Prediction in spoken language comprehension across linguistic levels: evidence from L2 speakers and adults with dyslexia.

Language comprehension is not only incremental but predictive: over the past two decades, extensive evidence has shown that comprehenders do not passively wait for linguistic input but anticipate words and structures before they are encountered, thereby facilitating efficient comprehension. While it is no longer surprising that prediction exists, important questions remain about how it operates across linguistic levels and population groups. Research on second-language (L2) speakers has produced conflicting findings, with some studies demonstrating predictive use of semantic, morphosyntactic, or phonological cues, while others report reduced or absent anticipatory processing. Even less is known about prediction in developmental dyslexia – very few studies have examined whether individuals with dyslexia can anticipate upcoming linguistic input, leaving this population under-represented in the literature. Yet understanding prediction in dyslexia is crucial, as the condition is characterised by difficulties in phonological and morphological processing, slower lexical retrieval, and differences in verbal working memory, which may affect anticipatory processing in both reading and spoken comprehension. This dissertation addresses these gaps by investigating predictive mechanisms at the semantic, morphosyntactic, and phonological levels, comparing neurotypical L1 adults with L2 speakers and adults with dyslexia. Two visual-world eye-tracking studies were conducted. Experiment 1 (Italian) tested anticipation of a highly expectable target and its phonological form, and the predictive use of grammatical gender cues on the determiner, in L1 Italian speakers, L2 speakers (Tyrolean L1), and adults with dyslexia. Experiment 2 (English) replicated Ito and Husband (2017), examining contextual, semantic, and phonological prediction, and included a semantic competitor condition to test whether related alternatives receive anticipatory activation. Results revealed a consistent ability across all groups to anticipate a highly predictable target word. Beyond this, the strength and timing of prediction varied. In Experiment 1, L2 participants showed somewhat shallower prediction with similar onset timing to L1 controls, while individuals with dyslexia displayed broadly preserved prediction. No group showed anticipatory looks driven by grammatical gender cues, although L1 and L2 participants showed a facilitation effect after word onset, which was absent in the dyslexia group. There was no evidence for phonological pre-activation. In Experiment 2, all groups anticipated the target, but semantic competitor effects were selective: only contextually supported competitors attracted anticipatory looks. Adults with dyslexia showed delayed competitor activation, while L2 speakers showed a trend towards slower semantic pre-activation. Again, no phonological competitor effects emerged. These findings indicate that spreading activation to semantically related but contextually implausible words is limited and constrained by contextual relevance. Strong contextual support appears to inhibit irrelevant alternatives, suggesting an efficient system that balances the benefits of anticipation with its cognitive costs.