Machine Translation (MT) systems typically translate sentences independently of each other, however, certain textual elements cannot be correctly translated without a wider conversational context, which may appear outside the current sentence. ADAPT was challenged to create an MT system that can take context from previous sentences into consideration in the translation of application in conversational e-commerce, Voice-UI, for localisation managers.
Adapt developed a Neural translation system that was attentive to the wider conversational context for a seamless multi-language experience using inputs from current sentences and previous source sentences.
Our novel combination of contextual strategies greatly outperforms existing models. This strategy uses the previous sentence as an auxiliary input and decodes both the current and previous sentence. Previous source sentences are integrated as contexts when translating the current sentence.
Our system improves translation quality by 20% and speed by 15% compared to the baseline.
Chatbots – international customer support
Voice – UI