The True, Deep Happiness: Towards the Automatic Semantic Classification of Adjective-Noun Collocations

Abstract

[Abstract] Most of the current techniques for the extraction of collocations from corpora either provide plain lists of collocations or focus on the retrieval of a few specific types of verb-noun collocations - support verb constructions being the most prominent of them. Only little work has been done so far on the extraction and semantic classification of adjective-noun collocations. In this paper, we concentrate on the classification of adjective-noun collocations according to an adjective-noun fragment of the fine-grained semantically oriented collocation typology - the lexical function typology as introduced in the Explanatory Combinatorial Lexicology. For the classification, we use different Machine Learning (ML) techniques that draw upon the semantic description of the adjective-noun bigrams. The descriptions are obtained from an external lexicosemantic resource. So far, experiments have been carried out mainly on Spanish. In experiments discussed in the paper, two different ML-techniques have been applied to bigrams from the semantic field of emotions: the Nearest Neighbor Classification and a variant of the Bayesian Network Classification - the Tree Augmented Network Classification. As the external lexico-semantic resource, the Spanish part of the EuroWordNet has been used.