A Language Detection System for Short Chats in Mobile Games

Machine Translation system accuracies are often brought down due to inaccurate Language Detection (LD) of input phrases. The Language detection accuracy is further affected when the inputs are short and contain ungrammatical phrases, especially in a multilingual mobile game setting. Chat messages in mobile games are often short as they are typed on mobile devices and contain slang as a common communication preference. Previous work has shown that LD systems have a drop in accuracy when the inputs are short messages instead of long ones. This paper targets LD for short chat messages in mobile games. We propose a novel LD system which integrates text-based and user-based methods to achieve significantly better performance over current state-of-the-art LD systems.