Semantic Representation for Understanding Meaning Based on Correspondence Between Meanings

Abstract

One of the biggest problems in natural language processing is that its processing target (i.e. the surface expressions of sentences) has a great deal of diversity. In order to reduce the difficulty, it is desirable to extract the semantic content denoted by a sentence in such a way that it does not depend on the surface expressions as much as possible. This paper proposes a new semantic representation and general interpretive procedures that enable us to obtain the result of semantic interpretation from a variety of surface expressions of the input independently of their dependency structures. In the semantic representation to be proposed, a variety of surface dependency relations are compressed into attribute nouns, and the meaning expressed by dependency relation is represented in a uniform style (i.e. attribute = value). This approach enables us to establish correspondence between meanings by using the attribute-value pair as a basic unit. With this semantic representation and the general interpretive procedures, the same interpretive result can be obtained from sentences with different dependency structures. We will further demonstrate that semantic contents of multiple sentences can be integrated by interpreting them based on the correspondence between meanings.