Effect of distributional shape on learning a target sound

Abstract

Vowel pronunciations (measured in Hz) form distributions that vary from normally distributed to quite skewed. Labov (2001) noted that vowels undergoing change tend to be skewed, while stable ones are not. We ask if the different distributions are causally related to change. We exposed participants (n = 238) to a positively skewed, negatively skewed, or normal distribution of pure tones varying in pitch (to mimic how vowels vary in quality). Participants were told they were listening to notes played by amateur musicians who’d been aiming for the same note. In each trial, participants listened to 20 tones then played the note they thought was the target. Output pitches were compared to the input. The question was whether learners would play a tone corresponding to the mean or the mode of the set or would instead “shift” the note. In the two conditions analyzed so far (normal and positively skewed), participants output pitches that were slightly higher than the mean of the input set, indicating shift in both conditions; however, the difference between conditions is not significant. There is also an effect of the final note in the set, evoking the well-established recency effect in memory. Analysis of the third condition is ongoing.