Issues in the measurement of perceptual assimilation

Abstract

This study examined the effect of methodological variables on the fit between predicted discrimination scores based on identification data and actual discrimination data in cross-language speech perception experiments. Such variables include (1) single versus multiple talkers in discrimination test trials; (2) different discrimination test types (e.g., AX, AXB, oddity); and (3) identification tests in which stimuli are presented individually versus stimuli being presented in the same context as they appear in discrimination tests. The optimal pair of identification and discrimination tests, yielding the best match between predicted and actual discrimination scores, can be used in subsequent studies examining perceptual category structure. These methodological variables were examined by presenting American English speakers with two Hindi contrasts, [b]–[p] and breathy voiced dental-retroflex, both in initial position and in an [i], [a], or [u] context. The stimuli appeared in a range of categorial discrimination and identification tests. Early results examining the third variable listed above demonstrate that identification tests that present stimuli in the same context as they appear in corresponding discrimination test trials correlate more strongly with discrimination scores (r=0.72, p<0.05) than identification tests that present stimuli in isolation (r=0.58, p<0.05).