Parallel processing of auditory and phonetic information in speech perception. II. Application of a probabilistic model

Abstract

At the last meeting of the Society I presented data suggesting that auditory and phonetic information in a speech signal can be processed in parallel. In that experiment, subjects identified either an auditory dimension (fundamental frequency) or a phonetic dimension (place of articulation of the consonant) of synthetic consonant-vowel syllables. Reaction times (RTs) were significantly shorter when the two dimensions varied in a completely correlated manner (correlated condition) than when each dimension varied alone (control condition). The purpose of this paper is to apply a simple probabilistic model to these results. The model makes two assumptions: (a) the component processes for auditory and phonetic dimensions can operate simultaneously, with the durations of each component process distributed statistically over trials; and (b) the subject's response on each trial in the correlated condition is based upon whichever component process is completed first on that trial. Equations based on this model accurately predict both the magnitude of the improvement and the form of the RT distributions in the correlated condition from the empirical RT distributions in the control conditions. In addition, the model makes precise quantitative predictions concerning the effects of (a) varying the relative discriminability of the two dimensions, and (b) adding additional redundant dimensions. The empirical validity of these additional predictions is currently under investigation.