Statistical learning as a domain-general mechanism of entrenchment.

Abstract

The birth of the statistical learning literature is often traced back to Reberʼs (1967) seminal study on implicit learning using an artificial grammar learning paradigm. However, to fully understand the relationship between such early implicit learning studies and the current notion of statistical learning, it is important also to consider its conception. The theory of perceptual learning by J. J. Gibson and Gibson (1955) paved the way for accounts of learning with a basis in sensory experience. In the Gibsonsʼ theory of perceptual learning, which has close parallels to current ideas about entrenchment (Schmid, 2007), repeated experience with a percept enhances oneʼs ability to discriminate between it and other percepts. This chapter argues that a communicative system characterized by entrenchment, as posited in this volume, likely relies to a considerable extent on the ability to track, learn, and use underlying associative relationships between linguistic elements and structures in comprehension and production. When considering the origin of statistical learning as a theoretical construct, it is also important to consider the early work of Miller and Selfridge (1950), who thought that a reliance on transitional probabilities may be similar to the way in which grammar is learned. Other research informed by both Millerʼs work and the theory of perceptual learning espoused by J. J. Gibson and Gibson (1955) demonstrated that frequent co-occurrence due to underlying structure improved participantsʼ recall of letter sequences (Miller, 1958) and that learning the positional relationships between linguistic units (i.e., morphemes) occurs as an experiential process of familiarization with the temporal positions in which such units are frequently encountered (Braine, 1963). This laid the foundation for future research investigating the close relationship between frequent co-occurrence and the strength and automaticity of recall at various levels of linguistic analysis. From the beginning, research on implicit learning related to language was focused on the way(s) in which units of linguistic information are formed. Some of the early explanations for the ways in which this learning happened relied on experiencebased accounts, as just described. However, experience-independent theories of language acquisition quickly became the dominant perspective primarily because of the widespread acceptance of the "poverty of the stimulus" argument (Chomsky, 1965; Crain, 1991). Saffran, Aslin, and Newportʼs (1996) research gave the psychology of

Deficits in statistical leaning of auditory sequences among adults with dyslexia

“…Although the exploration of performance in visual AGL is quite extensive, the investigation of the deficit in implicit sequential learning in other modalities of sequential stimuli in AGL task is still at its infancy. The focus on one modality is based on the original assumption of implicit sequential learning as a modal (Jost & Christiansen, ; Reber, ), an is in line with evidence among TD participants which reported above chance learning similar to the results observed in the visual AGL task in various auditory stimuli (spoken syllables (Gomez, ), spoken letters (Manza & Reber, ), sung syllables (Schön et al, ), pure tones (Altmann et al, ; Saffran, Johnson, Aslin, & Newport, ), and musical timbres (Tillmann & McAdams, ). Nevertheless, studies that focused on the effect of modality on performance in an AGL task showed that implicit sequential learning is indeed affected by the modality of the stimuli.…”

Mini Pinyin: A modified miniature language for studying language learning and incremental sentence processing

“…They also highlight that BROCANTO has provided a useful window onto the neurobiology of grammar learning; however, given that BROCANTO is an artificially constructed language withtypicallyno associated meaning, it is not possible to compare behavioural and neural findings with those reported in native speakers. Further, differences in language learning may be explained by more domain-general mechanisms, such as statistical learning ability (Erickson & Thiessen, 2015;Frost, Armstrong & Christiansen, 2019;Jost & Christiansen, 2017), over and above that of LAA.…”

Exploring and Exploiting Uncertainty: Statistical Learning Ability Affects How We Learn to Process Language Along Multiple Dimensions of Experience

“…Over the past two decades, evidence has amassed documenting the extent to which language cognition depends on learning spatiotemporal regularities in a probabilistic manner (Jost & Christensen, 2017). Recent studies highlight the existence of individual differences within this ability (Kidd, Donnelly, & Christiansen, 2018) and discuss the possibility that statistical learning might not be a unitary learning mechanism but a set of domain‐general computational principles that operate in different modalities (Frost, Armstrong, Siegelman, & Christiansen, 2015).…”