naacl-submission
GitHub针对NAACL投稿的ARR周期审计技能。检查当前ARR周期是否衔接NAACL，验证PDF格式、匿名化、限制声明及OpenReview表单合规性，规避撤稿风险。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill naacl-submission -g -y
SKILL.md
Frontmatter
{
"name": "naacl-submission",
"description": "Use when a paper aimed at NAACL is about to be uploaded to an ACL Rolling Review cycle — verifying that the chosen cycle can still reach the next NAACL edition, that the PDF passes ARR's format and anonymization gates, and that every declaration on the OpenReview form survives a desk-reject audit."
}
NAACL Submission
NAACL has no submission portal of its own. What you upload goes to an ACL Rolling Review (ARR) cycle on OpenReview; NAACL enters the picture weeks later, when you commit the finished review package to the conference. This skill audits the upload with that split in mind. The single most NAACL-specific check comes first, because it is the one authors from other venues never think to make.
Check zero: does this cycle reach a NAACL at all?
NAACL is not annual. The chapter historically stands down in years when the main ACL meeting is hosted in the Americas — there was no NAACL 2023 and no NAACL 2026 (ACL 2026 ran in San Diego, July 2-7, 2026). As of 2026-07-08 the next edition is NAACL 2027, whose dates, venue, and feeding ARR cycles were still unannounced (待核实 via https://2027.naacl.org/ and https://aclrollingreview.org/dates). Before polishing anything:
1. Open aclrollingreview.org/dates → does any listed cycle name NAACL?
2. If yes → that cycle's submission date is your real deadline.
3. If no → your reviewed paper can still wait in ARR and commit later,
or commit to a sibling venue (EMNLP/EACL/AACL) instead.
4. Record the answer with a date — this table changes without notice.
Historical anchor: NAACL 2025 (Albuquerque, April 29 - May 4, 2025) drew on the ARR October 2024 cycle — ARR upload October 15, 2024; reviews and meta-reviews back December 12, 2024; final versions for Main and Findings due February 8, 2025; all deadlines 11:59 PM UTC-12. Treat those numbers as shape, not schedule.
The gates ARR applies before any human reads the science
| Gate | Passing condition | Cost of failure |
|---|---|---|
| Length | Long ≤ 8 content pages, short ≤ 4; references and the Limitations section sit outside the count | Rejected without review |
| Template | Current *ACL style files, unaltered geometry | Treated as a length violation |
| Limitations | A real Limitations section after the conclusion | Explicit desk-reject ground |
| Anonymity | No names, affiliations, thank-yous, or de-anonymizing links anywhere in PDF or supplements | Rejected without review |
| Checklist | Responsible NLP checklist filed and consistent with the PDF | Desk reject for misleading answers |
| Resubmission link | Prior ARR ID plus a change summary if this work was reviewed before | Desk reject if omitted |
Two subtleties worth spelling out for NAACL-bound work specifically. First, papers on Indigenous or low-resource languages of the Americas often carry community acknowledgements and data-partnership credits — those must come out of the reviewed version and return only at camera-ready. Second, a preprint is allowed at any time under current ARR policy, but the form asks you to declare it; an inaccurate declaration is itself a desk-reject ground, and truthfully declaring "no preprint" buys borderline-case priority you forfeit by posting one mid-cycle.
Form fields that quietly decide your fate
- Preferred venue. When a cycle feeds more than one conference, the form asks where you intend to commit. Naming NAACL here routes your paper toward the chapter's area-chair pool; it is a signal, not a contract.
- Reviewing obligations. ARR expects author-side reviewing or nomination; ignoring the field can block the submission outright.
- Languages studied. Fill it precisely — a NAACL commitment later puts your paper before a community that reads "Spanish, Quechua, English" as substantive information, not metadata.
Supplements at upload time
- Software and data go in as archives (
.zip/.tgz) through the OpenReview fields, not as links to personal cloud storage that can log visitors. - Reviewers promise to judge the paper from its content pages alone, so a supplement may deepen evidence but must never complete an argument.
- Name files blandly (
code.zip,data_card.pdf); a lab-convention filename is an anonymity leak the sweep below should catch.
Worked micro-case
A team studying code-switched Spanish-English dialogue plans to submit in July 2026 "for NAACL." The calendar check shows no cycle currently names NAACL 2027. Their real options: enter the next open cycle now and let the reviewed package wait for a NAACL commitment window (reviews stay valid across cycles), target EMNLP/AACL via the May cycle instead, or hold for the cycle NAACL 2027 eventually announces. The audit's job is to force that decision explicitly rather than let a vague "submit to NAACL" drift into a missed conference entirely.
Ordered pre-upload pass
- Reconfirm the live cycle calendar and the NAACL question above.
- Rebuild the PDF from the untouched current style files.
- Run an anonymization sweep over text, footnotes, metadata, and every linked artifact; check supplement archives file by file.
- Reconcile each checklist answer against the section it points to.
- If resubmitting, write the change summary a skeptical AC would accept.
- Upload early enough to fix a rejected PDF before the AoE cutoff.
Mechanical sweeps before the AoE cutoff
pdfinfo main.pdf— creator/author metadata fields must be empty or generic; LaTeX toolchains leak usernames here.- Search the PDF text for your institution's name, grant numbers, and the string fragments of every author's homepage.
- Open each supplement archive on a machine that has never touched the project; hidden OS files and absolute paths surface immediately.
- Recompute the content-page count after the last figure move; page 8.05 is page 9 to a compliance script.
- Remember the clock: 11:59 PM UTC-12 is later than every timezone in the Americas, which breeds complacency — OpenReview load spikes at the hour are the real risk, not the math.
Output format
[Cycle sanity] <cycle name -> feeds NAACL? yes/no/unverified>
[Gate results] length / template / limitations / anonymity / checklist / resubmission
[Declaration risks] <preprint, venue preference, reviewing obligation>
[Blocking fixes] <ordered, with owner>
[Upload verdict] go / hold / re-plan cycle
Version History
- 9f86f09 Current 2026-07-19 17:05
