Document-Level Machine Translation with Context-Aware Memory
Xuan-Dung Doan, Viet-Hoang Vu, Ngoc-Dung Nguyen, Viet-Trung Tran · 2024
Document-level machine translation offers numerous advantages over sentence-level machine translation, particularly the ability to integrate contextual information between sentences. Recent research has explored context-awareness and recurrent memory to enhance machine translation accuracy. In this paper, we present VT-Trans, a transformer-based machine translation that incorporates recurrent memory and contextaware information in translation, improving the ability to remember contextual information. VT-Trans achieves state-of-the-art results in all three popular datasets, outperforming the previous works.