Search

This dissertation reports a series of three experimental studies that investigated how variation in speech clarity and intelligibility as well as the presence of semantic context affects listeners’ speech segmentation. An artificial language learning experiment in Study 1 (Chapter 2) showed that speaking clearly relative to speaking conversationally improved segmentation of nonsense words by statistical learning. However, the improvement was observed only in the quiet listening condition but not in noise. Using the visual-world eye-tracking paradigm, Study 2 (Chapter 3) examined the clear speech segmentation benefit during real-time processing of meaningful sentences in which the target word was temporarily ambiguous with a competitor across a word boundary. The results revealed that that relative to conversational speech, clear speech facilitated target word segmentation even before the target and competitor could be disambiguated based on phonemic information. The facilitation not only emerged in quiet but also extended to the noisy listening condition. Finally, built upon Study 2, Study 3 (Chapter 4) employed eye-tracking to further explore how such clear speech facilitation effect is modulated by semantic cues from the preceding context. It was found that while the clear speech segmentation benefit was eliminated when the context already biased listeners towards the target, it was still present when the context favored the unintended competitor. Taken together, the key findings from the three studies advance the understanding of the relative importance of signal-dependent and signal-independent sources of information during segmentation in realistic communicative settings. The results also provide novel insight into the well-documented clear speech processing benefits by demonstrating that improved segmentation may in part underlie these benefits. The dissertation also has theoretical implications, suggesting directions for refining the current spoken word recognition and segmentation models.