Universality versus language-specificity in listening to running speech

Abstract

Recognizing spoken language involves automatic activation of multiple candidate words. The process of selection between candidates is made more efficient by inhibition of embedded words (like egg in beg) which leave a portion of the input stranded (here, b). Results from European languages suggest that this inhibition occurs when consonants are stranded but not when syllables are stranded. The reason why leftover syllables do not lead to inhibition could be because they might in principle themselves be words; in European languages, a syllable can be a word. In Sesotho (a Bantu language), however, a single syllable cannot be a word. We report that, in Sesotho, word recognition is inhibited by stranded consonants, but stranded monosyllables produce no more difficulty than stranded bisyllables (which could be Sesotho words). This suggests that the viability constraint which inhibits spurious embedded word candidates is not sensitive to language-specific word structure, but is universal.