Generative Writing

Abstract

Writing isn’t a summative process. You don’t do all the work and then “write it up”. The act of writing suggests new experiments to run, reveals holes in your argument to fill with literature or analysis, and helps you communicate with your co-authors. Because writing helps you think (perhaps painfully), it is a generative process: it generates ideas.

To take advantage of writing’s generative properties, you need to write all throughout the production of research. You need to write frequently and copiously. If you hold the mindset of writing as a generative process, you will find many opportunities to write in the natural course of your work.

What purposes does generative writing serve?

Generative writing helps you generate new ideas in two major ways. In the act of writing, you record your thinking at this point in time, document what you’ve done, and your current reaction to it. In the act of reflecting, generative writing helps you think again about what you thought before, notice how your thinking has changed, and expand on (or amend) your prior work.

Generative writing helps you publish more papers in two major ways. In the act of writing, you practice articulating your ideas into words and develop your voice as a writer. In the accumulation of writing, you build a library of documentation about your project, from your ideas about theories to the details of what you did and how.

How much do I need to write?

You need to write copiously.

Your writing doesn’t have to be beautiful, it just has to be recorded. Later, you will update your writing to reflect new ideas. Much later, you will engage in wordsmithing: increasing the flow of your writing, reordering ideas, massaging prose. You should not do that now. Right now, you just need to record what happened, what you thought about, and how you are going to proceed.

How can I get words on a page?

Imagine that you are telling your collaborator what your interim results are. Explain what you did. Explain why. Explain how you are stuck. Write down your explanations.

Virtual Andy

My friend Brian has someone he calls “Virtual Andy”. Virtual Andy is not a real person. Virtual Andy lives in Brian’s head. He is Brian’s mental model of what Brian’s careful co-author would say to him. When Brian is stuck, he has “conversations” with Virtual Andy and writes them down, his parts and Andy’s parts.

Imagine that you are telling the story of what happened in the data to your spouse. What’s going on here? What parts are interesting to you? Your spouse loves you and doesn’t mind if the story is complicated or doubles back on itself. Your spouse appreciates all the work you’re doing and makes appropriate noises from time to time. Your spouse doesn’t care about your grammar. Write the story down.

Start with a specific writing prompt, and then write for 15 minutes. You don’t have to write beautifully in those minutes, but the words need to flow from your fingers. Some of my favorite prompts are:

• I want it to be true that …
• So far, I think that …
• I am really interested in …
• (when I’m stuck) If only I knew how to …
• (when I’m stuck) I think the problem here is …
• (when I’m working on significance) I want to claim that …
• (when I’m working on significance) The thing I want everyone to know is …

Can I draw instead?

My friend Brandi Agerbeck is a huge advocate for visual thinking as a way to record and develop ideas. It’s akin to generative writing, but more diagrammatic and visually compelling. If this is a skill you already have, go ahead and use it to generate new ideas! However, if you’re looking for words on a page (e.g. for later publication), it might not be what you need.

You may find that keeping a research blog (private, or just shared among your collaborators) is fruitful because it gets you into the habit of writing informally about research. Alternately, you may want to keep a pile of (searchable) plain text files in your dropbox and interact with them via your favorite text editor. You may want to use an online lab notebook or a wiki or a git repository. You might want to use little moleskine notebooks. Find a system that works for you, and use it. Use it all the time.

Should I use generative AI for generative writing?

Generative writing helps you generate ideas. Generative AI takes prompts from you, then generates (a lot of) text. Generative AI can be a useful thought partner in some circumstances, but it can hobble your development in others.

There isn’t a simple answer to this question, so here’s a complex answer.

How do I collect my writing into a paper?

There are two basic approaches to writing a paper generatively. You can write an outline and fill it in, or you can write an argument which grows into a paper. The first way works really well when you have an idea about your argument already and you need help structuring it. The second way works really well when you are already good at structuring arguments in general, but need to work on the content of this argument in specific.

Argument first

Every paper makes an argument about something. Your paper’s argument grows organically in the process of conducting research.

The seed of your argument is your claims: what’s going on here, and why does it matter? When you engage with your data, you work on what’s going on. When you engage with the research literature, you work on why it matters. You need to engage with your data and the literature in tandem to build a strong claim: a little bit of literature suggests what to look for; a little bit of analysis suggests a new avenue to look in the literature.

As you engage back and forth with data and literature, write down your tentative claims. Link data and literature. Write questions to be answered and hypotheses to be checked. Revisit old questions and refine or answer them.

Your claims document will grow and refine itself. You will develop many claims, some of which won’t be supported in further investigations (that’s ok), and some of which will mutate in response to different data or literature (that’s great!).

Eventually, you will have so many claims that you need to decide which ones to pursue now, which to pursue later, and which to develop into each paper. It’s ok (delightful, even!) to have several papers going at once, in various stages of completion. Having lots of papers insulates you from the effects of tardy co-authors.

Outline first

Here’s a generic outline for pretty much any paper in PER:

1. Introduction: where does this work come from, and why does it matter?
2. Claims: what contribution does this paper make to our understanding of the topic?
3. Theoretical Framework: why are these claims important?
4. Context and data collection: Where did the data come from, and how did you acquire them?
5. Data analysis methods: how did you use the theory and the data to generate your claims?
6. Data and analysis: what do the data actually say? how do you interpret that?
7. Discussion: what else does this research connect to? what are the limitations?
8. Conclusion: what contribution did this work make: a more nuanced discussion than the Claims section.

You paper might rearrange these sections a bit, or add subsections in different places. That’s ok -- that level of detail in the outline happens organically as you move through the process. Generally speaking, your argument shows up in parts 2, 3, 6, and parts of 7.

Now that you have a rough outline, comb through your existing writing and paste it into the appropriate sections. Next, write the missing sections, possibly by doing one of the exercises above. Don’t worry about wordsmithing now.

You may find that your outline needs to be adjusted in the course of writing your paper. That’s fine. But, if you find that you spend more time outlining than actually writing, then you need to abandon outlining temporarily and get back to putting words on the page.

Argument and Outline together

Actually, there’s a third way: do both. This way is attractive if you’re already pretty good at writing papers and you find the structural rules of the prior ways too constrictive. I don’t recommend starting out on the middle path; try one of the others first.

The middle path looks like this:

• Sketch your argument, possibly diagrammatically, on a whiteboard or large piece of paper.
• Abstract your argument into a paper structure (possibly informed by the generic structure above).
• Notice places where your argument is lacking because you have big holes in your already written text.
• Write that text, alter the outline to fit, or improve your argument.

Remember, you’re not wordsmithing right now. Don’t worry about if your text is beautiful; worry that your argument is sound.

Finding your audience

Your audience will determine how much detail you need in your arguments, which parts of your argument to expand, and how to situate your work within a broader field. You can do brilliant research, but without the right framing, journals will reject it.

Early in the process, while you are still designing your study, it’s probably too soon to think about where your paper’s journal home will be. However, once you’ve drafted your methods and claims, it’s a good time to start thinking about which journals tend to publish papers like yours. Sometimes you know very quickly that your paper would be perfect for X, but more often you should think more carefully about the best journal. How do you find the best journal?

The best journal for your paper is the one where:

• the mission of the journal is best aligned with your work;
• the audience of the journal is one you want to talk to; and
• the editors and reviewers of the journal will publish your paper.

You can’t know about the last criterion until you try, so let’s leave it aside for now. Notice that none of these are “the highest-impact journal you can find” or “the journals that your department accepts as valid”. I’ve seen a lot of people continually fail to publish their work because they’re trying to publish in high impact factor journals without considering how well their work fits those journals. While they get rejection after rejection (and have to rework the same paper again and again), more sensible authors publish their work in journals where it fits well. The APS and I agree: chasing impact factors is bad practice.

You have several avenues for figuring out a journal home for a prospective paper. The best, first route is to talk to your co-authors. If together you can’t figure it out, you need to seek outside expertise.

Write a one-page prospectus of your work. In your prospectus, you should highlight your claims, outline your methods and theoretical framework, and give a sense of the scope of the study. Your goal in writing the prospectus is to convince someone you know that this paper is worth publishing somewhere, so polish it well (using the flow handout). Send your prospectus to two or three people you trust who might have done similar work and ask for their advice about a journal home. If you don’t know any one, ask your co-authors to send out the prospectus. If no one knows any one (really?), send it to someone you cite a lot. If everyone you cite is dead, you need to do a better literature review.

Small tips without a home

• File management
  • Append the date to each file you share, so that you can keep the versions straight
  • Switch to LaTeX early to avoid having to spend a lot of time reformatting.
  • Journal-specific templates. Find them. Use them.
• Content management
  • BibTeX and Zotero form a bibliographic management system. Use them.
  • Use structural headings liberally.
  • For long papers, insert a table of contents using \tableofcontents (you can take it out for submission)
• Getting fruitful feedback
  • When you send a draft to collaborators or beta-readers, tell them what they should (and should not) attend to.
  • Find beta-readers.
• Other people have written about generative writing in other ways, with tutorials.

Additional topics to consider

History

This article was first written on January 1, 2014, and last modified on July 17, 2024.