Structure-aware modeling
Layout, reading order, hierarchy, and relations across text, tables, forms, figures, charts, and multi-page documents.
Why DocInsights
Documents are structured visual-textual evidence, not just plain text.
Real-world documents in science, healthcare, law, finance, government, and enterprise settings combine natural language with tables, forms, figures, charts, lists, layout cues, and cross-page relationships. Reliable document intelligence requires models and systems that reason across these elements while staying grounded in document evidence.
DocInsights 2026 brings together researchers and practitioners across NLP, Document AI, multimodal learning, information retrieval, knowledge representation, and human-centered systems to advance trustworthy, scalable document understanding.
Scope
Research Themes
Reasoning and grounding
Evidence selection, verification, numerical and temporal reasoning, attribution, and hallucination mitigation for document QA and generation.
Multimodal document understanding
Joint text-layout-vision modeling, OCR-free document models, charts, infographics, forms, and complex visual-textual evidence.
Long and cross-document intelligence
Linking, synthesis, retrieval, and reasoning across long documents, collections, and document-centered workflows.
Evaluation and robustness
Benchmarks, diagnostic analysis, uncertainty, stress tests, OCR/layout noise, and realistic failure modes.
Deployment and governance
Legal, financial, healthcare, scientific, enterprise, and government documents with privacy, auditability, and human-centered review.
Workshop Format
Program Preview
DocInsights 2026 takes place on Thursday, October 29, 2026, in Room F3. See the full program.
Invited talks
Agenda-setting talks from leaders in document AI, multimodal NLP, retrieval, grounding, and evaluation.
Oral presentations
Selected accepted papers presented as talks, with final timing determined after acceptance decisions.
Poster and demo session
Interactive discussion for accepted papers, systems, demos, diagnostic studies, and work in progress.
Shared task spotlight
Shared task details will be announced once the scope, timeline, and participation instructions are final.
Panel
Discussion on trustworthy document intelligence in science, legal, finance, healthcare, government, and enterprise settings.
Early-career spotlights
Short invited highlights and moderation opportunities for emerging work in document-centered NLP.
Challenges
DocSem and Dr.DocBench
Two complementary challenge tracks cover evidence-grounded quantitative reasoning and expert-level document parsing. DocSem has concluded and its final test leaderboard is public; Dr.DocBench runs August 10–October 10, 2026, with submissions closing at 12:59 PM UTC.
Explore both challengesUSD 5,000+ total prize pool
Challenge participants can still submit system papers by September 15, 2026 at 11:59 PM UTC. Authors may indicate an archival or non-archival preference; the review committee will make the final archival/non-archival decision.
Submit a system paperTimeline
Important Dates
All submission deadlines are 11:59 PM UTC-12:00 ("Anywhere on Earth") unless otherwise noted. OpenReview is the submission system of record for exact system timestamps.
Direct Submission Deadline
August 2, 2026 Original deadline
August 10, 2026
12:59 PM UTC (UTC+00:00)
The direct submission deadline has been extended. Submissions may be archival or non-archival.
Submit Direct PaperARR Commitment Deadline
August 30, 2026
11:59 PM UTC-12:00
For eligible papers with completed ACL Rolling Review reviews and meta-review.
Commit ARR PaperAcceptance Notifications
September 13, 2026
Notification date
Decisions will be released through OpenReview.
Camera-ready Deadline
September 27, 2026
Final version deadline
Applies to accepted archival papers.
Workshop Day
Thursday, October 29, 2026
Room F3
The full schedule, shared-task sessions, and poster session are on the program page.
View ProgramUpdates
Latest News
Workshop program released
The DocInsights 2026 program is now available. The workshop takes place on Thursday, October 29, 2026, in Room F3, with keynotes, the DocSem and Dr.DocBench shared-task sessions, and a poster session featuring all 48 accepted papers.
DocSem concluded — final test leaderboard released
DocSem submissions have closed and the final test leaderboard is now public. Thank you to everyone who participated. Participants can still submit a system paper by September 15, 2026 at 11:59 PM UTC. Authors may indicate an archival or non-archival preference; the review committee will make the final archival/non-archival decision.
DocSem held-out test submissions opened September 5
The public test split was released with 1,730 held-out tasks and PDFs without labels. The submission window ran from September 5 through September 10 Anywhere on Earth, with Hugging Face login required, up to three accepted submissions per account, and at least six hours between attempts. Final results were released after the competition concluded.
Three DocSem validation labels corrected
Three organizer-only validation ground-truth labels have now been corrected, most recently on September 3, following additional data review. All existing submissions were rescored, and the leaderboard now reflects the updated results.
Seven training annotations corrected in DocSem
The training split was updated following community feedback. The task definition and data format are unchanged; participants should use the latest release.
Shared-task paper submissions are open
DocSem and Dr.DocBench system paper submissions opened on OpenReview, with a deadline of September 15, 2026 at 11:59 PM UTC. Authors may indicate an archival or non-archival preference; the review committee will make the final archival/non-archival decision.
Dr.DocBench is live on EvalAI
The expert-level document parsing challenge is open through October 10, 2026 at 12:59 PM UTC, with up to USD 3,000 in prizes.
Direct submission deadline extended to August 10
Direct papers are now due August 10, 2026 at 12:59 PM UTC (UTC+00:00). The previous August 2 deadline has been superseded.
OpenReview submission links are live
Direct submissions and ARR commitments are available through separate OpenReview groups.
DocInsights is listed as an accepted EMNLP 2026 workshop
The workshop is officially listed as "DocInsights - Workshop on Document Intelligence and Understanding."
Website launched
Initial workshop information, CFP details, and organizer information were published.
Contact
Questions about the workshop?
Reach the organizing team by email, or follow DocInsights for updates on speakers, program details, and challenge releases.