The Production-Processing Paradox: Examining the Cognitive Effects of Code-Switching Via FEBLOC, a Novel Ecologically Valid Switching Corpus

Abstract

Many fluent bilinguals pass from one language to the another in a single conversation, seemingly without effort. However, in-lab experiments show that reading or hearing code-switching elicits cognitive costs (e.g., longer reaction times, distinct brain responses, etc.). The present dissertation teases apart this production-processing paradox and examines the cognitive consequences of code-switching according to modality. First, nineteen in-group pairs of French-English bilinguals (i.e., siblings, couples, parent-child, etc.) completed a series of speech elicitation activities in a private room. Their conversations were recorded, transcribed, and coded for language membership (English, French, code-switched). This speech data—coined the French-English Bilingual Loved Ones Corpus (FEBLOC)—resulted in a dataset of 36,000 utterances (18% code-switches). Exploratory analyses revealed that i) the amount of within-speaker code-switching did not impact performance on the speech elicitation activities; ii) between-speaker switching was related to decreased proficiency and performance; iii) members from the same testing pair tended to match each other's language usage, with no effect of relationship type (e.g., couples vs. relatives); iv) women tended to code-switch more often than men; and v) participants were inaccurate in self-evaluating their code-switching habits. Second, since FEBLOC contains multiparty speech (pairs of bilinguals), it was possible to compute proportions of produced switches (i.e., what partner-A said themselves) and heard switches (i.e., what partner-A heard from partner-B) for each participant. These indices were then examined in relation to the bilinguals’ performance on a subsequent behavioural Flanker task. The findings revealed a dissociation in the cognitive consequences of code-switching according to modality. Producing and hearing code-switches were both linked to ameliorated interference suppression (quicker responses to conflict trials), but interference disengagement was uniquely negatively associated to producing switches (slower responses to non-conflict trials). I speculated that producing dense code-switches engages proactive control (which impedes task disengagement), but that hearing inter utterance code-switches engages reactive control (which does not hinder task disengagement). This difference reflects the inherent nature of producing vs. processing: speech output can be biased according to a speaker’s own internal state (e.g., the accessibility of a given lexical item). By contrast, the listener is ‘at the mercy’ of the speaker’s internal state and must process input as it arrives. Finally, FEBLOC was utilized in a pilot EEG experiment. Ten new French-English participants listened to forty excerpts of genuine conversations from the corpus (half bilingual, half unilingual); they also completed a Flanker task, and five minutes of resting-state EEGs. The passive listening EEGs for each excerpt were split into equal-sized segments. Spectral components were then extracted and averaged across conditions, giving a measure of ‘global’ processing in the frequency domain. During both resting state and passive listening, language background characteristics (e.g., entropy) modulated gamma oscillations. Equivalent global neural responses were observed for densely code- switched excerpts and unilingual excerpts during the passive listening task. However, participants with larger N2 Flanker effects exhibited decreased code-switching listening costs, while participants with larger LPC Flanker effects exhibited increased code-switching listening costs. Under the assumption that early ERP components reflect bottom-up mechanisms and later components reflect top-down mechanisms, I speculated that processing code-switches is positively associated to (top-down) reactive control abilities. Nevertheless, the absence of global costs corresponds well with bilinguals’ feelings about code-switching: though bilingualism doubtlessly shapes cognition (even at rest), code-switching costs are probably functionally inconsequential outside of the lab. Overall, the present dissertation demonstrates that bilingual production and bilingual processing involve distinct mechanisms: producing (dense) code-switches engages (and therefore potentially trains) bottom-up proactive control, while processing switching is modulated by top-down reactive control. The original data included in the present dissertation circumvents many of the pre-existing issues in the code-switching literature (e.g., stigmatization, ecological validity, subjectivity), since (naive) in-group bilinguals were tested in a private room, and switching habits were operationalized objectively. With any luck, the corpus can continue to instruct future bilingualism research.