The Influence of Parallel Processing on Illusory Vowels

Research has shown that listeners perceive illusory vowels inside consonant clusters that are not allowed in their L1. This phenomenon has been examined using several psycholinguistic and computational models, including hidden Markov models (HMMs), applied to human phoneme perception. However, the inference algorithm of HMMs assumes that parallel processing, which has not been proven to have psychological reality, is a valid cognitive process. This study tested the psychological reality of parallel processing by attempting to duplicate two results from previous studies: First, listeners perceive an illusory vowel in consonant clusters that are not permissible in their L1. Second, the illusory vowel is based on the characteristics of the preceding consonant, indicating that listeners integrate phonotactics and acoustic information. The experiment manipulated the number of candidates that the model can refer to, and the algorithm can be considered parallel when it allows models to use more than two candidates that are stored in memory. In addition, the transition probabilities between consonants were manipulated to represent the different phonotactics. The results showed that only the parallel processing condition reproduced the two observations above, supporting the psychological reality of parallel processing.