Predicting Intrasentential Pauses: Is Syntactic Structure Useful?

Syntactic structure plays a role in prosodic phrasing, which in turn guides where speakers may pause during a sentence. To what extent does syntactic structure predict the locations and durations of unfilled, intrasentential pauses in conversational speech? Here we attempt to answer that question by using the word-aligned and syntactically annotated Switchboard corpus [3]. Automated binary classification revealed that while syntactic structure helps above and beyond lexical features such as POS, knowledge of constituent node labels was less useful than knowledge of the starts and ends of constituents. On the other hand, some POS and constituent labels were more predictive than others in predicting word boundary durations. Some labels indicated the presence of a disfluency; others picked out syntactic contexts such as between a noun phrase and a prepositional phrase, which was predictive of a lack of or shorter pause.