Effect of contextual tonal variation on speech recognition: Evidence from eye movements

While much is known about how lexical tones are perceived in isolation, little has been done on the perception of tones in connected speech. This study, via visual world paradigm, investigates effect of contextual tonal variation on speech recognition in Tianjin Mandarin, where three types of contextual tonal variation have been identified: Near-Merger Sandhi, No-Merger Sandhi, and No-Sandhi Coarticulation. Listeners were asked to identify the target amongst an array of four possibilities upon hearing a disyllabic collocation, while their eye movements were tracked. Results suggest that native listeners are sensitive to fine-grained phonetic details in contextual variation of lexical tones. NoSandhi Coarticulation was the easiest to recognize as participants fixated on the targets the earliest among three conditions. Near-Merger Sandhi was more difficult to process than No-Merger Sandhi, reflected in the overall less proportion of looks to target.