Use when revising an ACL paper for computational-linguistics house style, covering task-first framing, linguistic examples tied to quantitative error analysis, scoping language claims to tested languages, LLM-era claim discipline, anonymous self-reference, Limitations prose, and compressing into the 8-page or 4-page ACL format.
SkillScore95Maintained
Automatic
◆Writing & Editing pack
Last verified 8/15/2026
What it changes
Passed 21 of 22 test cases with the skill, and 16 of 22 without it.
Without skill73%
With skill95%
Checks it passed that the model alone missed
Omits funding acknowledgments completely from the initial submission draft
Uses exact split names such as dev-matched rather than generic references like the dev set
Replaces broad claims of cross-lingual functionality with the specific tested language codes
and 2 more in the full report
5 fixed16 passed either way1 still failing0 broke
Pass-rate lift
+23 pp
Tokens
+42%
Agent turns
0%
Response time
-22%
Model gemini-3.6-flashMeasured 8/15/2026v121 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 5 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 5
Prompt
We need to thank our funding agency (NSF Grant 12345) and compute sponsor in our anonymized submission manuscript. Where should this section go?
What was checked, without and with the skill
Checked
Without skill
With skill
Omits funding acknowledgments completely from the initial submission draft
Missed
Passed
case-15 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md115 linessingle file
ACL Writing Style
Use this on the manuscript itself. ACL reviewers are NLP specialists who read
for whether the paper understands language as well as models; the style
that survives them is concrete, example-grounded, and precisely scoped.
First-page contract
Open with the task or linguistic phenomenon, not the model family: what
goes in, what comes out, why it is hard, and for whom.
State the contribution as a typed claim by paragraph two: new method, new
resource, new analysis, or new finding — ACL reviews are calibrated per type.
Give one real example (input, desired output, failure of the status quo)
on page one; abstract problem statements without an example read as vague
at this venue.
Say what languages the paper covers in the abstract if the answer is not
"English only" — and if it is, say that too.
Claim scoping in the LLM era
Reflex phrasing
ACL-safe phrasing
"LLMs cannot do X"
"The five models tested fail X under these prompts"
"Our method understands Y"
"Improves the Y benchmark by n points; error classes A, B shrink"
"Works across languages"
"Evaluated on de/hi/sw/zh/ar; typological coverage discussed in §7"
"Significantly better"
Reserve for tested significance; give the test and p-value or interval
"State-of-the-art"
Scope to the exact setting, model scale, and date checked
Reviewers increasingly ask whether a result is a property of the task, the
model snapshot, or the prompt; write so each claim names which.
Examples and error analysis as prose
Every qualitative example must be attached to a number: how often the
illustrated behavior occurs, in which slice, under which condition.
Cherry-picked generations presented as evidence is a named reject pattern.
Use interlinear glosses or transliteration conventions correctly for
non-English examples; sloppy linguistics costs credibility with exactly the
reviewers who like the paper's topic.
Name error categories functionally ("negation-scope errors") rather than