Event-related potentials and the categorical perception of speech sounds

To determine whether there are physiological correlates of categorical perception.

Human evoked potentials were recorded in response to computer-modified speech sounds from a nine-stimulus continuum between /ba/ and /da/. In the first experiment, subjects listened to trains composed of 52% /ba/ or /da/ and 6% of each of the other eight stimuli and classified the stimuli as "ba" or "da." In the second experiment, subjects read a book and ignored trains containing a standard stimulus (p = 80%) and two deviant speech sounds (p = 10% each), one within the same category as the standard and the other across the category-boundary. The third experiment was similar to the first except that the subject was reading. The fourth experiment compared the responses to stimuli that deviated from standards in terms of their phonemic category or intensity.

An N2-P3 complex was evoked by those stimuli in the more improbable category when the stimuli were attended to in the first experiment. In the second and third experiments, there was a clear mismatch negativity (MMN) for the across-category deviant stimuli when the standard stimulus came from the /ba/ end of the continuum. However, when the standard stimulus came from the /da/ end of the continuum, there was no definite MMN. The overall frequency-content of our /da/ stimulus was broader than that of the /ba/ stimulus. A deviant stimulus from the /da/ end of the continuum thus contained frequencies which were not present in the /ba/-standard stimuli and these frequencies could elicit a MMN. In the fourth experiment the MMN evoked by a small change in intensity was much larger than that evoked by a change in phonemic category.

The N2-P3 complex accurately reflects the phonemic categorization of speech stimuli. The MMN evoked by changes in speech sounds may indicate the detection of acoustic rather than phonetic changes.