Consonant confusability and its relation to phonological similarity

Abstract

Gradient similarity avoidance patterns in Bengali echo reduplication suggest that the most similar consonants to /t/ are, in order, /t, th, d, t, s, th, k, .../. Converted to confusability, this ranking predicts that aspiration is most confusable, followed by voicing, minor place, continuancy, major place, and sonority. To confirm this, 24 native speakers identified syllables masked with babble, noise, or nothing (“clear”). Results indicate that confusability reflects similarity as predicted by avoidance patterns. In clear speech, most errors involved voicing and aspiration. Other errors reflected dialect-specific alternations. Noise introduced the percept of a loud burst: fricatives were often heard as affricates, non-alveolar coronals as alveolars, and non-coronals as coronals. Babble largely resembled noise. This pattern suggests that voicing is the most confusable feature, followed by aspiration, minor place and continuancy, major place, and lastly sonority. This ranking seems Bengali-specific, as studies of English find place to be significantly more confusable than manner and voicing. These results suggest that at least for Bengali, phonological alternations and perceptual confusability are argued to be better representations of how speakers judge similarity, rather than patterns in the lexicon or metrics such as shared natural classes metric of Frisch et al. (2004).