"Abstract" Understanding: The Relation between Language and Memory

Lebowitz, Michael

Natural language is primarily a tool of communication. This implies that whatever roles syntax, semantics, pragmatics, and world knowledge play in language understanding, the comprehension process must be driven by the need to understand the text or conversation, where understand means sufficiently relating the new information being conveyed to existing memories in order to remember the information and/or respond. The need for memory-driven text processing becomes especially clear during designed to read large the construction of a computer system numbers of texts and add them to a coherent memory. However, the recognition that information such as syntax and semantics play only a subsidiary role in text processing does not make understanding more difficult; rather, it makes it possible. A current natural language processing project underway at Columbia involves the creation of a computer program, known as RESEARCHER, that will read large numbers of technical abstracts, such as patent abstracts, and builds up a coherent memory based on the information obtained. This memory is then used in turn to help in the understanding process. RESEARCHER will use some of the same understanding principles as did IPP, a program that reads and remembers news stories [Lebowitz 80, Lebowitz 81]. One of the goals of RESEARCHER is to show that memory-based understanding techniques are as applicable to physical descriptions as to descriptions of events. In fact, due to the knowledge-intensive nature of technical descriptions, it is expected that the application of memory will be even more important in driving processing.


More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-012-81
Published Here
October 18, 2011