emnlp-workflow
GitHub用于规划EMNLP论文全流程的项目管理技能。涵盖ARR周期对齐、基于2026年日历的倒推实验与标注时间表、作者回复窗口安排，以及提交决策（Commit/Revise）和最终呈现后勤协调。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill emnlp-workflow -g -y
SKILL.md
Frontmatter
{
"name": "emnlp-workflow",
"description": "Use when planning an EMNLP project calendar end to end — choosing the ACL Rolling Review cycle that feeds the conference, backward-planning experiments and annotation to the ARR deadline, scheduling the author response window, deciding commit versus revise-and-resubmit, and mapping camera-ready and Budapest presentation logistics."
}
EMNLP Workflow
Use this as the project-management skill for an EMNLP paper. The defining fact: EMNLP has no direct paper deadline of its own. The clock that matters is the ARR cycle aligned with the conference, and missing it usually means committing from an earlier cycle or waiting for the next conference. All dates below are the 2026 configuration (checked 2026-07-08); re-anchor every one against the live calendar.
The 2026 spine
| Date (2026) | Event | Owner |
|---|---|---|
| May 25 | ARR May cycle submission deadline | ARR |
| May 27 | Author registration form (EoD AoE) | ARR |
| Jul 6 | Reviews due from reviewers | ARR |
| Jul 7-13 | Author response + author-reviewer discussion | ARR |
| Jul 30 | Meta-reviews released | ARR |
| Aug 2 | Commitment deadline for EMNLP | Conference |
| Aug 20 | Acceptance notification (Main / Findings / reject) | Conference |
| Sep 20 | Camera-ready | Conference |
| Oct 24-29 | EMNLP 2026, Budapest (hybrid) | Conference |
Parallel option: the Industry Track took direct OpenReview submissions until June 16,
2026 — a genuinely different pipeline (see emnlp-topic-selection for routing).
Backward plan from the ARR deadline
NLP papers die on data and annotation lead times more often than on modeling. Working back from a late-May deadline:
- 10+ weeks out — task definition frozen; annotation guidelines drafted and piloted; if human evaluation is planned, the protocol and pay budget approved now.
- 8 weeks — datasets licensed and downloaded; contamination check on every evaluation set you intend to use with pretrained or API models.
- 6 weeks — main experimental grid running; baselines reproduced first, because a baseline you cannot reproduce invalidates the comparison you were about to run.
- 4 weeks — error analysis begun on real outputs; this is content, not garnish, and it cannot be written the night before.
- 2 weeks — full draft in the ACL template; internal read by someone outside the project hunting for unscoped claims.
- 1 week — Responsible NLP checklist answered against the actual draft; anonymization sweep; citation-resolution pass.
- Deadline week — freeze, upload, calendar the author registration form.
The commitment decision — the step other venues do not have
Between meta-review (Jul 30) and commitment (Aug 2) sits a real strategic choice with a 72-hour fuse in 2026:
- Commit when the meta-review is positive or mixed-positive: SACs read the reviews, scores, and your response, and can accept to Main or Findings.
- Revise and resubmit to a later ARR cycle when reviews identify repairable gaps — the paper returns to the same reviewers where possible, so superficial edits are visible. But a later cycle means a later conference: check which venue the next cycle feeds before choosing patience.
- Withdraw and rework when reviewers rejected the framing, not the execution; resubmission with the same reviewers cannot fix a framing they already dislike.
Draft the commit/revise decision tree before July 30. Three days is enough to execute a decision, not to have the argument.
Coordination structure
Role map for a 4-author EMNLP submission
- Evidence owner: experiment grid, seeds, significance tests, error analysis
- Data owner: licenses, contamination checks, annotation pipeline, agreement stats
- Compliance owner: template, checklist, anonymization, citation resolution
- Response owner: drafts the July response, tracks each reviewer thread
(One person may hold two roles; no role may have zero owners.)
Archive the exact submitted PDF and checklist on deadline day — the July response must quote them precisely, and the camera-ready diff will be audited against them.
If May 25 slips
Because ARR runs continuously, missing the aligned cycle is a detour, not a year lost — but the detour has geometry worth knowing in advance:
- A paper reviewed in an earlier cycle can typically still be committed to EMNLP by August 2 — ACL-family conferences have historically accepted commitments of papers reviewed in preceding cycles. Confirm the eligible-cycle window in the current call; if it holds, this is the rescue path for a project that is review-ready in spring.
- A paper that misses May 25 entirely waits for the next cycle, which feeds whatever conference it is aligned with — possibly not EMNLP. Check the alignment table on the ARR dates page before promising the group "we'll just catch the next one."
- The Industry Track's June 16 direct deadline is not an overflow valve for main-track papers; it reviews against deployment criteria and will bounce a mis-routed research paper.
The week-of-deadline load profile
Plan for the fact that three distinct clocks tick in the same week: the paper PDF (May 25), the author registration form (May 27), and any reciprocal-reviewing qualification data the form demands. In the May 2026 cycle the reviewer pool ran at roughly one AC per twelve submissions, and ARR expanded reciprocal assignments to cope — meaning a multi-paper group could owe several reviews landing in June, exactly when the group planned to rest. Put the reviewing obligation on the same calendar as the submission, with a named owner per obligated author.
Failure modes by phase
- Annotation started at week 6 delivers labels at week 1 with no time to measure agreement — the reviewers will ask for kappa you cannot compute.
- Skipping baseline reproduction until late converts "our method wins" into "our method wins against a baseline we misconfigured," the most common EMNLP rebuttal disaster.
- Treating the July 7-13 window as vacation: it is the only synchronous contact with reviewers in the entire pipeline.
- Forgetting that Findings acceptance may carry no presentation slot — decide before commitment whether Findings satisfies the project's goals (funding reports, degree requirements, visibility) or whether a revise-and-resubmit aims higher.
Output format
[Current phase] scoping / data / experiments / writing / submitted / response / commitment / accepted
[Next hard deadline] <date + owner (ARR vs conference) + source URL>
[Critical path] <three tasks gating the next deadline>
[Commitment posture] commit / revise-resubmit / undecided — with rationale
[Role gaps] <unowned roles from the role map>
Version History
- 9f86f09 Current 2026-07-19 15:17
