In this work, we prioritized human users by taking a comprehensive approach to minimize risks transferred to them. We evaluated and implemented responsible AI techniques to ensure the quality of translations while considering ethical implications.
Low-resource languages were chosen for this study, which could improve world readiness and information access for many in these communities. However, it may also make groups with lower levels of digital literacy more vulnerable to misinformation or online scams. Bad actors might misappropriate our work for nefarious activities, which we consider as unintended use.
To address this, we optimized for translation quality while acknowledging that toxic, biased, or false outputs produced by the model could remain. These could have an adverse impact on those who rely on these translations to make important decisions, particularly when related to health and safety.
In summary, our work aims to improve translation quality while ethically considering the potential risks involved. We prioritize human users’ safety and security by addressing potential misuse of our work, ensuring responsible AI practices, and minimizing adverse impacts on vulnerable groups.
Computation and Language, Computer Science