Acquisition of semantic patterns from a natural corpus of texts

Abstract

In this paper we present a methodology and a program, PETRARCA, for the extensive acquisition of a case based semantic dictionary. The major limitation of existing NLP systems is a poor encoding of semantic knowledge; on the other side, it is unrealistic to assume a manual codification of word senses, including idiomatic expressions and metaphors.The system presented in this paper analyzes a large sample of sentences including a given word, and produces for that word one or more entries in the semantic dictionary (one entry for each word sense). Sentences are provided by an on-line corpus of press agency releases on finance and economics.In PETRARCA, a target word sense definition is represented by a detailed list of use-types, called surface semantic patterns (SSPs). SSPs mirror the way humans most naturally explain a new word sense; in fact, people tend to give associations related to words rather than to provide conceptual categories.To derive these associations, the system uses a high-coverage morphosyntactic analyzer, a catalogue of phrasal-patterns/semantic-interpretation pairs, and a set of selectional restriction rules on semantic interpretation types.This paper claims that surface semantics is a reasonable descriptive framework to build a working computer program for language processing. It is general, and makes it possible to establish in a systematic way the rules of semantic encoding. We believe this being a useful contribution towards a more complete system of language learning.