Language comprehension is largely supported by predictive mechanisms that account for the ease and speed with which communication unfolds. Both native and proficient non-native speakers can efficiently handle contextual cues to generate reliable linguistic expectations. However, the link between the variability of the linguistic background of the speaker and the hierarchical format of the representations predicted is still not clear. We here investigate whether native language exposure to typologically highly diverse languages (Spanish and Basque) affects the way early balanced bilingual speakers carry out language predictions. During Spanish sentence comprehension, participants developed predictions of words the form of which (noun ending) could be either diagnostic of grammatical gender values (transparent) or totally ambiguous (opaque). We measured electrophysiological prediction effects time-locked both to the target word and to its determiner, with the former being expected or unexpected. Event-related (N200-N400) and oscillatory activity in the low beta-band (15-17Hz) frequency channel showed that both Spanish and Basque natives optimally carry out lexical predictions independently of word transparency. Crucially, in contrast to Spanish natives, Basque natives displayed visual word form predictions for transparent words, in consistency with the relevance that noun endings (post-nominal suffixes) play in their native language. We conclude that early language exposure largely shapes prediction mechanisms, so that bilinguals reading in their second language rely on the distributional regularities that are highly relevant in their first language. More importantly, we show that individual linguistic experience hierarchically modulates the format of the predicted representation.

We investigated how native language experience shapes anticipatory language processing. Two groups of bilinguals (either Spanish or Basque natives) performed a word matching task (WordMT) and a picture matching task (PictureMT). They indicated whether the stimuli they visually perceived matched with the noun they heard. Spanish noun endings were either diagnostic of the gender (transparent) or ambiguous (opaque). ERPs were time-locked to an intervening gender-marked determiner preceding the predicted noun. The determiner always gender agreed with the following noun but could also introduce a mismatching noun, so that it was not fully task diagnostic. Evoked brain activity time-locked to the determiner was considered as reflecting updating/reanalysis of the task-relevant preactivated representation. We focused on the timing of this effect by estimating the comparison between a gender-congruent and a gender-incongruent determiner. In the WordMT, both groups showed a late N400 effect. Crucially, only Basque natives displayed an earlier P200 effect for determiners preceding transparent nouns. In the PictureMT, both groups showed an early P200 effect for determiners preceding opaque nouns. The determiners of transparent nouns triggered a negative effect at ∼430 ms in Spanish natives, but at ∼550 ms in Basque natives. This pattern of results supports a "retracing hypothesis" according to which the neurocognitive system navigates through the intermediate (sublexical and lexical) linguistic representations available from previous processing to evaluate the need of an update in the linguistic expectation concerning a target lexical item.

We investigated how native language experience shapes prediction mechanisms. Two groups of bilinguals (either Spanish or Basque natives) performed a word matching task (WMT) and a picture matching task (PMT). They indicated whether the stimuli they perceived matched with the noun they heard. Spanish noun endings were either diagnostic of the gender (transparent), or ambiguous (opaque). ERPs were time-locked to the gender-marked determiner preceding the predicted noun. In the WMT both groups showed a negative (~340 ms) effect. Basque natives displayed an earlier effect (~150 ms) for determiners preceding transparent nouns. In the PMT both groups showed an early effect (~160 ms) for determiners preceding opaque nouns.Transparent nouns' determiners elicited prediction at ~330 ms in Spanish natives, but at ~460 ms in Basque natives. It is concluded that bilinguals rely on the features of the L1 for predicting in the L2: native experience moulds prediction. Linguistic prediction is hierarchical, with different operation levels communicating at an interface stage.

Hearing spoken words can enhance visual object recognition, detection and discrimination. Yet, the mechanism underlying this facilitation is incompletely understood. On one account, words would not bias visual processes at early levels, but rather interact at later decision-making stages. More recent proposals posit that words can alter visual processes at early stages by activating category-specific priors in sensory regions. A prediction of this account is that top-down priors evoke changes in occipital areas before the presentation of visual stimuli. Here, we tested the hypothesis that neural oscillations can serve as a mechanism to activate language-mediated visual priors. Participants performed a cue-picture matching task where cues were either spoken words, in their native or second language, or natural sounds, while EEG and reaction times were recorded. Behaviorally, we replicated the previously reported label-advantage effect, with images cued by words being recognized faster than those cued by natural sounds. A time-frequency analysis of cue-target intervals revealed that this label-advantage was associated with enhanced power in posterior alpha (9-11 Hz) and beta oscillations (17-19 Hz), both of which were larger when the image was preceded by a word compared to a natural sound. Prestimulus alpha and beta rhythms were correlated with reaction time performance, yet they appeared to operate in different ways. Reaction times were faster when alpha power increased, but slowed down with enhancement of beta oscillations. These results suggest that alpha and beta rhythms work in tandem to support language-mediated visual object recognition, while showing an inverse relationship to behavioral performance.