Experiments with Term Translation

Proceedings of COLING 2012: Technical Papers, pages 67–82, COLING 2012, Mumbai, December 2012. Experiments with Term Translation Mihael ARCAN 1 Christian F EDERMANN 2 Paul BUIT ELAAR1 (1) Unit for Natural Language Processing, Digital Enterprise Research Institute National University of Ireland Galway Galway, Ireland (2) Language Technology Lab, German Research Center for AI Saarbrücken, Germany mihael.arcan@deri.org, paul.buitelaer@deri.org, cfedermann@dfki.de Abstract In this article we investigate the translation of financial terms from English into German in the isolation of an ontology vocabulary. For this study we automatically built new domain-specific resources from the translation search engine Linguee and from the online encyclopaedia Wikipedia. Due to the fact that we performed the translation approach on a monolingual ontology, we ran several sub-experiments to find the most appropriate model to translate the financial vocabulary. The findings from these experiments lead to the conclusion that a hybrid translation system, a combination of bilingual terminological resources and statistical machine translation, can help to improve translation of domain-specific terms. Finally we undertook a manual cross-lingual evaluation on the monolingual ontology to get a better understanding on this specific short text translation task. Keywords: Ontologies and terminology, Empirical machine translation. 67 1 Introduction Our research on the translation of ontology vocabularies is motivated by the challenge of translating domain-specific terms with restricted or no additional textual context that in other cases may be used to improve the translation. For our experiment we started by translating financial terms with the baseline systems trained on the JRC-Acquis (Steinberger et al., 2006) corpus and the European Central Bank Corpus (Tiedemann, 2009). Although both resources contain a large amount of parallel data, the translations were not satisfactory. To improve the translations of the financial ontology vocabulary we built a new parallel resource, which was generated using Linguee, an online translation query service. With this data, we could train a small model, which produced better translations than the baseline model using only general resources. Since the manual development of terminological resources is a time intensive and expensive task, we used Wikipedia as a background knowledge base and examined the articles tagged with domain- specific categories. With this extracted domain-specific data we built a specialised English-German lexicon to store translations of domain-specific terms. These terms were then used in a pre-processing method in the decoding approach. This approach incorporates the work by (Aggarwal et al., 2011), where the authors use the ontology structure to calculate the similarity between the labels. They combine the semantic, terminological and linguistic information for monolingual ontology matching, which can be extended to the multilingual scenario. We split the financial terms into n-grams and queried for financial sub-terms in Wikipedia, which we used to query Wikipedia. The remainder of the paper is organised as follows: In Section 2 we give an overview on the related work. In Section 3 we describe the ontology and the existing parallel resources, which were used for generating the translation and language model. Section 4 presents the new resources which were used for improving the …

Content truncated to 500 words