In the Words of Others: ERP Evidence of Speaker-Specific Phonological Prediction

Prediction models usually assume that highly constraining contexts allow the pre-activation of phonological information. However, the evidence for phonological prediction is mixed and controversial. In this study, we implement a paradigm that capitalizes on the phonological errors produced by L2 speakers to investigate whether specific phonological predictions are made based on speaker identity. L1 Italian speakers were asked to read sentence fragments, after which a final word was spoken by either an L1- or an L2-accented speaker. The spoken final word could be predictable or not, depending on the sentence context. The identity of the speaker (L1- vs. L2-accented) may or may not be cued by an image of the face of the speaker. Our main analysis indicated that cueing the speaker identity was associated with a larger N400 predictability effect, possibly reflecting an easier processing of predictable words due to phonological pre-activation. As visual inspection of the waveforms revealed a more complex pattern than initially anticipated, we used Temporal EFA (Exploratory Factor Analysis) to identify and disentangle the ERP components underlying the effect observed. In the L1-accent condition, predictable words elicited a posterior positivity relative to unpredictable words, possibly reflecting a P3b response, which was more pronounced when the speaker identity was cued. In the L2-accent condition, cueing the speaker identity was associated with a smaller N1 and a larger P3a response. These results suggest that phonological prediction for L1- and L2-accented speakers likely involves different cognitive processes.