Expanding wordnets to new languages with multilingual sense disambiguation

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 97–108, Osaka, Japan, December 11-17 2016. Expanding wordnets to new languages with multilingual sense disambiguation Mihael Arcan John P. McCrae Paul Buitelaar Insight Centre for Data Analytics, National University of Ireland, Galway firstname.lastname@insight-centre.org Abstract Princeton WordNet is one of the most important resources for natural language processing, but is only available for English. While it has been translated using the expand approach to many other languages, this is an expensive manual process. Therefore it would be beneficial to have a high-quality automatic translation approach that would support NLP techniques, which rely on WordNet in new languages. The translation of wordnets is fundamentally complex because of the need to translate all senses of a word including low frequency senses, which is very challenging for current machine translation approaches. For this reason we leverage existing translations of WordNet in other languages to identify contextual information for wordnet senses from a large set of generic parallel corpora. We evaluate our approach using 10 translated wordnets for European languages. Our experiment shows a significant improvement over translation without any contextual information. Furthermore, we evaluate how the choice of pivot languages affects performance of multilingual word sense disambiguation. 1 Introduction Princeton WordNet (Fellbaum, 1998) is a manually created resource that has been used in many differ- ent tasks and applications across linguistics and natural language processing. WordNet’s hierarchical structure makes it a useful tool for many semantic applications and it also plays a vital role in modern deep learning based NLP systems (Rychalska et al., 2016). However, Princeton WordNet is only avail- able for English and huge efforts have been made to extend WordNet with multilingual information in projects, such as EuroWordNet (Vossen, 1998), BalkaNet (Tufi¸s et al., 2004) and MultiWordNet (Pianta et al., 2002). However, most of the wordnet resources resulting from these efforts have fewer synsets than the Princeton WordNet and there are still many languages for which a wordnet does not exist or is not available to all potential users due to licensing restrictions, impacting applications in information re- trieval, word sense disambiguation, sentiment analysis or knowledge management that rely on Princeton WordNet. Most wordnets in languages other than English have followed an extend approach (Vossen, 2005), where the structure of Princeton WordNet is preserved and only the words in each synset are translated and new synsets are added for concepts, which are not lexicalized in English. Since manual multilingual translation and evaluation of wordnets using this approach is a very time consuming and expensive pro- cess, we apply statistical machine translation (SMT) to automatically translate WordNet entries. While an SMT system can only return the most frequent translation when given a term by itself, it has been observed that SMT provides strong word sense disambiguation when the word is given in the context of a sentence. As a motivating example, we consider the word vessel, which is a member of three synsets in Princeton WordNet, whereby the …

Content truncated to 500 words