Further modeling of the effects of lexical uniqueness in speechreading: examining individual differences in segmental perception and testing predictions for sentence level performance

Two experiments are reported that extend previous investigations of the influence of the lexicon on visual spoken word recognition. Experiment 1 investigated individual differences in segment level perception. Behavioral identification of consonants and vowels presented in either consonant-vowel or consonant-vowel-consonant nonsense syllables were collected from participant groups chosen to differ in their sentential speechreading ability (6 – more accurate versus 6 – less accurate). The results demonstrate that the two groups differed in vowel identification accuracy, but not in consonant identification accuracy. The groups also differed in the patterning of confusions between both consonant and vowel segments. This resulted in different degrees of modeled uniqueness and expected lexical equivalence class sizes between the two groups. Experiment 2 extended the lexical modeling approach [1] to sentence length stimuli. A set of sentences was developed that comprised key words predicted to be easy, moderate, or difficult to recognize. 123 (41easy, 41-moderate, 41-difficult) spoken sentences were presented visual-alone to 40 participants for open set identification. 301 keywords spoken in isolation were presented visual-alone to 23 participants for open set identification. Identification accuracy varied as a function of predicted difficulty for both words spoken in isolation and within sentence context. The effect of predicted difficulty on identification accuracy interacted with presentation context (sentence versus isolation). These results suggest that the advantage enjoyed by words predicted to be easy to recognize in isolation is reduced when those words are spoken in sentential contexts.