[2Win5] Poster session 2

[2Win5-52] Incorporating Reading Information in Japanese-to-English Machine Translation of Man'yoshu Monument Commentary Using a Large Language Model

〇Aoba Yamada¹, Katsuhito Sudoh¹ (1.Nara Women's University)

Keywords:AI, machine translation, GPT

This study tackles Japanese-to-English machine translation (MT) of Manyō poetry monument commentary using a Large Language Model (LLM). Japanese-to-English MT often faces difficulties translating Japanese words requiring prior knowledge, especially the reading of proper nouns. We also use an LLM to find words with unknown kanji reading before MT. Those words are then passed to the LLM with their reading information annotated manually, for Japanese-to-English translation.In our experiments, we observed some improvements in translation accuracy compared to LLM-based MT without reading information, especially when using romanized kanji reading.