ACL Submission
Run this audit before uploading to an ACL Rolling Review (ARR) cycle on OpenReview. ACL papers are not submitted to the conference directly: they enter an ARR cycle, collect reviews and a meta-review, and are later committed to ACL. Reopen the current ARR CFP at aclrollingreview.org/cfp and the target conference's call before trusting any number below.
Two-step reality check
- Confirm which ARR cycles feed your target ACL edition. ACL 2026 accepted work from the October 2025 and January 2026 cycles; committing requires a complete review package, so the last eligible cycle is earlier than the conference date suggests.
- The ARR deadline is the hard one. Commitment is a separate, later form with its
own deadline (March 14, 2026 for ACL 2026) — see
acl-workflowfor the calendar. - All ARR deadlines run 11:59 pm UTC-12 (anywhere on Earth).
Format gate
- Long papers: up to 8 content pages. Short papers: up to 4. Both get unlimited pages for references, plus unlimited space after the conclusion for the required Limitations section and an optional ethics statement.
- A missing Limitations section or overlength content is an explicit desk-reject condition under the ARR CFP — this is the most common self-inflicted rejection.
- Use the official ACL LaTeX/Word style files unmodified; margin, font, or spacing tampering is treated as a length violation.
- Appendices go after references in the same PDF; reviewers are not obliged to read them, so the 8 or 4 content pages must carry the whole argument.
Anonymity gate
- The submission itself must be fully anonymized: no names, affiliations, acknowledgements, or self-revealing "we previously showed" citations.
- Since January 2024 there is no anonymity period: you may post a non-anonymous preprint at any time. But you must declare preprint status at submission, and choosing "no non-anonymous preprint" earns award eligibility and priority for borderline decisions — then posting one anyway before meta-review release is a desk-reject offense.
- Supplementary archives, repo links, and dataset pages must be anonymized too; links to tracked hosting (personal Dropbox/Drive) are disallowed.
Declarations that gate the submission form
| Form item | What ARR checks | Failure mode |
|---|---|---|
| Responsible NLP checklist | Completed, consistent with the PDF | Misleading answers → desk reject |
| Resubmission link | Prior ARR submission ID + change summary | Unacknowledged resubmission → desk reject |
| Preprint declaration | Matches what is actually on arXiv | Broken "no preprint" pledge → desk reject |
| Dual submission | Not under review anywhere archival | Violation → reject at ARR or venue |
| AI assistance disclosure | Checklist item E answered honestly | Undisclosed generative-AI use → policy breach |
Resubmission discipline
ARR is built for revise-and-resubmit across cycles. If this paper was reviewed in an earlier cycle:
- Link the previous submission ID in the form — omitting it is desk-rejectable.
- Attach a summary of changes plus a point-by-point response to every weakness named in the old reviews.
- If the previous meta-review scored 1, a resubmission within six months without wholesale revision can be desk rejected; cosmetic edits waste a cycle.
- Decide whether to request the same reviewers — continuity helps when the revision genuinely answers them.
Final-72-hours sequence for an NLP paper
- Rebuild the PDF from the untouched style files and re-run the page count on content pages only.
- Re-read Limitations: it must be substantive (scope, languages, data, compute), not a ritual paragraph.
- Fill the Responsible NLP checklist against the actual PDF — every "yes" needs a section number.
- Sweep the supplement archive for author names in paths, notebook metadata, git history, and license headers.
- Verify title/abstract on OpenReview match the PDF, and the preprint declaration matches arXiv reality.
- Submit hours early; ARR deadlines are AoE but OpenReview outages are not an accepted excuse.
Cycle-choice quick checks
- Is the target conference actually accepting commitments from this cycle? Each edition names its eligible cycles; a mismatch strands the reviews.
- Does the team have bandwidth during this cycle's author-response window, roughly five weeks after the deadline?
- Would waiting one cycle allow the missing experiment that reviews will otherwise demand? A weak first cycle costs more than it saves, because the package follows the paper.
- Is any co-author's other submission in the same cycle a conflict for reviewer assignment? Register conflicts and profiles before deadline week.
Frequently mishandled cases
- Short paper padded to eight pages: reviewers rate contribution against format; a four-page idea submitted long scores as thin, not as generous.
- Appendix used as overflow body: content pages must carry the argument; a method section that says "see Appendix B" for its core mechanism fails the self-containedness expectation.
- arXiv v2 posted mid-cycle after declaring no preprint: this converts a declaration into a desk-reject liability — update plans, not the preprint.
- Dataset "available on request": allowed, but checklist answers must then explain reproduction honestly rather than implying release.
- Same work at a non-archival workshop: usually permitted, but most *ACL workshop proceedings ARE archival — verify the workshop's archival status in writing before assuming.
Output format
text[ARR readiness] Ready / Needs fixes / Not ready [Cycle + venue plan] <ARR cycle -> intended commitment venue> [Format gate] <length/Limitations/style-file findings> [Anonymity + declarations] <checklist/preprint/resubmission/dual-submission findings> [Desk-reject risks] <ranked list> [Fix order] <what to change before the cycle deadline>
