Guide
ACL Rolling Review: the 19% is not your acceptance rate
The famous 18.9% and 22.16% divide the acceptances by every submission in the feeder cycles, most of which never reached a venue. Count only the papers that did and the picture inverts.
Published 14 August 2026
Reviews already in? Then the commitment step comes fast and the response is the only lever left. How to write a rebuttal
The short version
- The hard part is not the acceptance rate at the end. It is getting to the point where committing is a real option.
- Then choosing inside a window three days wide. That window, not the rate, is what people miss.
The number nobody publishes
Two rates, one set of acceptances, different denominators
Both venues report every figure needed to compute the conditional rate, and neither computes it.
| Figure | ACL 2026 | EMNLP 2025 |
|---|---|---|
| Papers committed to the venue | 6,564 | 4,512 |
| Accepted, main conference | 2,296 | 1,811 |
| Accepted, Findings | 2,163 | 1,417 |
| Acceptance rate the venue publishes, main conference | 18.9% | 22.16% |
| Accepted as a share of papers committed | 67.9% | 71.5% |
| Reached commitment, as a share of the feeder cycles | 54.0% of 12,148 | about 55% of roughly 8,200 |
Counts and published rates from the ACL 2026 front matter and the EMNLP 2025 Findings front matter, read August 2026. The last two rows are our arithmetic; EMNLP's pool is back-solved from 22.16% and approximate. ARR's author guidance gives no Findings opt-in or opt-out at commitment.
The filter is attrition, not the accept rate
EMNLP 2025 says plainly that 778 of its submissions were withdrawn before three reviews and 670 desk-rejected, and that both stayed in the denominator. ARR's May 2026 letter names the pressure: 17,087 submissions against January's 10,518, while qualified area chairs went from 1,415 to 1,424 and reviewers from 8,633 to 10,636, which ARR says is not growth in proportion with submissions.
The three days
The meta-review lands 72 hours before the decision is due
In every cycle ARR has fully dated, and the cycle end is the venue's commitment deadline.
| Cycle | Submission | Meta-reviews | Cycle end, and the commitment deadline | Venue committing from it |
|---|---|---|---|---|
| May 2026 | 25 May | 30 July | 2 August | EMNLP 2026, AACL 2026 |
| August 2026 | 3 August | 8 October | 11 October | EACL 2027 |
| October 2026 | 12 October | To be announced | 23 December | NAACL 2027, COLING 2027 |
Dates from the ARR cycle calendar, which states no timezone; the three-day gap is our arithmetic. Withdrawal, author response and commitment rules from the ARR call for papers; the waiver from ARR's May 2026 cycle letter, about one cycle only. Read August 2026.
Three days is a choosing window, not a fixing window
Nothing about the paper can change in them: the only question they answer is which venue gets the reviews you already hold. The moment you could act was the author response, weeks earlier, and that is text only with no external links, or ARR says the paper will be desk-rejected.
Do not solve a bad cycle by withdrawing, unless ARR says otherwise
Once a paper has one review it cannot be quietly withdrawn and resubmitted as new, and withdrawing more than 48 hours after the submission deadline can block resubmission until the second subsequent cycle. Waiting costs less: after the meta-review the paper is no longer under review and can be committed elsewhere. One exception is on record. ARR's May 2026 cycle letter says that for authors who wish to withdraw their submissions in this cycle to avoid the high reciprocal reviewing load, it will waive the withdrawal penalty. Precedent, not policy.
The anchors, in the form's words
Soundness and excitement are not the same ruler twice
Reading one as though it were another is how authors talk themselves into the wrong panic.
| Score | Soundness | Excitement |
|---|---|---|
| 5 | Excellent. One of the most thorough studies the reviewer has seen, given its type. | Highly exciting. They would recommend the paper to others and attend its presentation. |
| 4 | Strong. Sufficient support for all of its claims. Extra experiments could be nice, but are not essential. | Exciting. They would mention it to others and make an effort to attend its presentation. |
| 3 | Acceptable. Sufficient support for the main claims. Minor points may need extra support or detail. | Interesting. They might mention some points to others, and attend the presentation if there is time. |
| 2 | Poor. Some of the main claims are not sufficiently supported. Major technical or methodological problems. | Potentially interesting. The paper does not resonate with this reviewer, but it might with others in the *ACL community. |
| 1 | Major issues. Not yet thorough enough to warrant publication, or not relevant to ACL. | Not exciting. It does not resonate with this reviewer, and they do not think it would with others. |
Anchors from the ARR review form, condensed from its first-person wording; the form calls excitement subjective, and its half steps of 4.5 to 1.5 carry no description. Three mandatory scores since February 2025, from the ARR reviewer guidelines. ARR's author guidance says scores do not directly indicate acceptance. Read August 2026.
The tinted row is the trap: a 2 means opposite things
Soundness 2 is a fault finding. Excitement 2 is a statement of taste, and the tinted row gives the form's own wording for both. Average them and you spend the author response arguing with the one number that was never a claim about the work.
Two rulers
Your resubmission rights are written on a scale ARR publishes twice
| Score | Current scale, since February 2025 | Superseded scale, still published |
|---|---|---|
| 5 | Consider for award, the form's top 2.5% | The paper is largely complete with no clear points of revision |
| 4.5 | Borderline award | No such score |
| 4 | Could be accepted to an *ACL conference | Minor points that may be revised |
| 3.5 | Borderline conference | No such score |
| 3 | Could be accepted to Findings | Major points |
| 2.5 | Borderline findings | No such score |
| 2 | Resubmit next cycle | Significant revisions to reach a publishable state |
| 1.5 | Resubmit after next cycle | No such score |
| 1 | Do not resubmit | Not likely to be publishable at an *ACL venue even after revisions |
The current column is the reviewer's overall assessment on the ARR review form. ARR's area chair guidelines say the metareview score changed in February 2025 to be more aligned with reviewer scores: 5 consider for award to 1 do not resubmit, same half steps. The superseded column is the action editor form page, source file September 2023, still linked from the call for papers as current. Read August 2026.
The tinted row is the one the old ruler cannot express
ARR's call for papers states that a prior version scored 1.5 cannot be resubmitted in the next cycle, that one scored 1 can never be resubmitted, and that it reserves the right to desk reject a paper scored 1 returning within six months without wholesale revision. Your rights are written in half steps that one of the two current scales does not have.
Preprints
Fully permitted, and it costs you the borderline
Permitted without qualification, and the tie-break is tilted against it anyway
ARR states there is no anonymity period and no limit on posting non-anonymous preprints while the work is under review. Its senior area chair guidelines then say that for borderline papers the ACL policy is that papers not preprinted have an advantage, and that SACs are asked to go looking if the authors said they were planning one. The penalty bites only at the borderline, which you cannot identify in advance.
Preprint permission from the ARR author guidance and the borderline advantage from the senior area chair guidelines, read August 2026.
Desk rejection
ACL 2026 rejected over 100 papers it had already accepted
Not at review or at commitment. At camera-ready.
| Ground for desk rejection | Note |
|---|---|
| Citations to non-existent literature | A clear violation of the ACL Policy on Publication Ethics, whether human-authored or LLM-generated |
| Format or anonymity violations | Repository links must be anonymised; Google Drive and Dropbox are not acceptable, they can track who downloads |
| Dual submission, or self-plagiarism | |
| Missing limitations section | |
| Appendices not in the double column format | Since July 2025 |
| Responsible NLP checklist incorrect, incomplete or misleading | Since December 2024 |
| Resubmission policy violations | Since December 2024 |
| Related work hidden in an appendix | ARR treats it as circumventing the page count |
| Prompt injections intended to manipulate generative tools |
The camera-ready rejections are in the ACL 2026 front matter; its programme chairs found them during final consistency checks and recommend the same checks at submission in future. The desk-reject checklist is in the ARR reviewer guidelines, the other grounds in the ARR call for papers. Read August 2026. Desk rejections can happen at any time in the cycle.
Free tool, no account
Where should this paper go?
Paste the title and the abstract. It ranks the venues that publish this kind of work, with the reach and the article charge OpenAlex has on record.
Everything that matters is decided before the cycle opens
Three days after the meta-review is not when a paper improves. The version you put into the cycle is the one your reviews follow, and the venue you commit it to is worth choosing before then.
Matching needs no account. A paid review does not predict a decision: what it cannot do is written down. Before the deadline, use the pre-submission checklist.
phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.