Abstract
Context is often proposed as a remedy for ambiguous interface translations, but evidence about document translation does not automatically establish an effect on web-application strings. This paper critically reviews thirteen selected research sources spanning software localization, contextual neural translation, document adaptation, and evaluation. The review distinguishes information supplied during inference from changes to training or architecture, and separates translation context from the context available to evaluators. The synthesis identifies evidence that contextual systems can improve particular discourse phenomena, counterevidence concerning the source of aggregate gains, and substantial limits on transfer to interface tasks. The paper's contribution is an evidence-transfer framework and a factorial estimand for separating local context, terminology instructions, and their interaction. The reviewed findings support targeted evaluation of context rather than a universal quality claim. No new translation experiment, system ranking, or estimate of Linguana performance is reported.
- 0
- Views
- 1
- Downloads
- —
- Citations
- 0
- Cited here
not yet indexed
papers on ScholarXIV
- Subjects
- cs.AI
- Keywords
- Software localization · document-level translation
- AI authorship disclosure
- Human Written
- Licence
- CC BY 4.0
- Submitted as
- PDF only
Submission history
-
First posted.