Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› eacl-author-response
eacl-author-response
GitHub指导在ACL Rolling Review rebuttal期间撰写EACL作者回复。涵盖区分阻断性异议与偏好、针对NLP常见质疑的应对策略、窗口期内实验执行纪律、向行动编辑提交元评论的要点，以及标记缺陷评审的流程，旨在最大化录用机会。
Trigger Scenarios
需要撰写EACL论文的反驳信
处理ARR rebuttal阶段的审稿意见
决定哪些实验需在反驳窗口内完成
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill eacl-author-response -g -y
SKILL.md
Frontmatter
{
"name": "eacl-author-response",
"description": "Use when drafting an EACL author response during the ACL Rolling Review rebuttal window, covering how to triage reviewer points, address the action editor who writes the meta-review, answer NLP objection patterns, run in-window experiments, flag deficient reviews, and decide respond-now versus revise-for-a-later-venue given EACL's single cycle."
}
EACL Author Response
Use this to work the ARR rebuttal window for a paper aimed at EACL. In ARR the response is addressed to reviewers and to the action editor who will synthesize a meta-review; that meta-review, not any single score, is what EACL's area chairs read at commitment. Reopen the current ARR author guidelines before drafting.
Read the room first
- Identify the action editor's lens: the meta-review is where reviews get weighed, so a response that moves the action editor matters more than one that merely satisfies the harshest reviewer.
- Separate blocking objections (a claim reviewers think is unsupported, a missing baseline, a reproducibility gap) from preference comments (a different framing, an extra dataset a reviewer would like). Spend the word budget on the blocking set.
- EACL's single cycle raises the stakes of this window: for an edition like EACL 2027 there is no next cycle to defer to, so a fixable objection should be fixed now if the experiment fits the window.
Objection-to-move table
| Reviewer objection pattern | Strong response move | Weak response move |
|---|---|---|
| "Claim broader than evidence" | Narrow the claim in text; point to the exact table | Argue the claim is "obviously" fine |
| "Missing tuned/LLM baseline" | Add it if runnable in-window; else scope the claim | Dismiss the baseline as unfair |
| "Not enough languages" | Add the runnable subset; state the residual in Limitations | Assert results "would generalize" |
| "Possible contamination" | Report an overlap check or decontamination result | Ignore or hand-wave |
| "Significance unclear" | Report a test / CIs over seeds | Cite a single-run delta |
Drafting rules
- Lead each reply with the concrete change ("We added X; see revised Table 2"), then the reasoning — reviewers skim.
- Do not introduce unsupported new claims or results you cannot show; a promise you cannot substantiate reads worse than an honest limitation.
- Keep the tone factual and non-defensive; the action editor is reading for whether you engaged, not whether you "won."
- Respect anonymity: no links that deanonymize, no "as we said in our NAACL paper."
Runnable-experiment discipline
For each requested experiment:
fits_window? = can it finish and be written before the response deadline?
changes_claim? = would the result actually alter a reviewer's objection?
-> run only if fits_window AND changes_claim
-> otherwise: acknowledge, scope the claim, and record it in Limitations
Do not start a sprawling experiment that cannot land in time; a half-reported result invites a sharper objection.
Flagging a deficient review
- If a review is unsupported, off-topic, or hostile, you may flag it to the action editor through the proper ARR channel rather than fighting it point-by-point in the public reply.
- Document the specific deficiency (factual error, no justification) — flagging is for genuine process failures, not for a low score you dislike.
When not to respond hard
- If the reviews reveal a structural gap no in-window experiment can close, the honest move
may be to accept the outcome and plan a later venue — but remember EACL's calendar: the
next *ACL opportunity is a different conference, so
eacl-topic-selectionandeacl-workflowshould drive that call.
Output format
[Response plan] <blocking objections, ranked>
[Action-editor angle] <what the meta-review needs to hear>
[In-window experiments] <run / defer, with reason>
[Reviewer replies] <one concise reply per reviewer>
[Deficient-review flags] <if any, with documented reason>
[Route decision] <respond-and-commit / revise-for-later-venue>
Version History
- 9f86f09 Current 2026-07-19 15:14
