Publishing in English when English isn't your first language
The short answer
Language rarely costs you a rejection on its own. It costs you the benefit of the doubt, and the benefit of the doubt is most of what a borderline decision is made of.
An editor and two or three reviewers are deciding, under time pressure, whether your work is what you say it is. Every sentence they have to read twice spends some of the attention they were going to give your argument. That is the whole mechanism. It is not about accent, and it is not about the occasional missing article. It is about how much effort your paper asks for before it starts paying out.
So this post is not about how to do research or how to pick a journal. It is about the four places where the language of a submission changes the decision: how you pitch a claim, what you write in the cover letter, how you read a reviewer's polite phrasing, and how you answer it.
Where the language actually costs you
Three specific places, in order of how much they cost.
The claim. Reviewers score contribution. If your central claim is stated so cautiously that they cannot find it, they will score what they found, which is less than you did.
The abstract and the first page. This is where the reader decides what kind of paper this is. A muddled opening makes an expert read defensively for the rest of the paper, and defensive reading finds more problems.
The rebuttal. The single highest-leverage piece of writing in the process, and the one nobody is taught. A good rebuttal turns a major revision into an acceptance. A bad one turns it into a rejection you did not need to receive.
Notice that none of these is grammar. You can write flawless sentences and still lose all three.
Hedging and claiming
The most common problem I see in second-language academic writing is not overclaiming. It is hedging in the wrong place, and hedging in layers.
English academic register has a narrow band. Too strong and a reviewer starts hunting for the counterexample. Too soft and there is nothing to score. Getting the band right is a learnable skill, and it is mostly a matter of hedging the interpretation while stating the measurement plainly.
Nine pairs. The pattern matters more than the specific words.
1. Solving versus reducing
Before: Our method solves the problem of domain shift. After: Our method reduces the accuracy drop under domain shift on the three datasets we tested.
"Solves" invites the reviewer to find one case where it does not. The bounded version is both truer and much harder to attack.
2. Stacked hedges
Before: It could perhaps be suggested that these results may indicate a possible improvement. After: The results show a consistent improvement across all five runs.
One hedge is a qualifier. Four is an apology, and it reads as a finding the authors do not believe.
3. Hedge the interpretation, not the data
Before: The mechanism is probably responsible for the effect, which we measured precisely. After: We measured the effect precisely. The mechanism we propose for it remains to be confirmed.
This is the one that costs people most. Writers nervous about their English hedge the entire sentence, which makes solid measurements sound uncertain. Split them: the number is the number, the story about the number is where the hedge belongs.
4. Common knowledge
Before: It is well known that X. After: X has been reported in three independent cohorts [refs].
"Well known" with no citation is the phrase reviewers circle. If it is well known, cite it. If you cannot find the citation, it is not.
5. The contribution sentence
Before: In this paper, we present a novel approach to Y. After: We show that Y can be done without the calibration step that previous approaches require.
Reviewers score the delta, not the adjective. Name what is different, and "novel" becomes unnecessary.
6. Limitations
Before: A limitation of this study is the small sample size. After: We tested 24 samples. That supports the direction of the effect but not its magnitude, which a larger cohort would be needed to fix.
A bare confession invites the reviewer to decide how much it undermines. Saying what the limitation does and does not touch keeps that judgement in your hands.
7. Impact
Before: This work will have a significant impact on the field. After: This removes the manual calibration step that currently limits throughput in clinical labs.
"Significant impact" cannot be checked, so it cannot be credited. A specific consequence can be both.
8. Previous work
Before: Previous methods failed to address this problem. After: Previous methods address the static case; the moving case has not been treated.
"Failed" reads as dismissive, it is usually inaccurate, and one of those authors may be reviewing your paper. The neutral version also states your contribution more clearly.
9. Causal verbs
Before: The treatment improved recovery. After: Recovery times were shorter in the treated group.
Unless the design supports causation, the causal verb is a claim the reviewer will reject on its own. Match the verb to the design.
Read your own abstract looking only for these nine patterns. Most drafts have four or five.
The cover letter
It is short, it is read by the editor and not the reviewers, and it answers one question: why this journal.
Three or four sentences is enough. What you found, in one sentence a non-specialist editor can follow. Why it fits this journal's scope, in one, ideally naming the kind of paper they publish rather than flattering the impact factor. Any declaration the journal requires. That is all.
Two things to leave out. Do not summarise the paper, because the abstract is attached. And do not apologise for your English. It plants the idea before anyone has read a sentence, and editors do not desk-reject for language when the science is in scope. If you want the paper read for its content, do not hand over a frame that says otherwise.
Reading a reviewer's real objection
Peer review is written in a politeness register that second-language readers routinely take at face value. The comment sounds mild, so it gets a mild response, and the editor sees an author who did not address the point.
A rough translation table:
"The authors may wish to consider..." means do this, or explain clearly why you did not. It is not optional.
"It is not entirely clear how..." means I could not follow it. Adding one clarifying sentence rarely fixes this; the passage usually needs rewriting.
"The novelty is somewhat limited." is a framing failure more often than a science failure. The reviewer did not find your delta. Go back to pair 5.
"The English could be improved." sometimes means exactly that, and sometimes means the reader had to work too hard. Ask the editor which sections were unclear, and treat structure as the first suspect.
"I would encourage the authors to temper their claims." means one specific sentence overreached. Find it, fix that one, and do not water down the whole paper in response.
And a pattern rather than a phrase: a list of twelve comments each labelled "minor" is not minor in aggregate. It usually means the reviewer found the reading effortful throughout.
The rebuttal
The structure that works is boring and mechanical, which is the point.
Open with two sentences: one thanking the reviewers without ceremony, one stating what substantially changed. Editors read this part.
Then number your responses to match theirs exactly. For each: quote the comment, state what you changed, and point to where. "Added, Section 3.2, paragraph 2" is worth more than a paragraph of explanation.
Concede precisely. Concede the point that was made, not the paper. There is a habit, more common in second-language writers, of accepting a criticism more broadly than it was offered, which quietly downgrades the contribution in the editor's mind.
Where you disagree, separate the observation from the inference. Agree with what they saw, disagree with what they concluded, and give the evidence. You are right that the effect is smaller in cohort B. We do not think that indicates instability, because the difference tracks the known age distribution, shown in the new Figure 4. That is a disagreement an editor can adjudicate.
Never argue about tone, never write "as we already said in the paper" even when it is true, and answer every point including the ones you consider wrong. Silence reads as agreement plus non-compliance.
What actually helps
Reading in your field helps, but slowly, and it teaches recognition rather than production. What builds the register faster is producing it under mild pressure and getting a reaction: write the claim, have it pushed back on, revise it, repeat. The nine pairs above are not facts to remember. They are patterns you need to be able to reach for while writing badly at eleven at night, which means having produced them yourself many times.
That is the gap I built nicheAce: language for. You describe the field you publish in, and it generates a fifteen-week journey through that field's own language: five exercises a day of about fifteen minutes, five days a week, ending each day in a role-play with a character who has their own goal, so an academic journey means practising the rebuttal and the defence against someone who does not agree with you yet.
The honest limits: about seventy-five minutes a day for fifteen weeks, Android in the EU today, and hands-free Audio Mode coming rather than shipped. It does not proofread your manuscript, and it is not a substitute for a colleague who knows your subfield reading your draft.
The part worth remembering
The reviewers are not marking your English. They are deciding, quickly, whether to believe you. Everything above is about making that decision cheap for them: one hedge instead of four, a delta instead of "novel", a rebuttal numbered to match their comments.
If you want the longer background on why the whole system runs in English in the first place, and what that costs the people it was not built for, that is here. If your next deadline is a funding proposal rather than a paper, the same register question shows up in a different form: what Horizon Europe actually requires of your English.