PulseAugur
EN
LIVE 23:23:59

New benchmark HardMTBench stress-tests Chinese-English translation

Researchers have introduced HardMTBench, a new benchmark designed to evaluate Chinese-English machine translation systems on knowledge-intensive domains. Existing benchmarks like FLORES-200 show saturation, with most large language models scoring similarly high. HardMTBench addresses this by covering 12 domains and using an LLM-based judge to assess translation difficulty, terminology, and knowledge density. This new benchmark widens the performance gap between systems and reveals domain-specific weaknesses that previous metrics overlooked. AI

IMPACT HardMTBench aims to reveal specific terminology and knowledge weaknesses in machine translation systems, pushing development beyond general quality metrics.

RANK_REASON The cluster contains an academic paper introducing a new benchmark for machine translation.

Read on arXiv cs.CL →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New benchmark HardMTBench stress-tests Chinese-English translation

COVERAGE [2]

  1. arXiv cs.CL TIER_1 English(EN) · Zheng Li, Mao Zheng, Mingyang Song, Tianxiang Fei ·

    HardMTBench: Stress-Testing Chinese-English Translation on Knowledge-Intensive Domains

    arXiv:2605.28315v1 Announce Type: new Abstract: General-purpose machine translation benchmarks such as FLORES-200 have reached a saturation regime on Chinese-English pairs, where modern large language models cluster within a narrow band of high scores. Across 22 systems, FLORES-2…

  2. arXiv cs.CL TIER_1 English(EN) · Tianxiang Fei ·

    HardMTBench: Stress-Testing Chinese-English Translation on Knowledge-Intensive Domains

    General-purpose machine translation benchmarks such as FLORES-200 have reached a saturation regime on Chinese-English pairs, where modern large language models cluster within a narrow band of high scores. Across 22 systems, FLORES-200 zh-en GEMBA scores fall in a 7.87-point range…