Responding to peer review

[article index] [] [@mattmight] [+mattmight] [rss]

When submitting an article for peer review, authors are often allowed to submit rebuttals to reviews before the decision to accept or reject.

Constructing these rebuttals requires delicacy.

Writing them often induces anxiety in graduate students.

The objective in a rebuttal is to convey confidence to reviewers:

We acknowledge your criticism and advice; we understand your misunderstanding; and we can fully integrate this feedback.

Having written dozens of these rebuttals and having read many more as a reviewer, I'm distilling my experience into concrete advice.

When writing a rebuttal, there are three general principles to follow: be polite, be conciliatory and be thorough.

You can exercise these through a five-step method that I outline below.

Types of rebuttals

There are two kinds of rebuttals: journal-submission rebuttals and conference-submission rebuttals.

Generally, you'll receive the option for a journal-submission rebuttal if the work is acceptable, but not in its current form.

For these rebuttals, you generally have a few weeks, and you're often expected to produce a new version of the paper that addresses reviewer concerns. The rebuttal can cite these changes.

You may also engage in multiple rounds of review with the reviewers for a journal submission, with acceptance only after all objections are satisfied.

In a conference-submission rebuttal, you may be considerably restricted: the rebuttal may be length-limited to 500-1000 words; you may have only 2-3 days to produce a response; and it may be the only chance to respond before receiving final judgment.

You can follow the same process for both journal and conference rebuttals, but tweak it to adhere to the particular constraints of each style.

General principles

However you decide to reply to a reviewer, several principles always apply:

  1. Be polite. Lashing out at a negative reviewer is not going to change their opinion.
  2. Be conciliatory. If a reviewer is wrong, don't say they're wrong. Say "there seems to be a misunderstanding." If the misunderstanding arose from poor presentation, admit it and offer to fix it.
  3. Be thorough. Don't dodge any points or refuse to answer any questions. If the reviewers found a flaw, admit the flaw and offer a fix (if you can find one).

Step 1: Indent

Once you've received your reviews, put them all in a file, and indent the whole file with >.

Suppose you get this review:

The paper is about [restatement of the abstract].

The main claimed contributions are:
1. X
2. Y
3. Z

However, I don't agree that X is novel.  
I've seen it in [Foo, 1989].

Moreover, Y is trivial.

And, Z is simply incorrect by the [some argument].

In conclusion: strong reject.

And this review:

The paper is a clever generalization of [Foo, 1989].

The main claimed contributions are:
1. X
2. Y
3. Z

But, I see the real contribution as W.

Indent everything:

> -- Reviewer 1 --

> The paper is about [restatement of the abstract].

> The main claimed contributions are:
> 1. X
> 2. Y
> 3. Z

> However, I don't agree that X is novel.  
> I've seen it in [Foo, 1989].

> Moreover, Y is trivial.

> And, Z is simply incorrect by the [some argument].

> In conclusion: strong reject.


> -- Reviewer 2 --

> The paper is a clever generalization of [Foo, 1989].
 
> The main claimed contributions are:
> 1. X
> 2. Y
> 3. Z

> But, I see the real contribution as W.

> In conclusion: strong accept.

Step 2: Reply to all points

Next, reply to every point of all reviews, minor or large, as you would in a traditional academic email:

> -- Reviewer 1 --

> The paper is about [restatement of the abstract].

The summary is accurate at a high-level, 
but misses some key details like A, B and C.

> The main claimed contributions are:
> 1. X
> 2. Y
> 3. Z

Agreed.


> However, I don't agree that X is novel.  
> I've seen it in [Foo, 1989].

While [Foo, 1989] does contain a method for doing X 
in specific cases, we have generalized the method 
for *all* cases.

Consider the following case:
 
 [motivating case]

This case is beyond the limits of [Foo, 1989], while 
it is in scope for our technique.

This is the key distinction in our work, and we discussed
this in more detail in the related work on page 11.


> Moreover, Y is trivial.

While we agree that Y is *simple* once the construction is seen,
we argue that it is not *trivial* to create this construction
in the first place.

If Y were trivial, we might have expected earlier work such as
[3] and [4] to use it, yet they did not.

Instead, they opted for a much more complicated partial solution.


> And, Z is simply incorrect by the [some argument].

It appears that there is a misunderstanding.  We realize
now that our presentation obscured some important facets of Z.

The reviewer seems to think that Z is Z', and we would agree
that Z' is incorrect.

> In conclusion: strong reject.

We respectfully disagree.


> -- Reviewer 2 --

> The paper is a clever generalization of [Foo, 1989].

This is an accurate summary.

> The main claimed contributions are:
> 1. X
> 2. Y
> 3. Z

We concur.

> But, I see the real contribution as W.

We thank the reviewer for raising the point.

We agree: W is a real contribution.

> In conclusion: strong accept.

We agree.

Step 3: Sort

If submitting to a conference, where reviewers may have to look at many rebuttals in a short period, make sure the most important points are at the top.

Sort all of your replies in order of importance, point by point.

For example:

> The paper is a clever generalization of [Foo, 1989].

This is an accurate summary.


> And, Z is simply incorrect by the [some argument].

It appears that there is a misunderstanding.  We realize
now that our presentation obscured some important facets of Z.

The reviewer seems to think that Z is Z', and we would agree
that Z' is incorrect.


> However, I don't agree that X is novel.  
> I've seen it in [Foo, 1989].

While [Foo, 1989] does contain a method for doing X 
in specific cases, we have generalized the method 
for *all* cases.

Consider the following case:
 
 [motivating case]

This case is beyond the limits of [Foo, 1989], while 
it is in scope for our technique.

This is the key distinction in our work, and we discussed
this in more detail in the related work on page 11.


> Moreover, Y is trivial.

While we agree that Y is *simple* once the construction is seen,
we argue that it is not *trivial* to create this construction
in the first place.

If Y were trivial, we might have expected earlier work such as
[3] and [4] to use it, yet they did not.

Instead, they opted for a much more complicated partial solution.


> But, I see the real contribution as W.

We thank the reviewer for raising the point.

We agree: W is a real contribution.


> The paper is about [restatement of the abstract].

The summary is accurate at a high-level, 
but misses some key details like A, B and C.


> The main claimed contributions are:
> 1. X
> 2. Y
> 3. Z

Agreed.


> The main claimed contributions are:
> 1. X
> 2. Y
> 3. Z

We concur.


> In conclusion: strong accept.

We agree.


> In conclusion: strong reject.

We respectfully disagree.

Step 4: Cut

If your reply is limited in length or if replies are redundant between reviewers, or if you simply don't need to addresss a point, trim your response, starting at the bottom (least important points) and working up.

For instance:

> The paper is a clever generalization of [Foo, 1989].

This is an accurate summary.


> And, Z is simply incorrect by the [some argument].

It appears that there is a misunderstanding.  We realize
now that our presentation obscured some important facets of Z.

The reviewer seems to think that Z is Z', and we would agree
that Z' is incorrect.


> However, I don't agree that X is novel.  
> I've seen it in [Foo, 1989].

While [Foo, 1989] does contain a method for doing X 
in specific cases, we have generalized the method 
for *all* cases.

Consider the following case:
 
 [motivating case]

This case is beyond the limits of [Foo, 1989], while 
it is in scope for our technique.

This is the key distinction in our work, and we discussed
this in more detail in the related work on page 11.


> Moreover, Y is trivial.

While we agree that Y is *simple* once the construction is seen,
we argue that it is not *trivial* to create this construction
in the first place.

If Y were trivial, we might have expected earlier work such as
[3] and [4] to use it, yet they did not.

Instead, they opted for a much more complicated partial solution.


> But, I see the real contribution as W.

We thank the reviewer for raising the point.

We agree: W is a real contribution.


> The paper is about [restatement of the abstract].

The summary is accurate at a high-level, 
but misses some key details like A, B and C.

Step 5: Polish

Now you've got a starting point by for editing.

If submitting to a conference, push the most accurate reviewer-written summary of your work to the top of your rebuttal to confirm this point of view. Comment on this summary to make it precise.

Since conferences rely on meetings to accept, other program committee members at the meeting may not have read the paper, and they probably won't read your full rebuttal either.

But, they might glance at the first couple paragraphs of the rebuttal.

Seeing an accurate summary from a fellow committee member will give them a quick sense of the work.

It won't hurt to add a "thank you" to the top of the review. Reviewing, positively or negatively, takes a lot of time.

As you edit your reply, try to place yourself in the reviewer's mind.

Would you be satisfied with your answers if you were them?

We thank the reviewers for the time and expertise 
they have invested in these reviews.


> The paper is a clever generalization of [Foo, 1989].

This is an accurate summary, and we'd like to amplify the
recognition of W as an additional contribution of the work
by reviewer 2.


We'll reply to individual points below:


> And, Z is simply incorrect by the [some argument].

It appears that there is a misunderstanding.  We realize
now that our presentation obscured some important facets of Z.

The reviewer seems to think that Z is Z', and we would agree
that Z' is incorrect.


> However, I don't agree that X is novel.  
> I've seen it in [Foo, 1989].

While [Foo, 1989] does contain a method for doing X 
in specific cases, we have generalized the method 
for *all* cases.

Consider the following case:
 
 [motivating case]

This case is beyond the limits of [Foo, 1989], while 
it is in scope for our technique.

This is the key distinction in our work, and we discussed
this in more detail in the related work on page 11.


> Moreover, Y is trivial.

While we agree that Y is *simple* once the construction is seen,
we argue that it is not *trivial* to create this construction
in the first place.

If Y were trivial, we might have expected earlier work such as
[3] and [4] to use it, yet they did not.

Instead, they opted for a much more complicated partial solution.


> But, I see the real contribution as W.

We thank the reviewer for raising the point.

We agree: W is a real contribution.


> The paper is about [restatement of the abstract].

The summary is accurate at a high-level, 
but misses some key details like A, B and C.

Step 6: Relax and revise

Once your rebuttal is in, relax.

Then, spend time incorporating the reviewer's feedback.

It will make for a better final version or a for a stronger resubmission, whatever the case may ultimately be.

Related pages