Domain adaptation for ontology localization

• Detailed description of an architecture and methodology for machine translation of ontologies.
• Methodology for extracting domain terminology from several resources.
• Statistical methods for the domain adaptation of machine translation systems according to ontologies.
• Detailed evaluation showing improvement in translation quality for a number of ontologies.

Ontology localization is the task of adapting an ontology to a different cultural context, and has been identified as an important task in the context of the Multilingual Semantic Web vision. The key task in ontology localization is translating the lexical layer of an ontology, i.e., its labels, into some foreign language. For this task, we hypothesize that the translation quality can be improved by adapting a machine translation system to the domain of the ontology. To this end, we build on the success of existing statistical machine translation (SMT) approaches, and investigate the impact of different domain adaptation techniques on the task. In particular, we investigate three techniques: (i) enriching a phrase table by domain-specific translation candidates acquired from existing Web resources, (ii) relying on Explicit Semantic Analysis as an additional technique for scoring a certain translation of a given source phrase, as well as (iii) adaptation of the language model by means of weighting nn-grams with scores obtained from topic modelling. We present in detail the impact of each of these three techniques on the task of translating ontology labels. We show that these techniques have a generally positive effect on the quality of translation of the ontology and that, in combination, they provide a significant improvement in quality.