Using natural language prompts for machine translation
This addresses the challenge of flexible and adaptable machine translation for users needing specific output properties or translation into rare languages, though it appears incremental as it builds on existing prompt-based methods.
The paper tackles the problem of controlling machine translation outputs, such as formality or dialect, using natural language prompts, and demonstrates that this enables positive transfer for unseen language pairs, allowing translation into languages not seen during fine-tuning by using their English names.
We explore the use of natural language prompts for controlling various aspects of the outputs generated by machine translation models. We demonstrate that natural language prompts allow us to influence properties like formality or specific dialect of the output. We show that using language names to control the output language of multilingual translation models enables positive transfer for unseen language pairs. This unlocks the ability to translate into languages not seen during fine-tuning by using their English names. We investigate how scale, number of pre-training steps, number of languages in fine-tuning, and language similarity affect this phenomenon.