Network analysis techniques applied to dictionaries for identifying semantics in lexical spanish collocations

Abstract

The definitions in dictionaries are a source of information to support the results obtained by the automatic extraction of collocations from a text corpus. Measures of association, which are generally used in this task, are useful tools to extract candidate combinations. However, they do not offer information about other features of the collocations. They do not distinguish whether a combination is categorized as a collocation because of its frequency properties or because of its structural properties. Moreover, they cannot distinguish between lexical collocations and functional collocations with delexicalized elements. In this paper, we use a graph database for representing collocations and relations between words retrieved from dictionaries. We consider relations between lemmas and definiens in dictionary entries as well as relations between two words used to define the same sense of another one. This allows us to use a clustering algorithm and measures of centrality and influence in networks to identify semantic characteristics of combinations. The aim is to enrich the information on the combinatorial restrictions of words based on frequencies obtained by means of corpus linguistic techniques.