The Correction Machine: A computer Model of Recognizing and Producing Belief Justifications in Argum

In discourse processing, the major problems are understanding the underlying connections between successive dialog utterances and producing coherent dialog responses. This dissertation presents a computer model that accomplishes these tasks for argumentative dialogs in which (1) each dialog response presents a belief, (2) each belief is presented to address some perceived disagreement, and (3) each belief is part of a justification for some other belief related to the dialog. In these dialogs, understanding a response involves relating a stated belief to beliefs appearing earlier in the dialog, and producing a response involves selecting a belief to justify and deciding upon the set of beliefs to provide as its justification. Our approach is knowledge based, using general, common-sense planning heuristics to recognize how a belief is being justified and to form new justifications for beliefs. This approach allows us to recognize and formulate novel belief justifications, a necessary capability for any system that participates cooperatively in dialogs involving disagreements. The model has been implemented in a computer program, The Correction Machine, that participates in dialogs in two disparate domains: arguing over the best way to perform everyday planning tasks such as keeping our Artificial Intelligence lab clean, and providing advice to novice UNIX users.