Dynamic Interaction of Syntactic and Semantic Analyses Based on the Equivalent Transformation Computation Model

Abstract

In much conventional natural language processing, syntactic and semantic analyses are executed sequentially, which is inefficient due to computation sequence inflexibility. In this paper, we present system design based on the equivalent transformation computation model to construct natural language understanding that executes integrated syntactic and semantic analyses and generates all correct answers with adequate efficiency. The advantages of using the equivalent transformation include speed, provable correctness, and rule-based intuitive design. Equivalent transformation changes declarative descriptions, each of which represents constraints, one step at a time. Each step is proven to preserve the semantic meaning of a declarative description. We use numerous constraints for formalization and processing sequence is decided dynamically at runtime based on context. We demonstrate flexible processing that reduces the amount of computation by explaining the analysis of a sample Japanese sentence describing a chess move on a chessboard.