Retrieving abstract information from multi-party conversations

This paper presents the use of dialog acts and addressee information in extracting domain-specific abstract information from multi-party conversations. Identification of dialog acts and addressee information is done in a domain-independent manner. This information is then used with a domain-specific rule set for identifying abstract information in chat conversations. To retrieve abstract information from a conversation in a new domain, the only requirement is to define a new rule set. This eliminates the need of domain-specific corpora. This paper also presents an improved version of a commonly used addressee recognition algorithm.