An Ontology-Based Translation Memory Model in Localization Translation

Translation constitutes an important part of the process of product or software localization. Unlike translators in the traditional sense, localization translators translate with the aid of the translation memory (TM) tools, which is the core of computer-aided translation technology. Semantic information at word and sentence levels is beneficial to efficient retrieval and higher translation quality. This paper analyzes the properties of localization translation and presents an ontology-based translation memory model that is specific to localization translation. This model owns two types of ontologies, i.e., the universal ontology and the specialized ontology. The former gives the semantic explanation of universal concepts and is based on HowNet and WordNet, the latter aims to define the semantics of concepts and terminology belonging to the localization domain and some specialized areas such as the domain of railways. A systematic workflow for constructing the specialized ontology is also illustrated. The semantic similarity metrics at the word and sentence levels are designed to measure the degree of similarity between the sentence to be translated and its candidate translated sentences. These metrics support retrieval from memory bases of similar sentences translated previously. It is revealed that applying ontologies in localization translation can improve the efficiency of localization translation, guarantee the translation quality and speed up the process of software localization.