Run this audit before uploading to the ARR OpenReview site. EMNLP does not operate its own
submission portal: the paper enters ACL Rolling Review, and the conference only sees it later
at commitment. Everything below reflects the EMNLP 2026 / ARR May 2026 configuration read on
2026-07-08 — reopen the live CFP and aclrollingreview.org before acting on any number.
Half of the submission rules belong to ARR, half to the conference, and confusing the owners causes real mistakes:
| Question | Who owns the answer | 2026 answer |
|---|---|---|
| Deadline for the reviewed PDF | ARR cycle calendar | May 25, 2026 (May cycle) |
| Page caps and template | ARR CFP | ACL templates; long ≤ 8 content pages, short ≤ 4 |
| Limitations section | ARR CFP | Mandatory, after the conclusion, uncounted |
| Checklist | ARR Responsible NLP checklist | Filed at submission; desk-reject enforced |
| Which venue the paper reaches | Conference commitment | EMNLP commitment due August 2, 2026 |
| Acceptance decision | Conference SACs and PCs | Notification August 20, 2026 |
The May 2026 cycle fed both EMNLP 2026 and AACL-IJCNLP 2026, and the form asked authors to declare the intended venue at submission time. Declare deliberately — the declaration shapes which conference's area chairs eventually argue over your paper.
ACL abolished the anonymity period: you may post and publicize a preprint at any time. The submission itself must still be double-blind. In practice this means the PDF cannot close the loop between itself and your public identity:
# Mechanical leak sweep on the submission PDF and supplementary archive
pdftotext main.pdf - | grep -nEi 'github\.com/[a-z0-9-]+|huggingface\.co/[a-z]|acknowledg|grant'
pdfinfo main.pdf | grep -Ei 'author|creator|producer'
unzip -l supp.zip | grep -Ei '\.git|username|/home/'
Since December 2024, ARR desk-rejects submissions whose Responsible NLP checklist answers are incorrect, incomplete, or misleading. Treat the checklist as testimony, not paperwork: answer it against what the PDF actually contains, and fix the PDF where the honest answer would look bad. Typical mismatches that get caught: claiming license documentation the paper lacks, "yes" on reported hyperparameters that appear nowhere, and human-annotation questions skipped because the ratings came from crowdworkers "informally."
The format choice is strategic, not cosmetic, and it interacts with the audit:
ARR's dual-submission rules prohibit the same work being under review at another archival venue simultaneously; the venue-preference declaration does not create an exception. What is not dual submission: a public arXiv preprint (explicitly allowed, no embargo), a non-archival workshop presentation, or a rejected earlier ARR cycle of the same paper. What is: a concurrent NeurIPS/journal submission of substantially the same content, and quietly hoping the calendars never collide. When any prior version of the paper exists — arXiv, workshop, previous ARR cycle — the OpenReview form has fields for saying so; fill them as disclosures, not confessions, because chairs check overlap mechanically and forgiveness for a declared overlap is routine while discovery of an undeclared one is not.
[EMNLP submission status] ready / blocked / at risk
[Format] template, content pages vs cap, Limitations present
[Anonymity] clean / leaks: <where>
[Checklist] consistent / contradictions: <items>
[Obligations] registration form, reciprocal reviewing, venue declaration
[Fix queue] <ordered actions before the ARR deadline>