Generate or Match? Towards Collaborative Interface for Translators in the Generative Era

To empower translators in a machine translation post-editing environment, we recognize the urgent need to focus on translator-centered interaction that facilitates translators to augment the translated output and, more importantly, the translation process. Recent advances in large language models (LLMs) have revolutionized machine translation (MT) and opened up new possibilities. When LLM performance no longer stands as a pause sign for its integration into the translation ecosystem, it is paramount to ensure that the translator-computer interface unleashes the full potential of the human-machine pair. We argue that one such overlooked aspect might be the effect of the generative paradigm brought about by LLMs on the translator-computer interaction. In this paper, we propose a taxonomy to characterize the translator-computer interaction, as well as to highlight the necessary changes when designing an LLM-based translation interface. By adopting the generative interface design, we argue that translators can regain supervisory control of both the translation process and the translated output, contributing to a more ideal and collaborative translator-computer interaction.