The Relative Weight of Statistical and Prosodic Cues in Speech Segmentation: A Matter of Language-(In)dependency and of Signal Quality

Abstract

In an artificial language setting, we investigated the relative weight of statistical cues (transitional probabilities, TPs) in comparison to two prosodic cues, Intonational Phrases (IPs, a language-independent cue) and lexical stress (a language-dependent cue). The signal quality was also manipulated through white-noise superimposition. Both IPs and TPs were highly resilient to physical degradation of the signal. An overall performance gain was found when these cues were congruent, but when they were incongruent IPs prevailed over TPs (Experiment 1). After ensuring that duration is treated by Portuguese listeners as a correlate of lexical stress (Experiment 2A), the role of lexical stress and TPs in segmentation was evaluated in Experiment 2B. Lexical stress effects only emerged with physically degraded signal, constraining the extraction of TP--words to the ones supported by both TPs and IPs. Speech segmentation does not seem to be the product of one preponderant cue acting as a filter of the outputs of another, lower-weighted cue. Instead, it mainly depends on the listening conditions, and the weighting of the cues according to their role in a particular language. Adapted from the source document