Mechanisms of language processing in monolinguals and bilinguals: prediction and production

Abstract

For both monolinguals and bilinguals, language comprehension is not merely a bottom-up passive process of waiting for words, but also involves actively generating top-down predictions about what is likely to occur. For example, if a speaker says It is raining outside, so you should take an …, the listener may predict that the next word is likely umbrella. For bilinguals, this prediction process is made more complicated because they can conduct language processing in two languages. That is, they can activate both their first language (L1) and second language (L2) in parallel during comprehension and production. However, it remains unclear whether this cross-language activation, especially during bilingual production, occurs in one-language contexts (where only the target language is used). In addition, while substantial evidence shows that L1 speakers can predict many aspects of upcoming words and that L2 speakers do so less effectively, the cognitive mechanisms underlying prediction—such as whether it requires cognitive resources or whether bilinguals co-activate L1 during L2 prediction—remain poorly understood. Therefore, this thesis investigated the mechanisms of language processing in monolinguals and bilinguals, focusing on prediction and production. It first examined whether prediction is constrained by cognitive resources by having L1 Mandarin Chinese speakers listen to Mandarin Chinese sentences with varying speech rates and levels of contextual predictability (highly predictive, moderately predictive, or unpredictive of a final word). We measured how quickly participants judged whether a given letter was contained in that final word (Experiment 1), named a picture corresponding to that final word (Experiment 2), and named that final word (Experiment 3). We found that participants responded more slowly at a faster speech rate, with greater effects in more than less predictable sentence contexts. These results suggested that a faster speech rate slows down prediction speed due to increased cognitive load, and that the detrimental effect of cognitive load is greater in more predictable contexts, where the prediction mechanism can be engaged to a larger extent. Overall, these results support the idea that prediction is cognitively demanding. Experiments 4 and 5 used a picture-word interference paradigm to investigate whether bilinguals activate their L1 during L2 production within a pure L2 context. We asked highly proficient Mandarin Chinese-English bilinguals (Experiment 4) and English native speakers who spoke no Mandarin (Experiment 5) to name pictures (e.g., hat, “mao zi” in Mandarin) in English while ignoring English auditory distractors that were phonologically related to the Mandarin translations of the picture names (mouth, phono-translation distractors). We found that both groups showed inhibition for semantically related distractors and facilitation for phonologically related distractors. Most importantly, Mandarin-English bilinguals were quicker to name pictures when just preceded by phono-translation distractors compared to unrelated distractors, but English native speakers were not. These findings suggest that information associated with words from the non-target language is activated even when that language is irrelevant to the task, but do not compete for selection. Experiments 6 and 7 directly tested whether bilinguals pre-activate L1 translations of predictable words during L2 comprehension and whether this depends on language contexts, using the visual-world paradigm. Specifically, Mandarin-English bilinguals listened to L2 English sentences containing a highly predictable word (e.g., You should take an umbrella with you, because there will be heavy rain at three o'clock this afternoon) while viewing a display containing a critical object and three distractors. We found that participants predictively fixated more on a competitor object whose Mandarin Chinese name was a homophone (e.g., feather [Mandarin: yu3]) of the Mandarin Chinese translation of the predictable word (e.g., rain [Mandarin: yu3]) than an unrelated object (e.g., comb [Mandarin: shu1]) when both languages were used (Experiment 7) but not when just English was used (Experiment 6). Our findings suggest that bilinguals predict across languages when both languages are contextually relevant but not otherwise.