Towards Distributional Semantics-Based Classification of Collocations for Collocation Dictionaries

Abstract

Automatic acquisition of raw source material is of great aid for the compilation of dictionaries, and, in particular, of specialized dictionaries such as collocation dictionaries. The extraction of collocations from corpora has been actively worked on since the late eighties. The quality of the state-of-the-art extraction algorithms allows the lexicographers to obtain lists of collocations they can work with. However, mere lists of collocations are not sufficient. In collocation dictionaries, collocations are grouped semantically, which also presupposes a semantic classification of collocations. In this article, a distributional semantics-based model is proposed that classifies collocations with respect to broad semantic categories as encountered in dictionaries. In experiments with Spanish verb-noun and noun-adjective collocations from the lexicographic field of emotion nouns, it is shown that the use of features extracted from the context of collocations is decisive for retrieval of draft entries for collocation dictionaries.

Collocation ranking: frequency vs semantics

“…Additionaly, Wanner et al (2017) Examining related literature, we can conclude that regardless of the fact that word embeddings are a very popular source of semantic information and that their usage as input features for making predictions in NLP has been considered a standard approach for years now, they have not yet been tested in a supervised learning setting on the task of general collocation ranking.…”

Using bilingual word-embeddings for multilingual collocation extraction

“…A different perspective on collocation extraction focuses not only on their retrieval, but on semantically classifying the obtained collocations, in order to make them more useful for NLP applications (Wanner et al, 2006;Wanner et al, 2016).…”

Multilingual Extraction and Categorization of Lexical Collocations with Graph-aware Transformers

“…Most of the research deals only with collocation identification (Smadja, 1993;Lin, 1999;Pecina and Schlesinger, 2006;Bouma, 2009;Dinu et al, 2014;Levine et al, 2020). Some works deal with the categorization of manually precompiled lists of collocations, either in isolation (Wanner, 2004;Wanner et al, 2006;Espinosa Anke et al, 2019) or with their original sentence-level contextual information (Wanner et al, 2017). Only a few works in the early phase of the neural network era of NLP address the problem of collocation identification and semantic categorization as a joint task in monolingual settings (Rodríguez-Fernández et al, 2015;.…”