Integration of Machine Translation Tools in Software Localization: mission (im)possible?

Abstract

The use of machine translation tools in various translation related activities has exponentially increased, not only because of the pandemic and post-pandemic context, the growth of the language industry, but also because of the current technological development of machine translation tools. Researchers, translators and developers actively discuss machine translation capabilities, its impact on translation and language, and challenges it brings. However, when it comes to software localization, possibilities of machine translation tools seem overestimated and limited, not only because of the quality of the output the machine translation tools produce, but also because of technical capacities to recognize programming language and handle difficult scenarios. This article aims to introduce possibilities and limitations of integrating machine translation tools in the process of software localization into the Lithuanian language. Here the term “integration” is used as the application of machine translation tools in the workflow of localization so as to speed up the process and help translators, but not as the technological integration when machine translation tools are integrated with computer aided translation (CAT) tools. The article presents an experiment during which several machine translation tools, such as Google Translate, DeepL, Vilnius University machine translation tool and Tildė machine translation tool, were tested with the Lithuanian language as the low-resourced language. The four machine translation tools were selected due to their popularity and current development in and for the Lithuanian language (Utka et al. 2020). The machine translation tools were given to translate .rc2 or .txt software-related resource files. The output quality produced in the Lithuanian language was compared in terms of text cohesion, term accuracy, identification of segments to be localized and damaged programming code. Moreover, the machine translation outputs were compared with the output of professional translation and localization CAT tools such as Passolo and Trados. The results showed that none of the machine translation tools used, despite the current integration of artificial intelligence solution, can produce high-quality translated text in the Lithuanian language due to the assumption that the Lithuanian language (with around 3 million speakers) is not commercially attractive. The output produced cannot be applied to speed up or ease localization in terms of the output text quality.