Citius at SemEval-2017 Task 2: Cross-Lingual Similarity from Comparable Corpora and Dependency-Based Contexts

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 226–229, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics Citius at SemEval-2017 Task 2: Cross-Lingual Similarity from Comparable Corpora and Dependency-Based Contexts Pablo Gamallo Centro Singular de Investigaci´on en Tecnolox´ıas da Informaci´on (CiTIUS) Universidade de Santiago de Compostela, Galiza pablo.gamallo@usc.es Abstract This article describes the distributional strategy submitted by the Citius team to the SemEval 2017 Task 2. Even though the team participated in two sub- tasks, namely monolingual and cross- lingual word similarity, the article is mainly focused on the cross-lingual sub- task. Our method uses comparable cor- pora and syntactic dependencies to extract count-based and transparent bilingual dis- tributional contexts. The evaluation of the results show that our method is com- petitive with other cross-lingual strategies, even those using aligned and parallel texts. 1 Introduction A comparable corpus consists of documents in two or more languages or varieties which are not trans- lation of each other and deal with similar top- ics. Comparable corpora are by definition mul- tilingual and cross-lingual text collections. The use of comparable corpora for word similarity is a well-known task (Fung and McKeown, 1997; Rapp, 1999; Saralegi et al., 2008; Gamallo, 2007; Gamallo and Pichel, 2008; Ansari et al., 2014; Hazem and Morin, 2014). The main advantage of comparable corpora is that the Web can be used as a huge resource of multilingual texts. In con- trast, their main drawback is the low performance of the extraction systems based on them. Accord- ing to (Nakagawa, 2001), word similarity extrac- tion from comparable corpora is a too difficult and ambitious objective, and much more complex than extraction from parallel and aligned corpora. However, the reasonable results our comparable- corpus method achieved in the cross-lingual sub- task of SemEval 2017 Task 2 (Camacho-Collados et al., 2017) show that the gap between paral- lel and comparable corpora for word similarity is shortening. In this article, we describe our comparable-corpus method for cross-lingual sim- ilarity in the next section (2).Then Section 3 de- scribes the experiments and the evaluation and, fi- nally, a discusion is addressed in Section 4. 2 The Cross-Lingual Strategy The best known strategy to extract bilingual cor- respondences from comparable corpora works as follows: a word w2 in the target language is se- mantically related to w1 in the source language if the context expressions with which w2 co-occurs tend to be translations of the context expressions with which w1 co-occurs. The basis of the method is to find the target words that have the most sim- ilar distributions with a given source word. The starting point of this strategy is a seed list of bilin- gual expressions that are used to build the context vectors defining all words in both languages. This seed list is usually provided by an external bilin- gual dictionary. In our approach, the seed expres- sions used as cross-language pivot contexts are not bilingual pairs of words as in related work, but bilingual …

Content truncated to 500 words