Towards Multimodal Simultaneous Neural Machine Translation
This work addresses the challenge of real-time translation for applications requiring timely multilingual understanding, though it is incremental as it builds on existing simultaneous translation methods by adding a modality.
The paper tackles the problem of insufficient input information in simultaneous translation by proposing multimodal simultaneous neural machine translation (MSNMT) that uses visual information, and it shows that MSNMT significantly outperforms text-only methods in low-latency scenarios on the Multi30k dataset.
Simultaneous translation involves translating a sentence before the speaker's utterance is completed in order to realize real-time understanding in multiple languages. This task is significantly more challenging than the general full sentence translation because of the shortage of input information during decoding. To alleviate this shortage, we propose multimodal simultaneous neural machine translation (MSNMT), which leverages visual information as an additional modality. Our experiments with the Multi30k dataset showed that MSNMT significantly outperforms its text-only counterpart in more timely translation situations with low latency. Furthermore, we verified the importance of visual information during decoding by performing an adversarial evaluation of MSNMT, where we studied how models behaved with incongruent input modality and analyzed the effect of different word order between source and target languages.