Influence of word size and tonal sequence probabilities on Mandarin segmentation errors

Abstract

The largely effortless process of segmenting a continuous speech stream into words has been shown cross-linguistically to be influenced by implicit knowledge about the distributional probabilities of phonotactics, stress assignment, and word size. Work in tonal languages provides compelling evidence that tones are also likely to be a source of probabilistic information for speakers of these languages. It has been shown that probability of syllable + tone combinations play a role in speech processing in Mandarin (Wiener & Ito, 2016; 2015; Wiener & Turnbull, 2013). Work in Cantonese has also demonstrated that transitional probabilities of lexical tones paired with vowels aid listeners in segmentation (Gomez, et al., 2018). Here we conducted two perception experiments with fifty native-Mandarin speakers and manipulated two potential segmentation cues: word size and tonal sequence probability. Contrary to our hypothesis that participants would make segmentation errors which reflect the most probable word size in Mandarin (two-syllable words) with highly probable tonal sequences, participants’ errors were overwhelmingly three-syllable words with low probability tonal sequences. This suggests that biases towards segmentation errors are not predictable based on straightforward probabilities.