Magnuson, J. S.
University of Connecticut and Haskins Laboratories
Among the most important tools in the psycholinguist\'s kit are corpora, which allow us to estimate individual language experience from large-scale averages from many sources. Corpus estimates are the basis for crucial theoretical constructs such as frequency and neighborhood density, and analogous dimensions of computational models. However, there are obvious challenges in applying such estimates to understanding variation in language ability. For example, findings that children with Specific Language Impairment do not show typical effects of phonological neighborhood density in spoken word recognition may indicate theoretically interesting differences in lexical organization, but they might also reflect simple differences in vocabulary -- it would seem that this issue is unresolvable without comprehensively documenting each child\'s lexicon. An alternative is to use \'human models\' that put participants on maximally equal footing, as with artificial lexicon learning tasks. I discuss recent work my colleagues and I have carried out using this approach, and complementary computational and statistical modeling of the dynamics of lexical processing. In particular, I will focus on recent results with a rigorously assessed group of low-literacy adults and discuss the potential for individual differences in spoken language comprehension to illuminate difficulties with reading.