Calendar of events

Title: Sequence length and time-course reveal limits of statistical learning theories

Abstract: Statistical learning – identifying coherent groups/sequences of information based on their co-occurrence in information streams – is a fundamental learning mechanism. However, most paradigms test performance at the end-point of learning, lacking sensitivity to long sequences and ignoring time-course behaviour that differentiates competing theoretical accounts. We remedy these problems using a new mouse-tracking visual statistical learning paradigm that records time-course behaviour and trains participants on a broader range of sequence lengths. Across six pre-registered online studies (N = 216) with uniform (2-, 3-, or 4-item) and mixed length sequences (2+4, 3+4, 2+3), we found: shorter sequences were learned more easily than longer ones (length effect); sequences were learned better in shorter contexts (e.g., 3-item sequences in mixed vs uniform designs) (context effect); and a trend for later items in a sequence to be learned more easily than earlier ones (order effect). We further tested these data against three prominent computational accounts (SRN, PARSER, TRACX). None fully reproduced relative differences across experiments, and all showed either absent or inconsistent order effects. We discuss how models might be refined to capture the dynamics of human statistical learning.

 Bio: Francesco Cabiddu is a postdoctoral researcher at Nottingham Trent University, working on a Leverhulme-funded project led by Professor Gary Jones that examines the mechanisms underlying the time course of human statistical learning. He previously worked at University College London on a project led by Professor Gabriella Vigliocco, investigating the role of multimodal teacher input in word and concept learning in both adults and children. He completed his PhD at Cardiff University, where he studied usage-based learning mechanisms in language development and processing. Francesco’s research interests centre on statistical learning and, more broadly, on usage-based approaches to language learning, which he has explored in both children and adults using experimental and computational methods.