Dissertation Defense WHAT: Peter Gorniak: "The Affordance-Based Concept" WHEN: Friday, July 22, 2005, 10:00 AM EST WHERE: Bartos Theatre, MIT Media Lab (E15) DISSERTATION COMMITTEE: Deb K. Roy Associate Professor of Media Arts and Sciences Massachusetts Institute of Technology Daniel C. Dennett Professor of Philosophy Tufts University Allen L. Gorin Director, Human Language Technology Research National Security Agency U.S. Department of Defense Leslie P. Kaelbling Professor of Computer Science and Electrical Engineering Massachusetts Institute of Technology ABSTRACT: Natural language use relies on situational context. The meaning of words and utterances depends on the physical environment and the goals and plans of communication partners. These facts should be central to theories of language and automatic language understanding systems. Instead, they are often ignored, leading to partial theories and systems that cannot fully interpret linguistic meaning. Gorniak introduces a new computational theory of conceptual structure that has as its core claim that concepts are neither internal nor external to the language user, but instead span the objective-subjective boundary. This theory proposes interaction and prediction as a central theme, rather than solely emphasizing deducing, sensing, or acting. To capture the possible interactions between subject and object, the theory relies on the notion of perceived affordances: structured units of interaction that can be used for prediction at certain levels of abstraction. By using perceived affordances as a basis for language understanding, the theory accounts for many aspects of the situated nature of human language use. It provides a unified solution to a number of other demands on a theory of language understanding including conceptual combination, prototypicality effects, and the generative nature of lexical items. To support the theory, Gorniak describes an implementation that relies on probabilistic hierarchical plan recognition to predict possible interactions. The elements of a recognized plan provide an instance of perceived affordances which are used by a linguistic parser to ground the meaning of words and grammatical constituents. Evaluations performed in a multiuser role-playing game environment show that this implementation captures the meaning of free-form spontaneous directive speech acts that cannot be understood without taking into account the intentional and physical situation of speaker and listener. MIT Media Laboratory Home Page | Events Main Index