Towards producing bilingual lexica from monolingual corpora

Towards Producing Bilingual Lexica from Monolingual Corpora Jingyi Han, Núria Bel Universitat Pompeu Fabra Roc Boronat, 138, 08018, Barcelona jingy.han@upf.edu, nuria.bel@upf.edu Abstract Bilingual lexica are the basis for many cross-lingual natural language processing tasks. Recent works have shown success in learning bilingual dictionary by taking advantages of comparable corpora and a diverse set of signals derived from monolingual corpora. In the present work, we describe an approach to automatically learn bilingual lexica by training a supervised classifier using word embedding-based vectors of only a few hundred translation equivalent word pairs. The word embedding representations of translation pairs were obtained from source and target monolingual corpora, which are not necessarily related. Our classifier is able to predict whether a new word pair is under a translation relation or not. We tested it on two quite distinct language pairs Chinese-Spanish and English-Spanish. The classifiers achieved more than 0.90 precision and recall for both language pairs in different evaluation scenarios. These results show a high potential for this method to be used in bilingual lexica production for language pairs with reduced amount of parallel or comparable corpora, in particular for phrase table expansion in Statistical Machine Translation systems. Keywords: automatic bilingual lexicon production, lexical resources, bilingual dictionaries. 1. Introduction Bilingual lexica are the key resource for many cross-lingual processing tasks and they are crucial components of machine translation systems. A large amount of research has focussed in the automatic production of bilingual dictionaries, mostly based on the large parallel or comparable corpora. However, such large bilingual related corpora are not readily available for many language pairs. Considering this data shortage problem, our objective is learning bilingual lexicons out of large monolingual, and not necessarily comparable, corpora in source and target languages. The first work in this area by Rapp (1995) was based on the hypothesis that translation equivalents in two languages have similar distributional profiles or co-occurrence patterns. Recently, Mikolov et al. (2013a) showed that word embeddings (distributed, dense and real-valued vector representations of words) indeed project word semantics into a vector space from their distributional characteristics. More interestingly, it is claimed that the relationship between vector spaces that represent different language word semantics can be captured by a linear transformation. Therefore, we propose to use word embedding vectors of translation word pairs to train a supervised classifier which can predict whether a new pair of words in two different languages can be under a translation relation, according to the regularities learned from a small training set. A previous attempt to use supervised classification for inducing bilingual lexica from non-parallel nor comparable corpora is Irvine and Callison-Burch (2013), although their approach is based on using non-distributional information (signals like temporal similarity, topical information, orthographical similarity, among others) to train the model. Moreover, the use of supervised methods has proved to be effective in statistical machine translation (SMT) too, a closely related task (Och & Ney, 2002). Our approach benefits from a supervised method and from the dense and small vectors produced by word2vec …

Content truncated to 500 words