Previously in the Gow Lab…
The Gow Lab started out with a focus on understanding how the structure of the speech signal facilitates language processing. Early work explored the interaction between word-initial acoustic cues to word boundary and the competitive dynamics of spoken word recognition. This led to the development of the Good Start Model, a model of lexical segmentation. According to this model, word boundary cues tend to enhance word-onset feature contrasts, creating islands of perceptual reliability that give early processing advantages to speaker’s intended words, and making it hard for unintended potential embedded words (e.g. the go in cargo) to compete for lexical activation. Later work incorporated evidence from phonology that word onsets tend to protected from phonological change across human languages.
For several years our work turned to the problem of how listeners recognize words that have undergone systematic phonological change. We wanted to know what role phonological representation or perhaps implicit knowledge of phonological rules played in listeners’ ability to compensate for lawful phonological alternations. We focused on the process of coronal place assimilation, a process found in English in which words that end with consonants with coronal place of articulation take the place of a following word-initial consonant. For example, the final /n/ green may sound like an /m/ in the phrase green beans, or an /ng/ in the phrase green kite. How could a listener who seldom confuses teen and team, recognize green prounounced greem? Our work showed that that assimilation actually facilitates word recognition by allowing listeners to anticipate upcoming words. Further behavioral and acoustic-phonetic work in English, Hungarian and Korean showed that English coronal assimilation is non-neutralizing, and that the processing of assimilated speech is a function of the nature of the degree to which phonological processes neutralize phonetic contrasts. The work collectively pointed the problem of feature cue parsing , or associating phonetic cues with individual segments.
In both of these lines of work we were interested in the potential role of top-down lexical influences on speech perception. While a number of experiment results clearly demonstrated that lexical knowledge influences phoneme interpretation, most of those results are consistent with either interactive mechanisms (in which knowledge of words alters phonetic interpretation) or modular mechanisms (in which phonetic and lexical analyses are only combined at a post-perceptual decision phase). We introduced effective connectivity analyses as a way to distinguish between these accounts. In a first study, we examined the Ganong effect, we found that lexical influences on speech categorization effects are accompanied by increased top-down influence of wordform areas on acoustic-phonetic areas. In related work, we found the same patterns of interaction accompanying lexical influences on the interpretation of assimilated speech, and sentential context effects on speech perception. The assimilation study also provided direct evidence of motor mediation of the perception of heavily coarticulated speech in the form of influence by premotor cortex on the superior temporal gyrus.