Abstract:

Lipreading is thought to be extremely difficult, and the difficulty is attributed to lack of perceptual distinctiveness among most visible spoken words. Therefore, it is thought that accurate lipreading requires reliance on postlexical processes, such as guessing. However, we [E. T. Auer and L. E. Bernstein, J. Acoust. Soc. Am. 102, 3704--3710 (1997)] predicted, based on computational modeling of the lexicon, that many words maintain their visual distinctiveness. In this study, we took only visual phoneme confusability and word frequency into account to predict the relative intelligibility of lipread words. We predicted that words without visual lexical competitors can be perceived as readily as heard words, and a systematic change in intelligibility would be observed as a function of word frequency and number of competitors. Predictions of near-100% accuracy for many words were confirmed in a perceptual word identification experiment with skilled deaf American lipreaders. Accuracy was not influenced by word length. When phoneme confusability in words was statistically controlled, the number of lexical competitors correlated significantly with word score, as did word frequency. These results will be compared with those of hearing participants, and the implications discussed with attention to bottom-up processes versus lexical competition in spoken word recognition. [Work supported by NIH/NIDCD 02107.]