Classification of functional-meanings of non-isolated discourse particles in human-human-interaction

Abstract

To enable a natural interaction with future technical systems, not only the meaning of the pure spoken text, but also metainformation such as attention or turn-taking has to be perceived and processed. This further information is effectively transmitted by semantic and prosodic cues, without interrupting the speaker. For the German language we rely on previous empirically discovered seven types of formfunction- concurrences on the isolated discourse particle (DP) “hm”. In this paper we present an improved automatic classification-method towards non-isolated DPs in human-human interaction (HHI). We show that classifiers trained on (HCI)-data can be used to robustly evaluate the contours of DPs in both HCI and HHI by performing a classifier adaptation to HHI data. We also discuss the problem of the pitchcontour extraction due to the unvoiced “hm”-utterances, leading to gaps and/or jumps in the signal and hence to confusions in form-type classifications. This can be alleviated by our investigation of contours with high extraction completion grade. We also show that for the acoustical evaluation of the functional-meaning, the idealized form-function prototypes by Schmidt are not suitable in case of naturalistic HHI. However, the precision of acoustical-meaning prediction with our classifier remains high.