Prosody, Discourse and Syntax in French Conversations: Resource creation and Evaluation

Overview & Synopsis

How syntax, prosody and discourse do interface with each other is a recurrent question in modern linguistics. While strong inter-dependencies have been established for read speech, the question is more challenging in the context of truly spontaneous conversational speech. To our best knowledge, there is not yet a reliable resource for addressing this question from a quantitative viewpoint, at least for French language. We present in this paper the process of creation of such a resource. We started from the Corpus of Interactional Data (CID) and organized an annotation campaign for marking prosodic phrasing, discourse segmentation and disfluencies tagging. Syntactic tagging and chunking have been produced automatically and the former had been manually corrected. We describe here the methodology, the challenges encountered and the results of the annotation campaign. Moreover, we evaluate the campaign and propose a reference dataset from the annotations produced. All the resources described are available trough Ortolang perennial language archive repository.