Automatic Validation of Terminology Translation Consistency with Statistical Method

This paper presents a novel method to automatically validate terminology consistency in localized materials. The goal of the paper is twofold. First, we explore a way to extract phrase pair translations for compound nouns from a bilingual corpus using word alignment data. To validate the quality of the extracted phrase pair translations, we use a Gaussian mixture model (GMM) classifier . Second, we quantify consistency of translation as a measurement of quality. With this approach, a quality assurance process for terminology translation can be fully automated. It can also be used for maintaining bilingual training data quality for machine translation.