Towards an understanding of tone category variability in Cantonese

Cantonese is typically described as having 6 lexical tones. There are reports, however, of three mergers-in-progress; the contrasts between Tone (T) 2 and T5 [Baueret al., LVC 15(2), 211–225 (2003)], T3 and T6, and T4 and T6 [Moket al., LVC 25(3), 341–370 (2013)] are becoming neutralized. Previous work examined the perception of merging tone pairs on an 11-step continuum, in addition to non-merging control pairs (T2-T3 & T5-T6), finding more discrete categorization performance for non-merging than for merging tone pairs. In this study, we examine these same Cantonese tone pair continua with a goodness rating paradigm to understand category-internal structure. Listeners rated the goodness of items on a 7-point scale with real word landmarks as endpoints. Listeners generally rated items across the continua in merging pairs as more acceptable. Items from non-merging tone continua presented a more curious pattern, suggesting that listeners perceived the mid-points of continua as the best-sounding or worst-sounding items. In an attempt to understand these results and those of a previous categorization task, we explore acoustic-auditory similarity measures (e.g., dynamic time warping, Frechet distance, area between curves) between the tones, comparing these measures to listener behaviour.