Kerz, E. 1 , Wiechmann, D. 2 & Christiansen, M. 3, 4
1 RWTH Aachen University
2 University of Amsterdam
3 Cornell University
4 Aarhus University
A substantial body of research has highlighted the central role of statistical learning (SL) in the acquisition of language in children and adults. This research has predominantly been conducted using artificial language learning (ALL) paradigms. However, it remains unclear to what extent such SL generalizes to a natural learning context. In contrast to the miniature language used in ALL tasks, natural language input is far noisier and rarely ever exhibits distributional patterns and statistical regularities with the same clear consistency. Language users are thus faced with the challenge of keeping track of the ever-changing statistics inherent in natural language input. This challenge is exacerbated by the fact that in natural language exposure many different types of linguistic patterns reflecting multiple distributional statistics are encountered across different communicative contexts. This study reports on a series of experiments examining the link between the distributional properties of natural language input and adult L2 processing across four communicative contexts - to assess L2 sensitivity to the variability in the statistics of multiword sequences. Distributional statistics (frequency estimates and entropy) were found to elegantly align with the behavioral data. Our study thus makes an important contribution to understanding statistical learning in the wild.