Multi-Pipeline Approach for Japanese-to-English Translation

This paper is a technical report detailing a multi-pipeline machine translation (MT) system aimed at translating repeated elements from Japanese to English. The system was developed for the Non-Repetitive Translation Task at the Ninth Conference on Machine Translation (WMT24) and focuses on reducing redundancy while ensuring high translation quality. It employs MeCab, a Natural Language Processing (NLP) tool, to identify repeated elements and utilizes the Large Language Model (LLM) Claude Sonnet 3.5 for translation and proofreading. The system successfully translated 89.79% of repeated instances in the test dataset, achieving an average translation quality score of 4.60 out of 5, which is significantly higher than the baseline score of 3.88. The report also discusses challenges encountered, particularly in identifying standalone noun-suffix elements and instances of consistent translations or mistranslations. The document outlines the methodologies and processes involved in the development and evaluation of the MT system.