[PS-2.19] Statistical learning in children's speech: are syllable-detection latencies a better measure?

Briscoe, J. , Prosser, R. . , Von Grebmer Zu Wolfsthurn, S. . & Kazanina, N. .

University of Bristol

Statistical learning (SL) captures regularities from the environment across multiple domains, including speech (Conway & Christiansen, 2005). Measuring SL from speech offers a useful insight to language acquisition, especially for atypical language development. Traditional methods rely on explicit recognition of tri-syllabic items at test, as evidence of SL, in adults and children. Learning is indexed through generalization from concatenated nonsense syllable streams that vary in probability during exposure, to word-like units at test. For children, recognition performance is typically above chance (e.g. Saffran et al., 1997), but not reliably so, suggesting that recognition performance varies with developmental factors (e.g. phonological memory). Our study combined implicit and explicit measures of SL to assess school-age children. Using a target-detection paradigm for syllables (following Franco et al., 2015) we sought to establish children?s sensitivity to SL on an implicit measure. We predicted that children, like adults, would identify targets from a presented stream at test, and that latencies would vary with probabilistic structure of exposure stream. With n=36 children aged 6-9 years tested, results are currently under analysis. Preliminary findings indicate variation in children?s ability to detect the target syllables, subsequent validation of latency data has implications for identifying children at language risk.