Partially automatically annotated corpus to predict gestural cues in Embodied Conversational Agents

#Structure of the corpus

This corpus has been built using speeches of Spanish politicians freely available here along with their transcriptions.

Each transcription has been analyzed in terms of:

Surface Syntactic Structure*

Deep Syntactic Structure*

Morphology (Part of Speech)*

Communicative Structure

Gestures (beat vs. no gesture tags) have been annotated using the videos.

*All those features have been automatically retrieved using the parser freely available in https://github.com/TalnUPF/miis. The other features have been annotated manually.

#Concerns about the corpus

This corpus has been mostly annotated manually. Annotation agreement has not been computed.

Moreover, it is small. In order to extract reliable correlations from it, it should be extended.