Context-Aware Machine Translation for Software Localization

Context-Aware Machine Translation for Software Localization Victor Munt´es-Mulero Patricia Paladini Adell CA Technologies WTC Almeda Park 08940 Cornell`a de Llobregat, Barcelona Victor.Muntes@ca.com Patricia.PaladiniAdell@ca.com Cristina Espa˜na-Bonet Llu´ıs M`arquez TALP Research Center Universitat Polit`ecnica de Catalunya cristinae@lsi.upc.edu, lluism@lsi.upc.edu Abstract Software localization requires translating short text strings appearing in user inter- faces (UI) into several languages. These strings are usually unrelated to the other strings in the UI. Due to the lack of se- mantic context, many ambiguity problems cannot be solved during translation. How- ever, UI are composed of several visual components to which text strings are as- sociated. Although this association might be very valuable for word disambiguation, it has not been exploited. In this paper, we present the problem of lack of con- text awareness for UI localization, provid- ing real examples and identifying the main research challenges. 1 Introduction Due to the rapid and worldwide development of Internet and IT applications, fast software local- ization is becoming essential, requiring user inter- faces (UI) to be translated to different languages. One of the main obstacles when translating UI is the word sense disambiguation problem since strings are usually independent from other strings in the UI and, therefore, it is not possible to infer semantic information from other parts of the text. In this paper, we want to show that the meaning of a string in this environment varies depending on its position in the UI. For instance, a word associ- ated to a menu may be interpreted as a name, but it may be an action if the same word appears on a button. Although enriching the translation pro- cess with this alternative contextual information would benefit quality, previous software localiza- tion techniques ignore, in general, this approach. c⃝ 2012 European Association for Machine Translation. It is commonly accepted that the number of words processed per day by a human translator is significantly increased when an efficient machine translation (MT) engine is used and human trans- lators intervene in the post-editing phase. Specifi- cally, it is becoming popular to use MT engines for software localization. Unfortunately, even if con- textual information about the UI components asso- ciated to strings was gathered, current localization procedures using MT engines are not devised to absorb and exploit it to improve MT quality. Vi- sually aided translation tools, like Passolo1 or Cat- alyst2, leverage contextual information and show it graphically to human translators. However, they depend on specific file formats which are not al- ways available. Improving the quality of the output of MT allows both (i) to reduce the cost of transla- tion by increasing translator’s throughput by up to 50%, based on CA Technologies3 experience, and (ii) to reduce the delay to market of the software products. The objective is simultaneous shipment. The main contributions of this paper are as fol- lows. Section 2 describes the most relevant state of the art. In Section 3, we define the problem of the lack of contextual information in UI localization providing examples extracted from …

Content truncated to 500 words