Adaptive Understanding: Correcting Erroneous Inferences.
Abstract
This thesis is about understanding potentially misleading stories. A reader cannot know ahead of time whether or not a story will turn out to contradict one of its own previous implications. Therefore, virtually every story is potentially misleading. Understanding a story requires the reader to be able to recognize when a story contradicts a previous inference, and to correct the erroneous inference by replacing it with a better inference. ARTHUR (A Reader THat Understands Reflectively) is a computer program that understands stories by inferring unstated connections among the statements in the text, and producing a representation of the story which includes these inferences. The inferences in this story representation must be continually updated in light of each new story statement read. ARTHUR can recognize and correct its own erroneous inferences during understanding, and hence it can understand stories which contain entirely novel information. ARTHUR demonstrates its understanding of a story by using its story representation to answer questions about the story. ARTHUR embodies a theory of adaptive understanding: it's own understanding processes are affected and altered by what it reads. ARTHUR's ability to understand depends on its knowledge of the situations that can appear in stories, and, reciprocally, its knowledge can be increased according to what ARTHUR reads. ARTHUR's operation is based on a theory of the organization of situational knowledge in an understander's memory, and a theory of the processes by which that knowledge is applied during story understanding.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1980
- Accession Number
- ADA081012
Entities
People
- Richard Horace Granger Jr
Organizations
- Yale University