE fellow, has recently presented at the 2016 LabPhon conference.
Individuals are attuned to process variation at all levels of linguistic production. Even at the level of phonetics, where phones vary due to coarticulation, physiology, and language experience (among other factors), speakers alter their production in response to the speech of their interlocutor (phonetic alignment; e.g., Pardo, 2006; Babel, 2012). Most research on phonetic alignment has investigated phonetic variation within a single phonological category, but individuals are also often exposed to systematic production variation across categories. Additionally, nearly all research on the topic has assessed phonetic alignment at only two time points (cf. Pardo, 2006; Delvaux and Soquet, 2007; Babel and Bulatov, 2011; Nielsen, 2011; Babel et al., 2013; Trofimovich and Kennedy, 2014; Hwang et al., 2015) rather than in real time, raising the question of whether alignment is an incremental process influenced by continued exposure or a rapid shift in production patterns due to discourse context.
The current study investigates the plasticity of phonological boundaries in discourse using a corpus of 34 Spanish-English bilinguals who converse with two Dutch confederates in English as a lingua franca across two speech styles (the Nijmegen Corpus of Spanish English; Kouwenhowevn et al., Forthcoming). When Spaniards converse with Dutch interlocutors in English as a lingua franca (ELF), they encounter two variables not present in their L2 speech: an English-like vowel contrast they have difficulty producing (/i/-/ɪ/; cf. Flege, 1991; Booij, 1995:1; Casillas, 2015) and a merger of two English vowels that habitually distinguish (/ɛ/-/æ/; cf. Booij, 1995:4-5; Archila-Suerte et al., 2012; Giacomino, 2012). We track the production of these contrasts during formal and informal speech using the Pillai score1 (e.g., Nycz and Hall-Lew, 2013; Hay et al., 2006; Babel et al., 2013) as an analog of category separation and mixed effects models of the corpus data (cf. Baayen et al., 2008; Baayen et al., 2015; Barr et al., 2013; Bates et al., 2015a; Bates et al. 2015b).
Results indicate that Spaniards aligned with Dutch confederates in their phonological category production, quickly merging their /ɛ/-/æ/ distinction and gradually separating their merged /i/-/ɪ/ category (see Figure 1), rather than adopting standard English production (a four-way contrast). We found greater merger in informal speech, but an interaction with time for the /i/-/ɪ/ contrast, which indicates that /i/ and /ɪ/ gradually separated in informal conversation. However, there was no effect of time for the /ɛ/-/æ/ contrast: Spaniards merged /ɛ/ and /æ/ more strongly in informal than in formal conversation, but the magnitude was stable throughout the conversation (see Figure 1). Finally, proficiency influenced alignment: the most proficient speakers separated /i/-/ɪ/ and merged /ɛ/-/æ/ more than less proficient speakers (see Figure 2). We situate these results alongside other research on phonetic alignment and speech production, stressing the importance of treating phonological categories as dynamic and interpreting phonetic alignment as a complex phenomenon that may be rapid or gradual, depending on the phonological categories under investigation.
Anne Beatty-Martínez, a 2015 PIRE fellow, has recently presented a poster at the 29th annual conference on human sentence processing.
Linguistic prediction allows individuals to link what comprehenders expect to hear with what speakers actually say. The Production-Distribution-Comprehension model (PDC; Gennari & MacDonald, 2009) states that listeners’ and readers’ sensitivity to distributional patterns in production can constrain the comprehension system. The PDC model inherently predicts variability among speakers who codeswitch and speakers who don’t codeswitch. We test this hypothesis at the neural level by examining the relationship between codeswitching production patterns and comprehension difficulty. Because exposure to codeswitching is predicted to impact comprehension, we recruited Spanish-English codeswitchers in the US and non-codeswitchers in Spain who were highly proficient in both languages. We focus on the production and comprehension of mixed noun phrases (mixed NPs; e.g., elM forkM). Evidence from naturalistic corpora suggests that bilinguals exhibit an overall preference for the masculine determiner (el), regardless of the noun’s gender in Spanish (Otheguy & Lapidus, 2003). In contrast, switches involving the feminine determiner (la) occur less frequently and are restricted to English nouns that are feminine in Spanish. To illustrate, el-codeswitches such as (elM forkM) and (elM spoonF) are extremely common. To a lesser extent, la-codeswitches involving feminine nouns (laF spoonF) have also been attested in bilingual speech, while those involving masculine nouns (*laF forkM) have not. We predicted that neurophysiological responses would reflect this asymmetry in codeswitchers but not in non-codeswitchers. Production was examined in a corpus of unscripted, taskoriented dialogues; comprehension was investigated in sentential contexts using event-related potentials. The same groups of bilinguals participated in the production and comprehension studies. As shown in Table 1, results for production show that proportions of noun phrase types (i.e., Spanish, English, or mixed NPs) differed across groups (X2 (2, N=6691) = 297.28, p<.000). Codeswitchers (N=14) produced more mixed NPs than non-codeswitchers (N=15) and these switches robustly reflected the aforementioned gender asymmetry. The ERP experiment showed that for bilinguals who codeswitch (N=15), switches consistent with attested distributional patterns were easier to process. This is reflected in the N400 response for less frequent compared to more frequent switches (see Figure 1). Crucially, no such effects were observed in non-codeswitchers (N=20). Our data suggest that individuals are highly sensitive to the constraints of their interactional context. In this way, converging evidence from codeswitching sheds light on how production choices speakers make can predict comprehension performance.
Alex McAllister, a 2015 PIRE student, presented data with his PIRE advisor, Matthew T. Carlson, at the 2016 LabPhon conference.
This study probed the relationship between automatic phonotactic repair and speech production, by asking whether the repair structure (a prothetic vowel) may be susceptible to reduction in speech. Spanish productively repairs word-initial /s/-consonant clusters (henceforth #sC) with a prothetic /e/ in both production and perception. We asked whether the initial vowel in Spanish #esC words like espalda ‘back’, which matches the default repair vowel, is more prone to reduction than other initial vowels, such as in aspirina ‘aspirin’. We explore this question in the speech production of 15 speakers of Andalusian Spanish who produced half #esC and half #asC words in isolation (578 tokens). Outright vowel deletion was uncommon, but was more likely with initial /e/ (5%) than initial /a/ (0.3%, one token). Moreover, when the /s/ was realized with greater duration (cf. the common tendency to lenite syllable-final /s/ in Andalusian), shortening of /e/, but not /a/, was observed. These findings provide evidence that reduction may be enabled when the reduced material can be perceptually repaired, leading to the occurrence of apparently illicit sequences in actual speech, e.g. espalda produced as [spalda]. The influence of articulatory, frequency, and other factors on reduction is also evaluated.
Grant Berry, a 2016 PIRE fellow, has recently presented a poster at the 2016 LabPhon conference.
We investigate the plasticity of phonological boundaries in discourse in a lingua franca by tracking the production of 34 Spaniards conversing with two Dutch confederates in English across two speech styles, and we focus on incremental changes in two key English vowel contrasts. The first, /i/-/ɪ/, is not made by Spanish speakers in aggregate but reliably produced by the Dutch confederates. The second, /ɛ/-/æ/, is produced by Spaniards, but not by Dutch confederates. We assessed degree of merger in each of ten normalized time bins with the Pillai score. Results indicate that Spaniards align with the Dutch confederates in their production of these contrasts, merging /ɛ/ and /æ/ and gradually separating /i/ and /ɪ/, rather than adopting standard English production. We found greater merger overall in informal speech, but an interaction with time for the /i/-/ɪ/ contrast indicates that /i/ and /ɪ/ gradually separate in informal speech. Conversely, there is no effect of time for the /ɛ/-/æ/ contrast: Spaniards merge /ɛ/ and /æ/ when conversing with the Dutch confederates significantly more in informal speech, but the magnitude is static throughout each interview. Finally, proficiency modulates alignment: the most proficient speakers separate /i/-/ɪ/ and merge /ɛ/-/æ/ more than less proficient speakers.
Jonathan Steuck, a 2015 PIRE fellow, has recently presented a poster at NWAV 45.
Prosody conveys not only linguistic meaning, but also extralinguistic information determined by the broader situational context (e.g. ‘speaker and addressee attributes’; Cole 2015; Henriksen 2013). By using intonation, speakers may encode sociolinguistic meaning via phonetic accommodation, “the process whereby speakers in an interaction modify their speech in response to their interlocutor” (MacLeod 2012:ii). While accommodation may be partly automatic and subconscious (Pickering & Garrod 2004; Trudgill 2008), sociophonetic research supports that speakers can exercise socially-motivated agency when accommodating (see e.g. Giles et al. 1991 for Communication Accommodation Theory; Babel 2010 for vowels; Babel & Bulatov 2011 for f0; Romera & Elordieta 2013 for group-level accommodation). Given these findings, to what extent do speakers of different dialects with contrasting intonation accommodate when they come into contact? To address this question, I quantify whether a speaker changes her fundamental frequency (f0) to converge with (accommodate to) or diverge from a Spanish speaker of another dialect when asking information- and confirmation-seeking yes-no questions—a crucial locus for analysis given the cross-dialectal variation in the nuclear accent and boundary tone configurations in Andalusian (rising: L*HH%; Henriksen & García-Amaya 2012), Porteño (falling: L+¡H*HL%; Gabriel et al., 2010, 2013) and Mexican (rising: L*LH%/H%; de-la-Mota et al. 2010) Spanish.
Data come from a two-session (S) experiment involving native speakers of Andalusian Spanish (N=3) or a foreign dialect (Porteño/Mexican; N=3) living in Granada, Spain (total speakers=6). In S1, speakers of the same dialect individually completed a baseline intonation survey and an Implicit Association Task (IAT; Greenwald et al. 1998, 2003), which gauges a speaker’s implicit socio-cultural bias for her interlocutor’s dialect (cf. Babel 2010). Speakers then completed a map task and 20 questions game to elicit yes-no questions and a paired sociolinguistic interview to capture sentiments regarding life in Granada. In S2, the same speakers completed another map task and 20 questions game in dialect-mismatched pairs. Information- and confirmation-seeking questions from both sessions (excluding interview) were hand-segmented by syllable and annotated in Praat (Boersma & Weenink 2015) using Sp_ToBI (Aguilar et al. 2009; Vilaplana & Prieto 2008). In the baseline and when speaking with someone of the same dialect, speakers usually produce the canonical contour (i.e. rising/falling) for their dialect. To capture more finely-grained phonetic effects, I compare nuclear accents (N=651) and boundary tones (N=651) by speaker using time-normalized f0 measurements and Smoothing Spline ANOVAs. Compared to baseline, all speakers of a foreign dialect phonetically accommodate f0 to an Andalusian while 2/3 of the Andalusian speakers phonetically diverge from a non-Andalusian in nuclear accents. At the phrase-final syllable, speakers do not categorically change their boundary tone (e.g. from rising to falling) when speaking with someone from another dialect. However, speakers display sensitivity to an interlocutor in terms of the percent of rising vs. falling contours used. Finally, results from the IAT suggest that bias for a particular Spanish dialect does not consistently predict degree of accommodation, a finding counter to Babel (2010). Nevertheless, this study provides further support for group-level accommodation (cf. Babel 2010; Romeira & Elordieta 2013).