Onnis, L. & Fong, S.
Nanyang Technological University Singapore
A crucial question about human statistical learning (SL) is whether it is stable across the lifespan, or modified by maturational and/or experiential factors. Here we hypothesized a link between SL and language experience. In particular, left-branching (LB) and right-branching (RB) languages are differentially predictive of linear and constituent word order, placing a parsing emphasis on forward (LB) or backward (RB) transition probabilities respectively (Onnis & Thiessen, 2013).
Bilingual speakers of LB and RB languages (Tamil and English) listened to an artificial grammar characterised by conflicting forward and backward probabilities. The crucial manipulation involved exposing participants to the same grammar instantiated with Tamil-sounding and English-sounding pseudo-words in separate sessions (within-subjects).
We expected participants to switch implicit preferences for forward probabilities in the Tamil session, and for backward probabilities in the English session. The study has been pre-registered and data will be analysed before the conference. If obtained, the results would show that bilinguals flexibly adapt statistical expectations on the fly based on the input they expect to process. Furthermore, the pattern of predicted results are interpretable in light of the 'paradox of success' hypothesis, whereby new learning is constrained by the mind being tuned and optimised for previously learned structures.