Domain generality versus modality specificity: the paradox of statistical learning

Abstract

Statistical learning is typically considered to be a domain-general mechanism by which cognitive systems discover the underlying distributional properties of the input. Recent studies examining whether there are commonalities in the learning of distributional information across different domains or modalities consistently reveal, however, modality and stimulus specificity. An important question is, therefore, how and why a hypothesized domain-general learning mechanism systematically produces such effects. We offer a theoretical framework according to which statistical learning is not a unitary mechanism, but a set of domain-general computational principles, that operate in different modalities and therefore are subject to the specific constraints characteristic of their respective brain regions. This framework offers testable predictions and we discuss its computational and neurobiological plausibility.

Generalization from newly learned words reveals structural properties of the human reading system.

“…The warping account we advance also offers a potential explanation for an important paradox in the statistical learning literature, which includesbut extends well beyondstudies of different facets of language learning (Frost, Armstrong, Seigelman, & Christiansen, 2015).…”

“…In addition, because PDP models are statistical learning systems, our data provide insight into an intriguing discrepancy in the statistical learning literature: why learning in different domains yields sometimes a high degree of generalization and sometimes stimulus-specific learning and minimal generalization (Frost, Armstrong, Seigelman, & Christiansen, 2015).…”

“…This is because the mappings between the structure of the surface form and these other representations is arbitrary in nature, as opposed to quasiregular (for discussion, see Plaut, 1997;Frost et al, 2015;McClelland et al, 1995).…”

Visual cortex signals a mismatch between regularity of auditory and visual streams

“…With respect to former, our current study fills a gap in current understanding, which has relied almost exclusively on studies of unisensory stimuli, and expands the scope of inquiry into the multisensory context. Furthermore, documenting a MMR signature would strongly inform a basic conundrum in the theoretical literature, which is whether there exist neural systems that are sensitive to regularity in multiple modalities (as reviewed in Frost, Armstrong, Siegelman, & Christiansen, 2015). With respect to theories of multisensory perception, our current study strongly bares on the question of whether there are brain systems that are sensitive to continuous, temporally extended features of multisensory streams.…”

“…Such results are consistent with behavioral findings, which yield little evidence for a single latent factor underlying sensitivity to statistics in auditory and visual streams . It also suggests that a domain-general capacity, if existent, would be subservient to modality-specific processing constraints (Frost, Armstrong, Siegelman, & Christiansen, 2015). However, all aforementioned studies share a core feature: they develop from, and evaluate theoretical models based on input streams presented within just a single modality -typically either auditory or visual.…”