The Effect of Lexical Distribution and Similarity Avoidance in Speech Segmentation

This thesis intends to compare the influence of lexical frequency with the influence of a constraint on statistical learning. There is evidence of a constraint restricting the occurrence of similar consonants in sequence in Dutch (Kager & Shatzman 2007). This constraint has been noted in various languages (e.g. Arabic, Hebrew, English, Japanese), suggesting that the constraint is not language-specific, and not based on lexical frequency. Arguably, the effects of the similarity constraint that these studies find are attributable to the subjects’ subconscious knowledge of lexical distribution. This argument can be backed by the evidence of human capacity for statistical and phonotactic learning and their ability to apply that to artificial language learning (e.g. McQueen 1998, Peña et al. 2002, Newport & Aslin 2004, Vitevitch & Luce 2005, Onnis 2005). Three experiments with Dutch native speakers attempt to tease apart the effect of the similarity constraint and the effect of under-representation in the lexicon. Ultimately, the results do not show a difference between the influence of lexical distribution and the influence of the phonotactic constraint on similarity. They do show the influence of phonological features for perceived similarity. Especially the feature [nasal] was seen to set apart the labial consonant /m/ from the other labials in this study. Clearly, a focus on specific elements in the speech input enables the learner to detect regularities, and to extract any rules more efficiently. In this fashion, phonological features of the phonemes influence statistical learning, and demonstrate how statistical learning interacts with linguistic properties of language