New paper argues for source-grounded machine translation evaluation

This paper argues that current machine translation evaluation methods, which heavily rely on references, are insufficient for accurately assessing translation adequacy. The authors propose reframing Quality Estimation (QE) as a primary approach for source-grounded adequacy evaluation, rather than a fallback. They advocate for hybrid metrics that prioritize faithfulness to the source text, using references only as supplementary evidence. AI

IMPACT Proposes a shift in machine translation evaluation, potentially improving the accuracy and fairness of automated assessment methods.

RANK_REASON The item is an academic paper discussing a novel approach to machine translation evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• machine translation
• Quality Estimation
• ScienceCast
• Source-Free MT Evaluation Is Not MT Evaluation

• paper
• other

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