Wikimedia invests in OKA
Funding Provisional 62% confidence first seen
The BBC reports that Wikimedia funded a study run by OKA (Open Knowledge Association), a Swiss non-profit that translates Wikipedia articles. The study analyzed 119 articles across 10 language pairs and more than 1,000 hours of translator work, finding that 27% of AI-generated text was corrected before publication. This matters because it quantifies how much human oversight may still be needed when AI drafts translations for high-trust reference sources like Wikipedia.
The deal
Deal terms as reported in the coverage below.
Decision brief
- What changed
- Wikimedia funded an OKA study on AI-assisted Wikipedia translation that reviewed 119 articles across 10 language pairs and more than 1,000 hours of translator work. The study found that 27% of AI-generated text was corrected before publication.
- Why it matters
- For leaders using AI in high-trust content workflows, this provides a concrete datapoint that machine-drafted translation still required substantial human correction in this sample. It suggests cost, staffing, and quality plans should assume ongoing editorial review rather than fully automated publishing, especially where accuracy and trust are central.
- Evidence
- The BBC reports the finding and attributes it to an OKA study funded by Wikimedia. The coverage presents one reported source and describes the result consistently as evidence that human oversight remains necessary in higher-stakes translation contexts.
- What remains uncertain
- The coverage does not specify the article topics, quality thresholds, model versions, or whether the 27% correction rate generalizes beyond Wikipedia translation. It is also unclear how correction rates would differ by language pair, domain, or with newer AI systems and review processes.
- Monitor next
- Watch for follow-up disclosures from Wikimedia or OKA on per-language results, methodology details, and whether they change operational guidance for AI-assisted translation workflows.
Analytical support, not advice — assumptions and open questions stated above.