中国科学院自动化研究所机构知识库

Text image machine translation (TIMT) which translates source language text images into tar- get language texts has attracted intensive at- tention in recent years. Although the end- to-end TIMT model directly generates target translation from encoded text image features with an efficient architecture, it lacks the rec- ognized source language information resulting in a decrease in translation performance. In this paper, we propose a novel Cross-modal Cross-lingual Interactive Model (CCIM) to in- corporate source language information by syn- chronously generating source language and tar- get language results through an interactive at- tention mechanism between two language de- coders. Extensive experimental results have shown the interactive decoder significantly out- performs end-to-end TIMT models and has faster decoding speed with smaller model size than cascade models.