Perceivable information structure in discourse prosody-Detecting prominent prosodic words in spoken discourse using F0 contour

Abstract

The present study is an attempt to show how information structure as well as discourse structure is represented via prosodic patterns in continuous speech through F0 features. The complementary relationship found between information and discourse structure reflected by prosodic feature F0 can account for prosodic contribution towards speech understanding. We assume that perceived emphases or foci (prominence) are important information assigned by information structure and marked by hump peaks in the F0 contour by prosodic units. These F0 peaks are first compared with locations of linguistic units, lexical entries (words) and prosodic units (the prosodic words PWs), respectively, to decide optimized units representing allocation of key information. While the PW is defined as perceptually identifiable units at the lowest-level in a prosodic hierarchy of spoken discourse, higher-level location consistency between PWs and information arrangements operates via prosodic units that are larger than words, suggesting that the PW is a plausible unit to derive key information. The information foci in PW units are further detected and compared before and after considering higher-level context of discourse prosody. Detection accuracy of information foci is significantly improved after removing contributions from discourse context across different speech genres and languages (English and Mandarin). Specifically how F0 peaks are correlated to key information content. The findings thus shed lights on how and why prosodic features significantly contribute to speech understanding, and at the same time imply how such findings could be applied to enhance technological development.