Translation review
Check a translation against its source, the glossary and the style guide.
Definition
Finding errors in a translation: mistranslations, omissions, additions, glossary violations and style problems, with the location and a suggested fix for each. It is the editor’s and reviewer’s step, not a new translation.
Input and output
- Input: A source text, its translation, and optionally a glossary and style guide.
- Output: A list of issues, each with location, type, explanation and suggested fix.
Examples
None yet. Add a short real example with its correct answer, and say where it comes from.
Edge cases
- Deliberate interpretive choices that are not errors.
- Omissions of a single particle that change the meaning.
- Style issues that depend on the audience.
Languages and traditions
- Each source and target pair is its own task variant.
How it is measured
No benchmark yet: help build one. See below.
Help wanted
Translations with reviewers’ real comments, so a benchmark can test whether a model finds the same issues. A first benchmark can be small: twenty to fifty items. See how to submit a benchmark.
Further reading
- Subtracting the Typological Baseline (Drupchen, 2026, Tibetan → French): why word-by-word comparison flags good translations, and a four-step review procedure that counts a loss only when the source made a real choice and the target language had a way to say it.
- Two strategies for improving machine translation from Tibetan (Drupchen, 2026): the practical guide.
Related tasks
Found in workflows
- Translating a Buddhist text (Kurt Keutzer)

BuddhistAI Knowledge base