Wednesday, December 19, 2012

Variability and Detection of Invariant Structure

Variability and Detection of Invariant Structure. Gomez, R. L. (2002). Psychological Science, 13, 5.

Recent theories of language learning suggest that language learners are particularly adept at tracking the statistical regularities that exist within natural language. For example, learners are able to sensitive to “adjacent dependencies”, that is, the higher probabilities that occur between adjacent syllables within words than between words. However, a key aspect of language learning involves learning patterns between “non-adjacent dependencies”. Some examples of nonadjacent dependencies experienced in everyday language use are the dependencies between auxiliaries and inflectional morphemes (e.g.: is running), and dependencies involving agreement (e.g.: the kids in the park are playing). What is of interest to researchers is how humans learn to track these nonadjacent relationships.

In an experimental task involving learning nonadjacent dependencies, subjects listen to a string of syllables and must learn that a particular non-adjacent pattern – for example, that a first element always occurs with a third element in the string. A key hypothesis in this learning regards the middle element. If the middle element varies more, the pattern between the first and third element becomes more salient and the learner shows greats sensitivity to learning the nonadjacent dependency.

In this study, adults and infants were exposed to an artificial language containing nonadjacent dependencies between words (e.g.: aXd, bXe, cXf). The set size of X varied between conditions, drawing X from a set size of either 2, 6, 12, or 24 elements. Learning was greater for when the set size from which X was drawn was highly variable (set size = 24). These results demonstrate that learning is adaptable may be dynamically guided by the statistical structure in their linguistic environment. It should be noted that these results have not been replicated in all subsequent studies.

Statistical learning in children with language impairment has been investigated only recently (see blogpost from Evans et al., 2009). It may be that these children have more difficulty recognizing such patterns in language. Current interventions involving explicitly teaching linguistic rules may facilitate this learning.

Blogger: Nicolette Noonan

No comments:

Post a Comment