Statistical learning abilities and their relation to language

Abstract

Numerous studies on statistical learning (SL) have demonstrated humans' sensitivity to complex statistical properties in their sensory environment. These observations have had a profound impact on the study of language, highlighting statistical aspects of the linguistic input that can be learned from experience, leading to the widespread claim that SL plays a key role in language acquisition and processing. But how can this theorized link be experimentally demonstrated? One increasingly popular avenue comes from studies of individual differences, which tie individual variability in SL to variance in linguistic behavior. This review presents the theoretical advances stemming from this line of research, as well as some of the challenges it currently faces. It contends that while previous studies had an important role in establishing the existence of some coarse‐grained link between SL and language, recent developments in SL research suggest that the exact nature of this relationship is more complex than originally conceived and is still far from being fully understood. I specifically discuss three outstanding challenges: (a) understanding individual differences in light of the componential nature of SL, (b) mapping the full array of SL processes given the complexity of real‐world statistics, and (c) estimating the strength of current empirical evidence while taking into account both positive and null findings. Confronting these issues, I argue, is a necessary step towards a full theory of the role of SL across language.

Short-Term Memory for Serial Order Moderates Aspects of Language Acquisition in Children With Developmental Language Disorder: Findings From the HelSLI Study

“…Recently, SL has also been shown to affect the accuracy of serial short-term memory tasks: better serial recall was seen for sequences that had first been included in an SL task (Isbilen et al, 2020). As SL has been shown to correlate with developmental language measures (for a review, see Siegelman, 2020) and has been suggested to play a role in DLD (Mainela-Arnold and Evans, 2014), possible modality-specific differences in the learnability of regularities in temporal SL sequences could also affect STM for serial order. From chunking explanations follows also an alternative hypothesis, i.e., that serial binding capacity reflected in STM tasks could constrain the ability of the SL mechanism to chunk the stream of incoming stimuli to build units in LTM.…”

Statistical and sequence learning lead to persistent memory in children after a one-year offline period

“…Detecting and extracting various kinds of regularities embedded in our environment is a fundamental component underlying human learning in all ages, enabling us to adapt to our surroundings and to predict future events 1 – 4 . Extraction of regularities is argued to be the basis of several motor and cognitive skills, including language 3 , 5 – 9 . The initially unstable representations of the detected and extracted regularities are converted into a more stable form via consolidation, allowing information to be preserved and retained later 10 .…”

Autism, constructionism, and nativism

“…Statistical learning should most probably be thought of as a domain-general capacity to detect the distribution of structural properties, which applies but is not limited to the processing of linguistic input. Even though the exact nature and scope of the role played by statistical learning in language acquisition remains an open question, probabilistic extraction of structural regularities is likely to play an important role in language development (Siegelman & Frost 2015, Frost et al 2019, Siegelman 2020. Using slightly different methods, two relatively recent meta-analyses concluded that there is an absence of impairment in implicit statistical learning in autism (Foti et al 2015, Obeid et al 2016.…”