The Role of Top-down Attention in Statistical Learning of Speech

Statistical learning (SL) refers to the ability to extract regularities in the environment and has been well-documented to play a key role in speech segmentation and language acquisition. Whether SL is automatic or requires top-down attention is an unresolved question, with conflicting results in the literature. The current proposal tests whether SL can occur outside the focus of attention. Participants either focused towards, or diverted their attention away from an auditory speech stream made of repeating nonsense trisyllabic words. Divided-attention participants either performed a concurrent visual task or a language-related task during exposure to the nonsense speech stream, while control participants focused their attention to the speech stream. Visual attention was taxed through the classic Multiple Object Tracking paradigm, requiring tracking of multiple randomly moving dots. Linguistic attention was taxed through a self-paced reading task. Following speech exposure, SL was assessed with offline tests, including a post-exposure explicit familiarity rating task, and an implicit reaction-time (RT) based syllable detection task. On the explicit familiarity rating measure, participants showed a reduction in learning when language-specific processing was taxed as compared to when visual resources were taxed. On the more implicit reaction time-based measure of SL, both divided-attention and full-attention controls performed comparably, all showing evidence of SL. These results suggest SL can proceed even when domain-specific (visual) resources are limited, but is compromised when more specific, language-related resources are taxed. These results offer insight into the neural cognitive underpinnings of SL and have exciting practical applications for improving adult second language acquisition