Knowledge-based parsing.

Abstract : A model for knowledge-based natural language analysis is described. The model is applied to parsing English into Conceptual Dependency representations. The model processes sentences from left to right, one word at a time, using linguistic and non-linguistic knowledge to find the meaning of the input. It operates in three modes: structure-driven, position-driven, and situation-driven. The first two modes are expectation-based. In structure driven mode concepts underlying new input are expected to fill slots in the previously built conceptual structures. Noun groups are handled in position-driven mode which uses position-based pooling of expectations. When the first two modes fail to account for a new input, the parser goes into the third, situation-driven mode which tries to handle a situation by applying a series of appropriate experts. Four general kinds of knowledge are identified as necessary for language understanding: lexical knowledge, world knowledge, linguistic knowledge, and contextual knowledge.