$M^2PO$ framework enhances LLM machine translation accuracy

A new framework called $M^2PO$ has been developed to improve machine translation by Large Language Models (LLMs). This method addresses a key issue where current models often favor fluent but inaccurate translations, overlooking partial errors like hallucinations and omissions. $M^2PO$ uses a dual-perspective approach to separate fluency from faithfulness and a multi-pair objective to better capture subtle errors. Experiments show that a 9B parameter model using $M^2PO$ can match the performance of proprietary models like GPT-4o and Gemini-2.0-Flash on benchmarks such as WMT23, WMT24, and FLORES-200+. AI

IMPACT This research could lead to more accurate and reliable machine translation systems by addressing subtle error types.

RANK_REASON This is a research paper detailing a new framework for machine translation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• FLORES-200+
• Gemini 2.0 Flash
• GPT-4o
• Hao Wang
• Large Language Models
• $M^2PO$
• machine translation
• WMT23
• WMT24

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →