Statistical learning in infants
Open Access
- 18 November 2002
- journal article
- editorial
- Published by Proceedings of the National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 99 (24) , 15250-15251
- https://doi.org/10.1073/pnas.262659399
Abstract
Statistical learning has become the subject of some considerable debate within cognitive psychology (1, 2). The debate reached particular prominence with the advent of sophisticated computational models of such learning based on neurally inspired notions of spreading activation within highly distributed systems of interacting units, so-called neural networks (3). One important development in this field was the notion that representations of higher-level structure might “emerge” on the basis of an initial sensitivity to low-level statistical cooccurrence phenomena (4). Thus, within the field of language research, higher-level theoretical constructs such as the grammatical class of a word (as noun or verb, for example) would emerge within a system that was sensitive only to the statistical distributions of words within sentences. Importantly, this emergence represented little more than simple statistical clustering; the internal representations of words that would tend to occur in similar distributional contexts would cluster together, and because nouns tend to occur in particular sentential contexts and verbs in others, the clustering of words into these two classes (and others with even finer distinctions between the classes) was in some sense a statistical inevitability. One important feature of such models was that although learning within these models was based on a sensitivity to statistically predictable variation in their input, beyond that, the real-world nature of that input (whether pertaining to words, sounds, letters, or shapes) did not matter; the experimenter might deem a particular pattern of activation across the “input units” to represent a particular kind of linguistic stimulus, but these inputs could be deemed just as easily to represent visual stimuli. It is thus significant that Fiser and Aslin (5) have demonstrated the sensitivity of infants to statistical properties of visual input. Evidence of a statistical underpinning to aspects of cognition has provoked considerable controversy with respect to …Keywords
This publication has 19 references indexed in Scilit:
- RETRACTED: Rule learning by cotton-top tamarinsCognition, 2002
- Learning and development in neural networks – the importance of prior experienceCognition, 2002
- Rule Learning by Seven-Month-Old Infants and Neural NetworksScience, 1999
- Rule Learning by Seven-Month-Old InfantsScience, 1999
- Rethinking Eliminative ConnectionismCognitive Psychology, 1998
- Tracking the Time Course of Spoken Word Recognition Using Eye Movements: Evidence for Continuous Mapping ModelsJournal of Memory and Language, 1998
- On the Inseparability of Grammar and the Lexicon: Evidence from Acquisition, Aphasia and Real-time ProcessingLanguage and Cognitive Processes, 1997
- Thematic Roles as Verb-specific ConceptsLanguage and Cognitive Processes, 1997
- Statistical Learning by 8-Month-Old InfantsScience, 1996
- Modality independence of implicitly learned grammatical knowledge.Journal of Experimental Psychology: Learning, Memory, and Cognition, 1995