The specificity of sequential statistical learning: Statistical learning accumulates predictive information from...

Learning statistical regularities from the environment is ubiquitous across domains and species. It might support the earliest stages of language acquisition, especially identifying and learning words from fluent speech (i.e., word-segmentation). But how do the statistical learning mechanisms involved in word-segmentation interact with the memory mechanisms needed to remember words - and with the learning situations where words need to be learned? Through computational modeling, we first show that earlier results purportedly supporting memory-based theories of statistical learning can be reproduced by memory-less Hebbian learning mechanisms. We then show that, in a memory recall task after exposure to continuous, statistically structured speech sequences, participants track the statistical structure of the speech sequences and are thus sensitive to probable syllable transitions. However, they hardly remember any items at all, with 82% producing no high-probability items. Among the 30% of participants producing (correct) high- or (incorrect) low-probability items, half produced high-probability items and half low-probability items - even while preferring high-probability items in a recognition test. Only discrete familiarization sequences with isolated words yield memories of actual items. Turning to how specific learning situations affect statistical learning, we show that it predominantly operates in continuous speech sequences like those used in earlier experiments, but not in discrete chunk sequences likely more characteristic of early language acquisition. Taken together, these results suggest that statistical learning might be specialized to accumulate distributional information, but that it is dissociable from the (declarative) memory mechanisms needed to acquire words and does not allow learners to identify probable word boundaries.