Exploring the Context of Lexical Functions

摘要

We explore the context of verb-noun collocations using a corpus of the Excelsior newspaper issues in Spanish. Our purpose is to understand to what extent the context is able to distinguish the semantics of collocations represented by lexical functions of the Meaning-Text Theory. For experiments, four lexical functions were chosen: Operl, Reall, CausFuncO, and CausFuncl. We inspected different parts of the eight-word window context: the left context, the right context, and both the left and right context. These contexts were retrieved from the original corpus as well as from the same corpus after stopwords deletion. For the vector representation of the context, word counts and tf-idf of words were used. To estimate the ability of the context to predict lexical functions, we used various machine-learning techniques. The best F-measure of 0.65 was achieved for predicting Reall by Gaussian Naieve Bayes using the left context without stopwords and word counts as features in vectors.