COMPUTABLE SEMANTIC DERIVATIONS (AND THEIR APPLICATION TO NATURAL LANGUAGE TEXTS).
INSTITUTE FOR FORMAL STUDIES LOS ANGELES CALIF
Pagination or Media Count:
A system of semantic analysis is described which derives structured representations for whole paragraphs of natural language text from word codings and semantic rules that can be put in phrase-structure form. From this representation an ambiguity resolution for the words of the paragraph can be read off, since a derived structure contains only a single selection from the coded senses of the text words. Thus the system is first and foremost an ambiguity resolver however, the structured representation itself remains available for potential further use in, for example, an MT program. The system is implemented in LTSP 1.5 and applied to ten paragraph length texts. The operation of the system rests on assumptions 1 of the radical sense-ambiguity of the words of natural language texts and 2 that meaningful texts are coherent because of a certain amount of conceptual repetition they contain, this repetition constituting a discourse structure sufficient to resolve the sense-ambiguity of the text. The semantic rules referred to explicate this conceptual repetition, or closeness, in terms of semantic compatibility between successive fragments of text. In certain cases the system is able to construct a new sense-entry for a word so that the resolution procedure for a paragraph can be completed. Several connections are defined between the conventional logical structure of the arguments contained by the paragraphs and the semantic structure derived by the present system. Author