Sonya M. Sheffert
Haskins Labs., 270 Crown St., New Haven, CT 06511
Univ. of Connecticut
It is generally agreed that listeners have a lexicon where words are stored as abstract types, rather than token instances. This assumption has led many researchers to posit ``normalization'' processes that strip away all but the abstract linguistic information represented in lexical entries from the acoustic signal. However, recent research finds that idiosyncratic speaker information, such as voice, is not discarded. For example, word recognition is facilitated when a word-voice pairing is preserved across study and test, suggesting that voice information is represented in memory and can serve as a retrieval cue for lexical access [e.g., Palmeri et al. (1983)]. Three experiments explored the effects of voice and visible speaker change on spoken word recognition using implicit and explicit memory measures. Subjects watched a videotape of speakers producing words. When a word was repeated, the visible speaker or his or her voice were either the same or different from the first presentation. Subjects performed a word recognition task. Some subjects were also required to explicitly encode either the speaker's voice or face. The results demonstrate that detailed information about a talker's voice and appearance is not lost during normalization, but is retained alongside the word in memory. [Work supported by NIH Grant No. HD-01994.]