Parsing as search

Abstract

Developing laboratory assignments on natural language processing in an introductory AI course is difficult. One reason is that the interpreter that applies a grammar to input is often a complex, hard to understand system. One is forced either to treat the interpreter as a black box, restricting students to extending the grammar that it applies, or to spend an inordinate amount of time mastering the interpreter. We present an alternative that enables students to master an interpreter in fairly short order. Approaching parsing as a search problem, an interpreter is presented that is based on a search engine with which students have had previous experience. This paper discusses an interpreter for an RTN grammar. A subsequent paper shows how the interpreter can be modified to handle an ATN grammar. The grammar and interpreter are available via ftp.