Identifying Events using Similarity and Context

Identifying Events using Similarity and Context Dominic R. Jones and Cynthia A. Thompson School of Computing University of Utah Salt Lake City, UT 84103 dominicj@cs.utah.edu, cindi@cs.utah.edu Abstract As part of our work on automatically build- ing knowledge structures from text, we apply machine learning to determine which clauses from multiple narratives describing similar sit- uations should be grouped together as descrip- tions of the same type of occurrence. Our approach to the problem uses textual similar- ity and context from other clauses. Besides training data, our system uses only a partial parser as outside knowledge. We present re- sults evaluating the cohesiveness of the aggre- gated clauses and a brief overview of how this work fits into our overall system. 1 Introduction Early work in natural language processing included am- bitious research on the representation and use of informa- tion about commonly experienced situations (Schank and Riesbeck, 1981). The concept of a script was introduced in this research, to explain how people understand these situations and make inferences about them. A script is a stereotypical sequence of events that occur as part of a larger situation and can be used to infer missing de- tails from a partial description of the larger occurrence, in essence providing a means for extracting information that is not actually present in a text. Research on scripts includes demonstrations of hand built scripts (Cullingford, 1978) and sketchy scripts (De- Jong, 1982) and the adjustment of hand-built scripts using a genetic algorithm (Mauldin, 1989). Work on learning schemata under constrained circumstances (Mooney and DeJong, 1985) pursues similar goals. Our research has indicated that scripts may not explic- itly occur in common types of text, such as newspaper stories or incident reports. Other research also appears to support this conclusion (Clark and Porter, 1995). There- fore, we are investigating event correlations as a more appropriate and extractable knowledge structure. In gen- eral, it appears that long event sequences do not reliably recur in our data. We instead look for reliable correlations between a small number of events. Our goal is to automatically extract correlated events from text, using only a partial parser as outside infor- mation. To support this goal, we need to group clauses from distinct texts into coherent events, handling several sources of variety in descriptions of the same type of oc- currence. Synonymy and abbreviations are two common contributors. A more important phenomenon is the ex- istence of semantic categories keyed to the events them- selves. A number of different objects may participate in an event, and yet have dissimilarities that place them in different conventional semantic categories. For example, a tree and a parked vehicle may both be collided with in different aircraft crashes, yet it is difficult to conceive of a reasonably specific semantic category that contains both. Each is a physical object, yet there are a large number of other physical objects that would not reasonably partici- pate in a crash in the same way (books, hamburgers, and moons are a …

Content truncated to 500 words