Measuring phoneme boundaries four ways

In order to determine whether the concept of “phoneme boundary” has any validity independent of a particular experimental paradigm, four procedures were used to locate the boundary on two different acoustic continua: a VOT continuum. ranging from [ba] to [pha], and a formant transition continuum, ranging from [bæ] to [dæ]. One procedure consisted of a standard identification test (boundaries were later determined by regression). The PEST procedure determined the position of the phoneme boundary adaptively. In the up/down procedure, subjects simply indicated when ascending or descending series of stimuli crossed the boundary. In the fourth procedure, subjects adjusted the stimulus presented until it fell on the boundary. The procedures differed in reliability, and there were substantial individual differences in the locations of the phoneme boundaries, but the correlation between boundary locations for each subject found by the different procedures was quite high. This result suggests that, despite recent suggestions that phonetic categorization is susceptible to cognitive biases, phoneme boundaries do exist independent of the particular methods used to measure them. [Work supported by NSF, NIMH, NICHD.]