POLIBITS, Vol. 60, pp. 43-56, 2019.
Abstract: Verb-noun collocations, lexical functions, hypernyms, context, supervised learning
Keywords: In this work we study semantic and contextual characteristics of four types of verb-noun collocations in Spanish. Each type corresponds to a different lexical function defined in the works of Žolkovskij and Mel’čuk [1, 2, 3] and further elaborated by Apresjan [4, 5]. First, we explain how the typology of lexical functions can be viewed as a consistent way to classify collocations according to their semantic and syntactic patterns. Then, using four lexical functions as well as free word combinations as classes of verb-noun pairs, we examine how they can be identified automatically by supervised learning methods. To build a semantic representation of verb-noun pairs, we used WordNet hypernyms of the verb and the noun. To study contextual properties of the classes, we experimented on a corpus of news. The highest F1-score achieved in the experiments was 0.81 for CausFunc 1 using hypernyms. We found that contextual characteristics were not powerful enough to discriminate among subtle semantic differences of lexical functions: the best F1-score of 0.62 for Real 1 was achieved by GaussianProcessClassifier using raw frequency of context words after removing stopwords from the corpus. Discussing our results, we looked for features which could account for higher or lower results.
PDF: Dictionary and Corpus-Based Study of Lexical Functions in Spanish
PDF: Dictionary and Corpus-Based Study of Lexical Functions in Spanish
https://doi.org/10.17562/PB-60-6
Table of contents of POLIBITS 60