Grape Expectations: A collection of EEG stories on form-based prediction in natural language contexts

Abstract

It is no longer controversial to say that language comprehension involves prediction. Decades of psycholinguistic research have demonstrated that comprehenders reliably anticipate both the meaning and the form of upcoming words (DeLong et al., 2005; Federmeier & Kutas, 1999; see Kuperberg & Jaeger, 2016; Kutas & Federmeier, 2011). The next frontier in predictive processing research is understanding the mechanisms by which prediction arises and how those mechanisms develop in the world’s many languages. To do this, the field must begin to characterize how prediction works in more naturalistic contexts and across a wider range of modalities and populations. In this dissertation, I present three EEG experiments that aim to understand the nature of predictive processing in ordinary storytelling contexts. In each experiment, we used a novel naturalistic story paradigm in which participants simply comprehend rich, naturally produced narratives that have experimental manipulations injected into them. As participants comprehended these narratives, we recorded their neural responses to a set of manipulated target words, allowing us to assess the nature of their linguistic predictions. In Paper 1, we investigated prediction in Spanish-English bilinguals while they listened to short stories that were presented in English with occasional Spanish words. We found that bilinguals’ predictions were lexically specific, meaning that they had generated expectations for a particular word in a particular language. This work provided initial evidence that predictions in naturalistic settings are form-based and move beyond just anticipating the gist of upcoming linguistic material (see Yacovone, Moya, & Snedeker, 2021). In Paper 2, we further assessed the nature of form-based predictions by asking whether English-speaking adults predict the sounds of upcoming words during comprehension. In this experiment, participants watched a cartoon narration of a children’s book, which had a set of manipulations spliced into it. Specifically, we identified highly predictable and unpredictable words, and then replaced them with nonwords that sounded similar or dissimilar to the original word (e.g. cake, ceke, vake). Results indicated smaller neural responses to baseline words (cake) and similar nonwords (ceke) relative to dissimilar nonwords (vake). This reduction in neural responses, however, was only observed for predictable words, replicating the prior findings from similar studies with more tightly controlled experimental designs. Thus, Paper 2 demonstrated that listeners are able to predict the sounds of upcoming words in natural language contexts. In Paper 3, we explored the nature of form-based prediction in a different modality—namely, sign language. To do this, we had deaf signers of American Sign Language (ASL) watch a narrative in which we had manipulated a set of target signs. Following the logic of Paper 2, we identified both predictable and unpredictable signs, and then replaced them with similar and dissimilar non-signs. We found tentative evidence that signers anticipate the handshape of upcoming signs during naturalistic comprehension, mirroring the findings from Paper 2. Taken together, these studies present a body of work that highlights the importance of understanding prediction in natural language contexts and across different modalities and populations.