Electrophysiological signatures of statistical segmentation: A systematic review and meta-analysis

Le résumé fourni par la source

Statistical learning (SL), the ability to extract regularities from continuous sensory input, is widely studied using segmentation in embedded pattern learning (EPL) paradigms. Different electrophysiological measures, including event-related potentials (ERPs), steady-state responses (SSRs), and event-related spectral perturbations (ERSPs), have been proposed as neural signatures of statistical segmentation. However, the extent to which these measures provide consistent evidence remains unclear. We conducted a systematic review and meta-analysis of EEG and MEG studies to assess the evidence for these signatures in the current literature. After screening 635 records, 39 studies (46 experiments) contributed to the quantitative synthesis. We conducted separate meta-analyses of four signatures: N1 and N400 pattern-onset effects, emergence of pattern-rate SSRs, and suppression of item-rate SSRs. The results provided some evidence for an N400 effect and strong evidence for pattern-rate SSR enhancement, whereas neither N1 nor item-rate SSR effects reached statistical significance. No moderation effects were observed across the tested characteristics, with the exception of contrast type, which significantly moderated the N400 effect. A comparative model further indicated that pattern-rate SSRs provided the most robust evidence of statistical segmentation. We discuss how our results inform the interpretation of these signatures as neural indices of statistical segmentation and outline directions for future research.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.