Surprising Evidence for Lexical Specificity in Selective Attention
 
Word learning is a particularly challenging form of category learning. Learners 
must recognize a word despite substantial surface variation from things like 
talker, speaking rate and ambient room acoustics, while also differentiating 
it from countless other, similar-sounding words. These factors make words an 
example of extremely dense categories; there is large between-category overlap 
and substantial within-category variance. Despite the intriguing structure of 
lexical categories, little work has considered how word learning relates to 
other forms of category learning. Is category learning domain general? Or do 
major domain differences between visual and auditory stimuli elicit unique 
forms category learning? We conducted three experiments to investigate how 
category learning deploys selective attention when adults learn lexical 
categories defined by novel features. The first experiment showed that 
attention is deployed to features that are relevant for categorization, at 
the expense of irrelevant features, in line with how selective attention 
operates in the visual domain. However, the subsequent experiments showed 
that these effects were highly context specific; they did not generalize to 
novel voices, nor to novel words; in the visual domain, selective attention 
effects generalize readily across contexts. These results suggest that 
categorization does not operate isomorphically across domains. I discuss 
several possible sources of domain differences, as well as deeper implications 
of this finding.