Towards automated Android app internationalisation: An exploratory study

Abstract

Android has become the most popular mobile platform with over 2.5 billion active users who use many different languages across many different countries. In order for Android apps to be useable by all of them, app developers usually need to add an internationalisation feature that adapts the app to the users’ linguistic and cultural requirements. Such a process, including the translation from the default language to up to thousands of languages, is usually achieved via manual efforts and hence is resource-intensive, time-consuming, and error-prone. Automated approaches are hence in demand to help developers mitigate such manual efforts. Since there are millions of apps proposed already for Android users, we are interested in knowing to what extent internationalisation has been supported. Our experimental results show that Android apps, at least the ones released on online markets, have mostly been equipped with internationalisation features, with the number of supported languages varies significantly. By mapping the actual term translations among different languages, we further find that the translations tend to be consistent among different apps, suggesting the possibility to learn from this data to achieve automated app internalisation. To explore this idea we implemented a Transformer-based prototype approach Androi18n, that learns from developers’ practical translations to achieve automated mobile app text translations. Experimental results show that Androi18n is effective in achieving our objective, and its high performance is generic across the translations of different languages.