Coherence and Interpretation in English Texts

Abstract

A theory and system are described for the semantic analysis of complex, coherent English texts. The principal question addressed in this paper is how the meanings of the smaller elements of language compose into the meanings of larger stretches of text. Within sentence boundaries, this is achieved by an operation, called predicate interpretation, which provides a mechanism for general words, especially those having a spatial flavor, to acquire specific interpretations in context. Eeyond sentence boundaries it is achieved by an operation which matches successive sentences against a small number of common patterns and builds up a tree-like structure representing the text's patterns of coherence. In isolation, words and sentences do not have specific meanings so much as they have the potential for acquiring a variety of specific meanings in particular contexts. In this paper we ask how the meanings of the smaller elements of language compose into the meanings of larger stretches of text. The answers proposed come out of work on an inferencing system being developed for the Semantic Analysis of complex, coherent Texts in English (called SATE). This system is intended to be general and is being applied to sets of directions (Hobbs 1975), algorithm descriptions (Hobbs 1977), and complex expository texts (Hobbs 1976b). Section 1 of this paper briefly describes the system and the collection of world knowledge axioms it rests on. Section 2 considers the question within sentence boundaries, addressing the problems of how words should be defined, how context influences the meanings of words, and how metaphors are to be interpreted. Sections 3 and 4 discuss how the relations between sentences can be discovered and hence how the structure of paragraph-length texts can be built up.