EACL Submission
Run this audit before uploading to the ACL Rolling Review (ARR) cycle that feeds your target
EACL edition. EACL papers are never submitted to the conference directly: they enter an ARR
cycle on OpenReview, collect reviews and a meta-review, and are later committed to EACL,
which decides main / Findings / reject. Reopen the current EACL CFP and aclrollingreview.org/cfp
before trusting any number below.
The one-cycle reality (EACL's defining trap)
- Confirm which single ARR cycle feeds your edition, and treat it as unrepeatable. For
EACL 2027 the organizers named the August 3, 2026 cycle the only viable cycle — the
conference sits early in the year, so there is no later cycle to slip into. Unlike ACL,
a missed or wasted cycle here loses the whole edition.
- The ARR deadline is the hard one. Commitment is a separate, later step (October 11, 2026
for EACL 2027) — see
eacl-workflow.
- With no fallback, a thin-but-submittable draft is a real dilemma: submitting burns the only
cycle, skipping loses the year. Decide deliberately, not by deadline reflex.
Format gate
- Long papers: up to 8 content pages. Short papers: up to 4. Both get unlimited pages
for references and appendices, 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 — 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 and can be rejected without review.
- 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 must be fully anonymized: no names, affiliations, acknowledgements, or
self-revealing "in our prior work we showed" citations.
- Since January 2024 there is no anonymity period: a non-anonymous preprint is allowed at
any time. But declare preprint status at submission; choosing "no non-anonymous preprint"
earns award eligibility and borderline priority — 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
personally identifying hosting are disallowed.
Declarations that gate the ARR 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 EACL |
| AI assistance disclosure |
Checklist item answered honestly |
Undisclosed generative-AI use → policy breach |
Long vs short
A short paper is a genre, not a truncated long paper: one claim, four pages. A compressed
eight-page study reads as thin (see the fastText exemplar in
../../resources/exemplars/library.md). See eacl-topic-selection for the full choice.
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 (languages, data, compute, scope), 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.
- Confirm the OpenReview title/abstract 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,
and there is no next cycle.
Frequently mishandled cases
- Assuming a fallback cycle exists. For a single-cycle edition it does not; plan as if this
is the only attempt.
- Appendix used as overflow body. Content pages must carry the argument; "see Appendix B"
for a core mechanism fails self-containedness.
- arXiv v2 posted mid-cycle after declaring no preprint: a desk-reject liability.
- Dataset "available on request": allowed, but checklist answers must then explain
reproduction honestly rather than imply release.
Output format
[ARR readiness] Ready / Needs fixes / Not ready
[Cycle plan] <single viable cycle -> EACL edition; fallback = none>
[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 one cycle deadline>
1---2name: eacl-submission3description: Use when auditing an EACL submission before its single ACL Rolling Review cycle deadline, covering ACL-template compliance, the 8-page long / 4-page short content caps, the mandatory Limitations section, the Responsible NLP checklist, anonymization, preprint and resubmission declarations, and the no-fallback-cycle timing trap.4---56# EACL Submission78Run this audit before uploading to the ACL Rolling Review (ARR) cycle that feeds your target9EACL edition. EACL papers are never submitted to the conference directly: they enter an ARR10cycle on OpenReview, collect reviews and a meta-review, and are later *committed* to EACL,11which decides main / Findings / reject. Reopen the current EACL CFP and `aclrollingreview.org/cfp`12before trusting any number below.1314## The one-cycle reality (EACL's defining trap)1516- Confirm **which single ARR cycle feeds your edition, and treat it as unrepeatable.** For17 EACL 2027 the organizers named the **August 3, 2026 cycle the only viable cycle** — the18 conference sits early in the year, so there is **no later cycle to slip into**. Unlike ACL,19 a missed or wasted cycle here loses the whole edition.20- The ARR deadline is the hard one. **Commitment is a separate, later step** (October 11, 202621 for EACL 2027) — see `eacl-workflow`.22- With no fallback, a thin-but-submittable draft is a real dilemma: submitting burns the only23 cycle, skipping loses the year. Decide deliberately, not by deadline reflex.2425## Format gate2627- Long papers: up to **8 content pages**. Short papers: up to **4**. Both get unlimited pages28 for references and appendices, plus unlimited space after the conclusion for the **required29 Limitations section** and an optional ethics statement.30- A missing Limitations section or overlength content is an explicit desk-reject condition31 under the ARR CFP — the most common self-inflicted rejection.32- Use the official ACL LaTeX/Word style files unmodified; margin, font, or spacing tampering33 is treated as a length violation and can be rejected without review.34- Appendices go after references in the same PDF; reviewers are not obliged to read them, so35 the 8 or 4 content pages must carry the whole argument.3637## Anonymity gate3839- The submission must be fully anonymized: no names, affiliations, acknowledgements, or40 self-revealing "in our prior work we showed" citations.41- Since January 2024 there is **no anonymity period**: a non-anonymous preprint is allowed at42 any time. But declare preprint status at submission; choosing "no non-anonymous preprint"43 earns award eligibility and borderline priority — then posting one anyway before meta-review44 release is a desk-reject offense.45- Supplementary archives, repo links, and dataset pages must be anonymized too; links to46 personally identifying hosting are disallowed.4748## Declarations that gate the ARR form4950| Form item | What ARR checks | Failure mode |51|---|---|---|52| Responsible NLP checklist | Completed, consistent with the PDF | Misleading answers → desk reject |53| Resubmission link | Prior ARR submission ID + change summary | Unacknowledged resubmission → desk reject |54| Preprint declaration | Matches what is actually on arXiv | Broken "no preprint" pledge → desk reject |55| Dual submission | Not under review anywhere archival | Violation → reject at ARR or EACL |56| AI assistance disclosure | Checklist item answered honestly | Undisclosed generative-AI use → policy breach |5758## Long vs short5960A **short paper is a genre, not a truncated long paper**: one claim, four pages. A compressed61eight-page study reads as thin (see the fastText exemplar in62`../../resources/exemplars/library.md`). See `eacl-topic-selection` for the full choice.6364## Final-72-hours sequence for an NLP paper65661. Rebuild the PDF from the untouched style files and re-run the page count on **content pages67 only**.682. Re-read Limitations: it must be substantive (languages, data, compute, scope), not a ritual69 paragraph.703. Fill the Responsible NLP checklist against the actual PDF — every "yes" needs a section71 number.724. Sweep the supplement archive for author names in paths, notebook metadata, git history, and73 license headers.745. Confirm the OpenReview title/abstract match the PDF and the preprint declaration matches75 arXiv reality.766. Submit hours early; ARR deadlines are AoE but OpenReview outages are not an accepted excuse,77 and there is no next cycle.7879## Frequently mishandled cases8081- **Assuming a fallback cycle exists.** For a single-cycle edition it does not; plan as if this82 is the only attempt.83- **Appendix used as overflow body.** Content pages must carry the argument; "see Appendix B"84 for a core mechanism fails self-containedness.85- **arXiv v2 posted mid-cycle** after declaring no preprint: a desk-reject liability.86- **Dataset "available on request":** allowed, but checklist answers must then explain87 reproduction honestly rather than imply release.8889## Output format9091```text92[ARR readiness] Ready / Needs fixes / Not ready93[Cycle plan] <single viable cycle -> EACL edition; fallback = none>94[Format gate] <length/Limitations/style-file findings>95[Anonymity + declarations] <checklist/preprint/resubmission/dual-submission findings>96[Desk-reject risks] <ranked list>97[Fix order] <what to change before the one cycle deadline>98```
brycewang-stanford