My research focuses on incorporating sociolinguistic and corpus-based work into psycholinguistic and lab-based experiments. Recent evidence suggests a tight link between production and processing, and as such I wish to explore this link through bilingualism and codemixing (the fluid integration of two languages during production). My dissertation seeks to explore how community norms of codemixing, and its pattern of usage in everyday life, interact with the cognitive limitations of language processing.
Another interest of mine comes in the neurocognitive application of theoretical models of language, including looking at aspects of Emergentism and Usage-Based grammar, and integrating these theories into psycholinguistic models of sentence processing, both in perception and production. More information on my current and past research projects are given below!
Current Research Projects
The Psycholinguistic Status of Lone Other-Language Items: Nonce or not?
The ability for bilinguals to integrate two languages within a single utterance has yet to receive an agreed-upon definition within linguistics. Broadly construed as language mixing, research on this topic has been guided by two fields: sociolinguistics, which seeks to delimit the social and structural conditioning of language mixing (when, where, and why it occurs); and psycholinguistics, focusing instead on the cognitive mechanisms required to control two languages. While much has been learned from each field independently, the disconnect between the two is no more apparent than at the single-word level. Sociolinguistic work on lone other-language items (LOLIs)—single words of one language inserted into another—has shown that their behavior differs systematically from multi-word instances of language mixing. In particular, LOLIs tend to surface morphosyntactically integrated into the surrounding language: they carry the morphological markings and are inserted into the syntactic positions of words congruent with the surrounding language and incongruent with their language of origin. The psycholinguistic study of language mixing has eschewed such findings and instead focused on the cognitive costs associated with switching languages in laboratory settings . Recent psycholinguistic work incorporating insights from sociolinguistic studies on naturalistic language mixing has demonstrated that such paradigms change the ways in which bilinguals process language-mixed input: when experimental design deviates from naturalistic patterns, the generalizability of results is affected.
Sixty highly proficient early Spanish-English bilinguals were recruited from the University of Puerto Rico in Río Piedras, all self-identifying as avid codeswitchers. Pupillometry was selected as the method of data collection because changes in pupil size have been associated with real-time changes in cognitive load. Participants listened to sentences containing noun-adjective phrases in four conditions: (1) unilingual Spanish, 2) Spanish with an English LOLI as the target noun, 3) Spanish with an established English borrowing as the target noun, and 4) Spanish with an English MWCS consisting of both the noun and adjective. Singe larger pupil sizes are indicative of greater cognitive load, a larger pupillary response was predicted for MWCS compared to established borrowings. If LOLIs are treated as nonce borrowings, the pupillary response was predicted to pattern like established borrowings and unlike MWCS.
Unilingual targets elicited the smallest pupillary response, with LOLIs, MWCSs, and established loanwords all showing significantly larger responses. Importantly, LOLIs did not pattern like established loanwords, which showed the largest pupillary response, but instead were more like (though still different from) MWCSs. Established loanwords seem to stand out the most, perhaps due to their relative salience when presented alongside unilingual ‘standard’ Spanish. These results suggest that, with respect to the comprehender, both LOLIs and established loanwords differentiate themselves from unilingual speech and each other, implying that morphosyntactic integration alone is not the defining characteristic of nonce borrowings. Rather, a gradient emerges across the four conditions, perhaps driven by aspects of salience and surprisal. In summary, the present study argues against a qualitative distinction between LOLIs, MWCS, and established loanwords.
When the going gets tough, the tough get codeswitching: Increased cognitive demand facilitates the processing of codeswitched sentences
One assumption in the literature on bilingual codeswitching is that it is cognitively effortful: switching languages requires the deactivation of the irrelevant language, and activation of and orientation towards the relevant language. Recent research, including some of my own, indicates that codeswitching need not be cognitively demand and, in fact, may serve as a way to alleviate processing demands. In this study, highly proficient Spanish-English bilingual codeswitchers read unilingual and codeswitched sentences in two blocks: in one block, each sentence was followed by a comprehension question with no additional task; in the second block, participants completed a math verification task between each sentence and comprehension question, resulting in increased cognitive demands during processing. While reading times on codeswitched sentences were greater than unilingual sentences in the no-delay condition, reading of codeswitched sentences was faster than unilingual sentences when bilinguals had to also complete the math verification task. The results of this study are in line with recent research on conflict adaptation that show that engaging in an already difficult task (in this case, the math verification task) facilitates another difficult task (comprehending codeswitched sentences), suggesting that codeswitching need not always be costly to the comprehender.
The Easy-First Bias in Bilingual Speech: Evidence from Language Mixing
The Easy-First bias was proposed by MacDonald (2013) as a general processing constraint underlying speech production: easier elements in production (be them more accessible due to priming, more frequent, shorter in length, etc.) will tend to occur earlier in the utterance compared to harder (less accessible, less frequent, longer) elements, allowing more time for the processing system to plan these harder elements. In order to test this, I am currently in the process of conducting a corpus analysis on the New Mexico Spanish-English Bilingual corpus (NMSEB; Torres Cacoullos and Travis, 2018) under the supervision of Dr. Rena Torres Cacoullos and Dr. Giuli Dussias. In this project, I am coding for the location of single-word insertions as well as the start of mutli-word stretches of language mixed speech at the level of the intonation unit and the prosodic sentence. Preliminary results suggest that, while language mixing tends to occur throughout the prosodic sentence, language mixing tends to be delayed in the shorter, more prosodically coherent, intonation units. In the case of single-word insertions, they occur overwhelmingly as the final word of the intonation unit, suggesting a unique prosodic/discourse function.
To further test this bias, this corpus study will be followed-up with a series of psycholinguistic experiments testing whether language mixing is more "difficult" when it occurs earlier in the sentence compared to when it occurs later, at both the single- and multi-word levels. We predict that if distributional patterns in the input (i.e., the corpus) suggest the presence of an Easy First bias then language mixers will also demonstrate this bias in their processing, given the preponderance of "late" language mixes. This would suggest that language mixing is not costly, per se, but rather that the bilingual production/processing system has exploited general cognitive mechanisms to ease the integration of language mixed speech.
Revisiting Asymmetric Switch Costs: Refined Evidence from Pupillometry
In the study of cued language-switching (e.g., when bilinguals are told which language to produce, normally naming images or numbers), the literature has often, though not always, reported asymmetric switch costs. Paradoxically, the cost of switching is most notable in the dominant language, while limited or even reversed in the non-dominant language. The classic story, dating back to Meuter and Allport's seminal 1999 study, was that producing the non-dominant language requires strong inhibition of the dominant language; thus, when switching back into the dominant language, that inhibition must be released, a process which is difficult and thus leads to switch costs. Specifically, these asymmetric switch costs are often attested in reaction times (the time it takes to correctly name the picture, for example), but further studies using neuropsychological methods have cast doubt on this assumption.
In the present study, early Spanish-English bilinguals named images in both Spanish and English, cued by the color of the border surrounding the image. Reaction times were recorded, as was the size of the pupil time-locked to the onset of the picture. The pupillary response has been reported to index such processes as focused attention, arousal, and processing load. In the reaction times, a classic asymmetric switch cost emerges: participants are longer to name in English (their dominant language) after having previous named in Spanish (their non-dominant language) than in English. In Spanish, however, there are no differences in naming times. In the pupillary response, however, we see robust switch costs in Spanish but a reversal of the switch effect completely in English. In other words, the pupillary response is larger when the participant names in English after naming in English than when the participant names in English after naming in Spanish. These findings, we propose, help us to refine the prior assumption that the release of inhibition of the dominant language is effortful: if the pupillary response is indeed indexing cognitive load, then it suggests that switching back into the dominant language does not require the effortful release of inhibition, but rather that release of inhibition just takes time. Likewise, switching into the non-dominant language requires inhibition of the dominant language to be "ramped up", resulting in a robust switch effect in the pupillary response.
The Effects of Mixing and Blocking on the Comprehension of Code-Switched Sentences
Given the link between language usage and language structure, one goal of lab-based approaches should be to draw on corpus-based studies to answer questions concerning the production and processing of language. In the psycholinguistic study of codemixing, this is particularly lacking: many studies do not take into account the social contexts in which codemixing occurs. Therefore, the presentation of stimuli in lab-based studies often does not align with how code-switches are encountered in naturalistic bilingual discourse. For example, bilingual corpora suggest that intra-sentential codemixing is not particularly dense, with codemixes localized and surrounded by stretches of unilingual discourse. This is not represented by a blocked design, commonly used in lab-based studies of codemixing, where codemixes occur for long stretches of time within a single block. A mixed mode of presentation more adequately approximates what we find in bilingual corpora. Here we examine how the method of presentation—mixed vs. blocked—affects the comprehension of code-switches, with the goal of determining which method more adequately reflects bilinguals’ experiences.
In a within-subject design, we recorded eye-movements while habitual Spanish-English code-switchers read sentences containing determiner-noun switches (La mujer planchó el suit para su hijo) and unilingual Spanish sentence (El jóven exploró la casa con sus amigos) in two sessions: in the blocked session, participants read a block of unilingual Spanish sentences followed by a block of code-switched sentences in a counterbalanced fashion; in the mixed session, unilingual Spanish sentences were interleaved with code-switched sentences. Linear mixed-effects models revealed that codemixing in the mixed session yielded faster reading times than in the blocked session, suggesting that a long string of codemixing is more cognitively costly than interspersing codemixing with unilingual sentences. This is in line with corpus-based evidence which shows that codemixing is not particularly dense or frequent, suggesting that a mixed mode of presentation may be more suitable when studying Spanish-English bilingual codemixing. However, community norms must still be kept in mind, as not all codemixing communities behave equally. As such, it is imperative that psycholinguists understand the populations that they are working with--not only from a cognitive, but also a social perspective.