An AI-assisted translation and collaboration workbench for professional translation teams.
DeepTrans Studio starts from the observation that high-quality translation depends not only on one-shot model output, but also on expert intervention, shared team judgment, and consistent terminology and style. The project builds segment-aligned editing, team memory, terminology management, multi-agent translation, and quality review for multilingual document translation.
Professional translation is difficult because of repeated revision, terminology consistency, layout preservation, and reuse of team knowledge. Standard machine translation returns outputs but rarely captures why experts changed them or how future work should inherit those decisions.
The system implements segment alignment, post-editing, terminology extraction and locking, team-memory retrieval, multi-agent candidate generation, quality checks, and layout-preserving PDF translation, turning expert edits into reusable team knowledge.
The project produced a usable translation workbench and demo system, with research outputs connected to a CSCW '26 Demo paper and the ACL 2026 BabelDOC system paper. The CSCW '26 Demo paper does not yet have a public paper page, so no paper link is shown here for now.