Haiyue Song, Johannes Eschbach-Dymanus, Hour Kaing, Sumire Honda, Hideki Tanaka, Bianka Buschbeck, Masao Utiyama
The paper introduces Format Reinforcement Learning (FormatRL) to improve structured document translation by optimizing structure-aware rewards, achieving better results on SAP documentation benchmarks.
This research tackles the challenge of translating structured documents like XML or HTML, which contain complex formatting beyond simple sentences. The authors developed a new method called Format Reinforcement Learning (FormatRL) that uses advanced techniques to improve how well the translated document retains its original structure and meaning. By focusing on both the translation quality and the structural similarity of the document, their approach shows significant improvements over existing methods, especially in translating technical documentation.
