How to Teach Writing Research Papers

    1 Context

    2 Symptoms

    3 Diagnosis

    4 Proposal

    5 Analysis

    6 Finding a Voice

    7 Conclusion

1 Context

This article is about teaching research students how to write research papers. To make it concrete, I’ve written it from the perspective of the species that is most subject to writing advice: PhD students. However, it should generalize to anyone on whom you spend similar amounts of time. Also, it’s written from the perspective of computer science; certain aspects may not apply to other disciplines (I’ll let you find the embedded assumptions).

2 Symptoms

You’re a PhD advisor trying to teach your student to write. You spend hours marking up their drafts. (Perhaps like this well-intended professor who really cares for their students.) Hours! And yet your student barely seems to get any better. Perhaps this student really can’t write?

The student has the dual perspective. They spend days writing a draft and send it to you. It comes back “more red then black”. They aren’t sure where to start, how to proceed, or when to stop. They sense there’s something they should be doing more than superficially correcting every red mark, but what is it? They conclude, “perhaps this person really can’t teach writing”.

I’m going to direct this article to the advisor, because they’re the ones who control the process, and they’re the ones best placed to fix it.

3 Diagnosis

The problem is this.

What you’re trying to do is to train a classifier. You need your student to learn good writing from bad, at least as you define it. You’re providing lots of input to that classifier. But your student may actually just be getting overwhelmed. You need to
  • Reduce the cognitive load being imposed them.

  • Reduce their sense of being overwhelmed.

  • Use your instincts as a teacher (assuming you have good ones!), not as an author or editor or any other role.

4 Proposal

Here’s something that has worked really well for me. My student writes, and we co-edit. We do one paragraph or at most two. We go over every sentence, even every word.

For earlier stage students, I usually say, “here’s the way I would change and here’s why”. I wait for them to digest that, perhaps asking a question to make sure they’re building the right model. The critical thing is that we do this live. (This way, they also see that I change and revise and sometimes even revert. I’m not perfect either!)

Once they have some experience, I flip it around. I point to a sentence or even word and say, “What do you think I’m going to say about this?” They conjecture; I either confirm or correct. And I explain why I have that opinion. See, classifier.

How long does this take? Longer than you might expect! Each sentence can take several minutes, so one paragraph can take over half an hour. The reason for limiting this to 1–2 paragraphs is to bound the time of the whole activity to 45–60 minutes. Beyond that, everyone is exhausted.

5 Analysis

I think there are several benefits to this approach. (Yes, it’s a bit like pair-programming, but I don’t want to stretch that analogy too far because it’s not the same thing.)

  • It bounds the amount of time they spend. Instead of spending hours trying to guess my intent, they know that the session will last at most an hour.

  • It bounds the amount of time I spend. I don’t have to spend hours and then feel like I have nothing to show for it.

  • It saves me from doing a bunch of markup with mediocre tools. Nothing is quite as effective as just pointing and editing. We might even learn editing tricks from each other!

  • It’s interactive. The student can ask questions.

  • It models behavior. The student may pick up useful secondary hints from it.

For instance, I keep a todo list and/or glossary at the top of every file. They can see how I modify that as I’m writing (because they can see me live-editing). It’s much easier to learn from a concrete demonstration than from an abstract statement to “have a glossary”.

With modern technology, you have a choice of how to do this. You can do it in person, or you can do it on Zoom. The benefit of doing it on Zoom is that you can have your student record the interaction, so that they can re-watch any parts that they were confused about or have forgotten. Or, of course, you can combine the two: do it in-person but record it anyway.

Technology aside, the most critical impacts are that students are not overwhelmed. It’s depressing to send a professor a 10 page paper and get 10 pages of “more red than black”. The cognitive load is far too high. The brain shuts down.

Yes, working one paragraph at a time is slow progress. You’re better off doing this a while before a deadline rather than at the very end.

But consider the following:
  • For a lot of advisors, your current system isn’t quite working for you anyway.

  • If you’re using \(N\) hours to comment, I claim you’ll get a two- to ten-fold benefit from doing \(N\) one-hour live sessions instead.

  • It’s an S-curve phenomenon. After a while you really see effects blow up (in a good way).

The key thing for this to succeed is that students have to pick up patterns. That is, if you point to something in this paragraph, the same phenomenon may occur three paragraphs later. When you did the single-pass edit through the whole paper, you marked all those instances. The good news is, you gave them an opportunity to notice and generalize. The bad news is, you may have left them so overwhelmed that they didn’t notice and didn’t connect and hence never generalized.

There’s a large body of literature on learning from examples versus learning from rules. I won’t go into it here, in part because not a lot of it is in the context of an advanced learner. Suffices it to say you want to give students both examples and rules. That is, when you make a specific edit, there are many lines that can pass through that point; also give them a more general rule that shrinks the number of lines, thereby improving their classifier.

6 Finding a Voice

You’ll note that above I wrote,

You need your student to learn good writing from bad, at least as you define it.

That is, what you’re really doing is teaching a student to write in your voice. Voice manifests at many levels, from the overall paper structure to how one words individual sentences (and may include uses of the passive and active voise).

A voice is something every PhD advisor imparts onto their students. Good or bad, they’re implicitly conveying it; I’m just making this explicit. (You may now be thinking of Nipper…)

However, this is not necessarily the student’s natural voice! Certainly, my PhD advisor and I have extremely different voices. So what to do about this?

I say: it’s okay. Early on, I want students to cultivate something similar to my voice. The focus should be not on individualism but on first getting good habits under the belt. After all, we’re writing papers together, and I may have a better sense of what voice will get papers accepted and read. As a student matures, their own voice starts to come out—indeed, that is one of the signs of maturity. By the time they graduate, they’re hopefully comfortable writing papers largely by themselves, and these will naturally gravitate towards their preferred voice. But until they get there, having at least one successful style under their belt will stand them in good stead.

7 Conclusion

If you try this out, let me know how it goes! I’d love suggestions on how to improve this, including feedback on why it doesn’t work at all.