Using Prosody to Classify Discourse Relations

JUAN MANUEL PÉREZ; JANINE KLEINHANS; CATHERINE LAI; MIREIA FARRÚS; LEO WANNER; GRAVANO, AGUSTÍN

Estocolmo

Conferencia; Interspeech 2017; 2017

International Speech Communication Association

This work aims to explore the correlation between the discoursestructure of a spoken monologue and its prosody by predictingdiscourse relations from different prosodic attributes. For thispurpose, a corpus of semi-spontaneous monologues in Englishhas been automatically annotated according to the RhetoricalStructure Theory, which models coherence in text via rhetori-cal relations. From corresponding audio files, prosodic featuressuch as pitch, intensity, and speech rate have been extractedfrom different contexts of a relation. Supervised classificationtasks using Support Vector Machines have been performed tofind relationships between prosodic features and rhetorical rela-tions. Preliminary results show that intensity combined withother features extracted from intra- and intersegmental envi-ronments is the feature with the highest predictability for adiscourse relation. The prediction of rhetorical relations fromprosodic features and their combinations is straightforwardlyapplicable to several tasks such as speech understanding or gen-eration. Moreover, the knowledge of how rhetorical relationsshould be marked in terms of prosody will serve as a basis toimprove speech synthesis applications and make voices soundmore natural and expressive.