ISCA Workshop on Plasticity in Speech Perception (PSP2005)
Senate House, London, UK
Japanese vowel length distinction is phonemic and is difficult for adult native English (NE) speakers to acquire. A Japanese long vowel is longer in duration than a short vowel by definition (with little difference in formant frequencies), but a long vowel spoken quickly can be shorter than a short vowel spoken slowly. Native Japanese speakers accurately identify length of vowels at varied rates by using the rate of the sentence as a perceptual cue, i.e., by way of rate normalization (Hirata and Lambacher, 2004). However, it is not known to what extent NE speakers can normalize rate and accurately identify Japanese vowel length at varied rates. If rate normalization is part of a general auditory process (Pisoni et al., 1983), and therefore, to occur even for beginning second language (L2) learners, their identification accuracy would not differ between sentences spoken quickly and slowly. However, if rate normalization is not an automatic but a learned process, beginning L2 learners' accuracy might differ between different speech rates, perhaps biased by absolute vowel duration.
Participants were sixteen NE speakers who had studied Japanese for fifteen weeks in the U.S. Stimuli were three types of Japanese disyllabic nonwords, /mVmV/ (two short vowels), /mV:mV/ (long and short vowels), and /mVmV:/ (short and long vowels) spoken in a carrier sentence at fast and slow rates by a native Japanese speaker. All the three word types had a pitch accent on the first syllable. The stimuli with two rates and three word types were all mixed and randomized. Participants choose one of the four alternatives: short+short, long+short, short+long, and long+long. Results showed no main effect of rate: the overall identification accuracy did not significantly differ between the fast (69.8%) and slow rates (74.2%). Thus, participants were overall able to adjust their identification according to the rate of stimuli. The effect of word type was significant: long+short type (86.9%) > short+short type (72.9%) > short+long type (56.1%). The lowest accuracy for the short+long type was probably because NE speakers were biased to perceive unaccented vowels as short (Minagawa et al., 2002). A significant interaction of rate and word type was found: only for the short+long type, accuracy was higher for the slow (65.8%) than the fast rate (46.5%). This implies that rate normalization is not an automatic process for beginning L2 learners when dealing with a very difficult word type, and that slower rates help for accurate identification.
Bibliographic reference. Hirata, Yukari (2005): "Effects of speaking rate on native English speakers' identification of Japanese vowel length", In PSP2005, 173 (abstract).