Conversation Understanding Through Belief Interpretation Sociolinguistic Modeling (CUBISM)
Abstract
The CUBISM project aimed to bring together research related to dialogue understanding along two different analytical dimensions, namely (1) participants mental content, and (2) participants social roles and relationships. With respect to mental content, we originally intended to extend the ViewGen belief ascription system (Wilks and Ballim, 1990) to maintain beliefs and other propositional attitudes for individuals and groups and to model the change and exchange of beliefs/attitudes within and amongst individuals and groups. This work had already been extended in ONR work. The aim was to populate this belief engine with semantic content extracted from dialogues and attributed to participants, so as to track their changing beliefs as dialogues progressed. The plan was to be the first system in which a belief engine was combined with real content extracted from dialogue corpora. The belief engine was re-implemented but data availability became a problem but we were able to shift the emphasis of the project to knowledge basis population. The other major component by UAlbany, was successful in competitions on sentiment extraction. The work was based on the assumption that social relations such as leader or influences can provide important social information about participant roles and relationships. Such social information is latent in dialogue and derivable via sociolinguistic features as locator of sentiment. The major research achievement of the project has been to show that belief and sentiment change in dialogue are in fact correlated and one can be used as a predictor of the other.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 2018
- Accession Number
- AD1051880
Entities
People
- Wilks Yorick
Organizations
- Florida Institute for Human and Machine Cognition