Context-Aware Neural Speech Translation in Multilingual Code-Switched Conversations
Abstract
As the world becomes more connected, with multiple languages being used in each conversation, multilingual communications play a much greater role in our daily lives. An example of this would be code-switching, where participants will move between languages on the fly throughout their conversations. Although significant advancements have occurred in the field of voice translation technologies recently, the majority of current systems are still based on processing at the sentence level and cannot address the dependencies of context that exist between the spoken turns within a conversation. This limitation often causes the translated output to be wrong, not make sense, or be semantically inconsistent, especially in situations where there are multiple languages present and where code-switching occurs. This work proposes to address these challenges with a new context-aware neural voice translation system that includes context awareness as part of its approach to translation