Statistical inference in the perceptual learning of non-native speech category.
Yuan Zhao · The Journal of the Acoustical Society of America · 2008
The study investigated the influence of category bias and category variance in English speakers’ learning of two lexical tone categories (high versus low). In Experiment 1, all subjects completed an AX discrimination pretest of adjacent tone pairs on a ten-step pitch continuum. They were then trained to associate a “low tone” with the meaning “pink” and a “high tone” with the meaning “blue” via a picture-sound association task. Subjects received either a biased training with a 5:1 ratio between high and low tones or an unbiased training with a 1:1 ratio between the two categories. It was then followed by an identification post-test. The results showed that poor pitch discriminators categorized significantly more sounds into the biased category. Good discriminators categorized significantly more sounds into the other category. Therefore, good discriminators have a novelty preference, while poor discriminators have a familiarity preference. In Experiment 2, subjects received either a small or a big variance training. The results showed that poor discriminators performed significantly better if trained with a big-variance input. However, good discriminators’ categorization performance was significantly better if trained with an input of a small variance. The study suggests that statistical cues are utilized differently among subjects of different auditory sensitivities.