Natural Language Sourcebook.

This sourcebook is intended to provide researchers and users of natural language computer systems with a classification scheme to describe language-related problems associated with such systems. Methods from the disciplines of artificial intelligence (AI), education, linguistics, psychology, anthropology, and psychometrics were applied in an effort to characterize better the scope, quality, and effectiveness of AI systems. The sourcebook provides examples of representative types of problems in natural language understanding drawn from literature in AI, computational linguistics, and cognitive science. Each example consists of: (1) one or more sentences, a fragment of dialogue, or piece of text that illustrates a conceptual issue; (2) a reference; and (3) a discussion of the problem a system might have in understanding the example. Examples present many types of processing problems. Two alternative classification schemes, one designed by a computational linguist and the other by a cognitive psychologist, are also offered as a basis for cross-referencing and exploration of alternate relations among examples given. Examples are in four categories: single-utterance; connected-utterance; true-dialogue; and ill-formed input. A glossary, a 100-item list of references, and an author index are appended. (MSE)