The goal of the CUBISM project is to develop and implement an algorithm for the deep and robust analysis and understanding of multi-party conversations — including conversations that are only partially and inaccurately transcribed, and likely to contain obscure and even deceptive language. We will accomplish this by developing a novel algorithm to systematically expose pragmatic knowledge in the conversation that is otherwise contextually bound and only implicitly expressed. In particular, the proposed algorithm will make explicit the conversation’s dynamics (e.g., critical topic shifts and other sociolinguistic features of dialogue) through which the interlocutors’ beliefs and intentions as well as those of third parties can be modeled and projected. The algorithm is applicable to both English and foreign language conversations (e.g., Chinese, Farsi), including multi-party (more than two participants) conversations and one-sided conversations — two-party conversations where one party’s utterances are unavailable.
For more information, please contact:
Prof. Tomek Strzalkowski, Principal Investigator
University at Albany, SUNY
ILS Room 262B, Social Science Building
Albany, NY 12222.
Email: tomek [at] albany [dot] edu
Phone: 518-442-3082; Fax: 518-442-2606
The purpose of our algorithm for deep analysis is to enable the larger DEFT system to discover, summarize, and alert regarding meaningful dialogue content that is responsive to such queries as:
· What is the state of belief of dialogue participants?
· What is the nature of the influence and leadership exerted in the dialogue?
· How quickly and in what directions are belief states evolving?
· What topics are discussed and what are the attitudes of participants towards those topics?
· Are there anomalies indicative of intentional deception?
The project is supported by the Defense Advanced Research Projects Agency (DARPA) as part of the DEFT Program.