Vowel Acquisition Based on an Auto-Mirroring Bias with a Less Imitative Caregiver

摘要

Regardless of interaction with less frequent imitative caregivers, infants can obtain the vowels of the caregivers' mother tongues by finding the correspondence between their own vowels and the caregivers' vowels. This paper proposes a learning method based on auto-mirroring bias (AMB) with a self-evaluation mechanism to find such correspondence. AMB is the robot's anticipation of being imitated by its caregiver and has a role of narrowing the candidates for the correspondence. The self-evaluation mechanism biased by AMB works as outlier (incorrect mapping) rejection, expecting that the outliers appear less consistently than the correct mappings do in the interaction. Results from several computer simulations with real sound wave recording from a human experimenter show that the robot could successfully achieve being imitated by the proposed method even if interacting with a caregiver who would seldomly imitate its utterances.