Complex events in an ontological -semantic natural language processing system
Abstract
The goal of this dissertation is to elucidate principles for representing complex-event knowledge (or “scripts”) for use in an ontological-semantic natural language processing system, specifically the Mikrokosmos system. Complex events are, simply, events that comprise other events. A shorthand example is: the event of buying may comprise picking out merchandise, bringing it to the cash register, offering money to the cashier, receiving change, and leaving the store with the merchandise. As previous research has shown, texts make widespread use of such “world knowledge” by leaving many events implicit, relying on the hearer/reader to infer this information within the discourse context. The challenge for natural language processing programs is to construct a model of this world knowledge to fill in these gaps. Previous programs armed with such knowledge, e.g., SAM (Cullingford 1978) and Ms. Malaprop (Charniak 1977), have made some advances, but several design problems prevented real progress. It is argued in this dissertation that the ontological-semantic paradigm generates semantic descriptions of texts rich enough to make use of complex-event knowledge, thereby eliminating one barrier to implementation of such knowledge. Furthermore, this dissertation develops specific modifications to the ontological-semantic system that enable representations of complex-event knowledge to: (i) represent, in principle, any sequence of “real-world” events, (ii) achieve broad conceptual coverage, (iii) discern “fine-grained” differences in conceptual information, and (iv) significantly reduce redundancy in knowledge representation. Several constructed complex-event descriptions are adduced as evidence of feasibility of the formalism. Though this knowledge has not yet been implemented, the suggestions made in this dissertation should prove to be a significant step in representing the immense body of complex-event knowledge for natural language processing systems.
Degree
Ph.D.
Advisors
Raskin, Purdue University.
Subject Area
Linguistics|Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.