Temporal predictability in speech: Comparing statistical approaches on 18 world languages

Temporal predictability in speech: Comparing statistical approaches on 18 world languages

Yannick Jadoul, Andrea Ravignani, Bill Thompson, Piera Filippi, and Bart de Boer

Artificial Intelligence Lab, Vrije Universiteit Brussel

Temporal regularities in speech, such as interdependencies in the timing of speech events, are often thought to scaffold early acquisition of the building blocks in speech: by providing on-line clues to the location and duration of upcoming syllables, temporal structure may aid segmentation and clustering of continuous speech into separable units. This hypothesis tacitly assumes that learners exploit predictability in the temporal structure of speech. Here, we test whether syllable occurrence is predictable over time. Existing measures of speech timing: (i) tend to focus on first-order regularities among adjacent units, and (ii) are overly sensitive to idiosyncrasies in the data they describe. Instead we pursue a two-pronged strategy to quantify predictability in a sample of 18 languages, integrating several statistical methods. First, we analyse distributional regularities using two novel techniques: a Bayesian ideal learner analysis, and a maximally simple distributional measure that nevertheless correlates with the common, more complex measure nPVI. Second, unlike previous approaches, we model higher-order temporal structure – regularities that arise in an ordered series of syllable timings – testing the hypothesis that non-adjacent temporal structures may explain the gap between subjectively-perceived temporal regularities, and the absence of universally-accepted lower-order objective measures. Together our analyses provide weak evidence for predictability at different time scales, though it is difficult to reliably infer predictability at higher-orders. We conclude that any temporal predictability in speech may arise from a combination of individually weak perceptual cues at multiple structural levels, but is challenging to pinpoint with confidence at any particular locus.