DiscoX: Benchmarking Discourse-Level Translation task in Expert Domains

TL;DR

DiscoX is a new benchmark for discourse-level Chinese-English translation in expert domains, highlighting the challenges faced by current models and providing a system for automatic, fine-grained evaluation aligned with human judgment.

Contribution

It introduces DiscoX, a comprehensive discourse-level translation benchmark with an associated evaluation system, filling a gap in expert domain translation assessment.

Findings

Current LLMs lag behind human experts in discourse-level translation.

Metric-S correlates well with human judgments and outperforms existing metrics.

DiscoX reveals significant performance gaps, emphasizing the need for further research.

Abstract

The evaluation of discourse-level translation in expert domains remains inadequate, despite its centrality to knowledge dissemination and cross-lingual scholarly communication. While these translations demand discourse-level coherence and strict terminological precision, current evaluation methods predominantly focus on segment-level accuracy and fluency. To address this limitation, we introduce DiscoX, a new benchmark for discourse-level and expert-level Chinese-English translation. It comprises 200 professionally-curated texts from 7 domains, with an average length exceeding 1700 tokens. To evaluate performance on DiscoX, we also develop Metric-S, a reference-free system that provides fine-grained automatic assessments across accuracy, fluency, and appropriateness. Metric-S demonstrates strong consistency with human judgments, significantly outperforming existing metrics. Our experiments reveal a remarkable performance gap: even the most advanced LLMs still trail human experts on these tasks. This finding validates the difficulty of DiscoX and underscores the challenges that remain in achieving professional-grade machine translation. The proposed benchmark and evaluation system provide a robust framework for more rigorous evaluation, facilitating future advancements in LLM-based translation.