Translation Canvas: An Explainable Interface to Pinpoint and Analyze Translation Systems
Published in Empirical Methods of Natural Langauge Processing, 2024
Translation Canvas is an advanced explainable interface designed for machine translation researchers to deeply analyze and compare the performance of various translation models. It serves as a sophisticated tool that not only highlights common errors within translations but also facilitates instance-level and system-level performance evaluations. By offering intuitive visualizations and powerful analytical capabilities, Translation Canvas helps users pinpoint specific weaknesses in translation models, enabling more informed model improvements.
Recommended citation: Chinmay Dandekar, Wenda Xu, Xi Xu, Siqi Ouyang, and Lei Li. 2024. Translation Canvas: An Explainable Interface to Pinpoint and Analyze Translation Systems. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 344–350, Miami, Florida, USA. Association for Computational Linguistics.
Download Paper