6

What your preferred approach to writing a paper with a word count limit? The two obvious approach are to

  1. Try to write so that you don't exceed the word limit in the first place
  2. write however you want, then shorten.

I would assume not worrying to much about the length and then shortening is easiest.

Regarding shortening: How do you go about it? Any specific suggestions on the use of LLMs?

9
  • 4
    another approach is just send the works to these journals that are possible to be explained appropriately within the page limits inherently, so select the journals based on all features of the work not just the title. A work that needs to be explained with a lot of details and experimental results, is not suitable to be shortened and sent as a short report, unless you want to report about it! (the same is true for the reverse situation)
    – m123
    Apr 11 at 10:31
  • 14
    I'd say the first step comes before you start really writing, in that you choose a scope for the paper that can fit.
    – Peter Flom
    Apr 11 at 11:13
  • 2
    @PeterFlom Or you write the paper, make it as concise as you can consistent with getting the point across, and then choose a journal that will accommodate it. Apr 11 at 11:52
  • 3
    Write around 20%-40% too much and shorten. The paper will be much better. Apr 11 at 23:50
  • 1
    Find a different journal.
    – Matsmath
    Apr 12 at 2:23

7 Answers 7

14

First regarding using LLM, it is increasingly becoming a commonplace to use AI to generate text for the papers. I think this practice should be discouraged not only by the editors/reviewers/publishers but also by the authors themselves (self accountability). Ironically speaking, there are so many cases reported on the LinkedIn showing the papers containing the unedited text from AI. The academic world is just getting weirder.

Regarding your question related to word count, it not always fixed word length but an estimated word length especially when it comes to long format publications such as journals, patents, and so forth. Yes the best bet should be write completely and then revise it considering the word count. You can also check other papers that have already been published in the same journal if they follow the word count guidelines strictly.

13
  • 22
    @RandomName42 Having an LLM revise or rewrite rather than starting from scratch does not prevent it from hallucinating errors and changing the meaning of your writing. I would not advise it for anything important.
    – Bryan Krause
    Apr 11 at 13:30
  • 3
    This is a Problem only if I don't understand the LLM's output enough to correct it or if I just blindly accept it, which should obviously not be the case. Apr 12 at 6:04
  • 5
    @cbeleitesunhappywithSX Strongly disagree. The LLM can't know what the human author does and doesn't know/understand. The human author can. Revising is way more work than initially writing. Don't outsource your intellect to something that's provably not intelligent.
    – Bryan Krause
    Apr 12 at 12:21
  • 7
    @ChellCPlus No, you're not expected to do that, so don't do it. Especially don't expect an LLM to "upscale" your understandable writing to something opaque. Terrible idea.
    – Bryan Krause
    Apr 12 at 12:23
  • 2
    @RandomName42 Those checks are more work than the initial effort would be if they are done correctly; I'm suspicious that they ever are. In any case, by using an LLM you're letting statistics about what is found in writing generally to make decisions about what is and isn't important, with a model based on an old corpus and not actually intelligent in the first place. Do you really want that thing making decisions about what is and isn't important to your novel research publication? Writing isn't some game, it's a critical piece of the research process.
    – Bryan Krause
    Apr 12 at 12:26
11

How do you go about it?

Cut out a lot of discussion, replace all convoltued forms ("it is possbile" with "may") ...

Any specific suggestions on the use of LLMs?

Yes: check carefully if and where the text you give is saved. You may be breaking confidentiality, in the not so far future you may be exposing yourself to plagiarism (what if the LLM you use is sentient and submit the publication before you?)

4
  • 5
    I like your scenario. An LLM with a scientific ego. Apr 11 at 23:50
  • +1 for checking carefully. However, preparing a publication means that the risk of breaking confidentiality is low since the text is anyways meant to be public. (IMHO generating texts for internal use is a very different scenario in that respect. And yes, there may accidentally be confidential things in drafts that are removed before publication - but then the baseline to compare to would be the risk of accidentally exposing confidential information and not realiziing this before publication). Apr 12 at 10:57
  • 1
    Wrt LLM submitting faster: I guess for now the risk would be rather the LLM generating text for someone else who then publishes faster. Anyways, writing e.g. with version control means that there are time stamps which can be used to contest such plagiarism. (And in turn, one should maybe be careful when the LLM expresses good ideas that were not in the input...) Apr 12 at 11:00
  • 1
    My dad taught me a good rule: delete the second sentence in every paragraph. Sounds weird, but it's amazing how often it works. If you don't believe it, try it on some of the answers to your question. Apr 23 at 17:14
8

I would recommend you do both: write a first draft concisely, then revise to shorten if necessary. It is good practice to be short and to the point (step 1) even if you do not have a word limit. Your readers and reviewers will thank you for it.

This is not easy, especially at first. You will get better with experience.

I have not tried using LLMs to revise text. I agree that this is a task they may be good at. However, I would recommend you revise the result thoroughly; LLMs do have a tendency to hallucinate, and I would be very careful as to whether their shortened text still conveys correct information and kept the important bits.

7

The only correct answer to this is, "Do what works for you," because there's no reason that anyone else's work process should align with how your mind works.

But in the spirit of helpful examples:

  1. Outline
  2. Write at length
  3. Trim
  4. Trim
  5. Trim
  6. Polish
  7. Trim

My natural writing style is expansive and verbose. The only way I can work my way around to 'terse' is by writing what I want to write and hacking away at it.

The same applies to page count maxima, which are arguably even worse. I've never been so tempted to throw my laptop out the window, as the day I added a word-- a short word-- to a paper draft on Page 1, and the various rules in the venue's mandatory Word template turned that into two or three extra lines on Page N+1 of an N-page-max paper.

3
  • 2
    You missed a few steps. 8. Polish. 9. Polish. 10. Trim. 11. Add a section of explanation. 12. Trim. 13. Trim. 14. Curse the Writing Gods. 15. Polish. 16. Remove the section added in step 11. 17. Revise. 18. Trim. 19. Polish. &c. Apr 12 at 13:12
  • 7
    @XanderHenderson I had to cut those to maintain the word count.
    – Anonymous
    Apr 12 at 13:44
  • 2
    +1 because I came here to give the same answer and the opposite example :D . My writing style (at least in academic wiriting) was rather dense, my first draft of a paper was usually below the page/word limit, so what I did was 1. Outline, 2. write, 3. expand explanations here and there, 4. polish, 5. find that one word that makes LaTeX wrap every paragraph so oddly that the bibliography ends up beyond the page limit, 6. polish, 7. upload a few hours before the submission deadline, even though the paper feels far from perfect.
    – Sabine
    Apr 12 at 15:50
5

I tackle this by thinking in units much larger than words. For example, in paragraphs, or pages, or sections. Take a document you've done before, and see how many words the paragraphs tend to be, just roughly. Many of mine are 200 words, so if I have a "budget" of 1000 words, that's 5 paragraphs. You can't express 20 ideas in 5 paragraphs, right? This sets the scope of what you will write.

If you have 5000 words, they will probably be introduction, thing, thing, thing, conclusion, each about 1000 words. Or of course you may have required sections like Summary, Background, Method, etc. What is a long one of those for you? What is a short one?

This will let you get the scope right. You won't find yourself needing to cut a thing in half. Then you may find that you need to chop 10% or add 10%, and that can be done by removing a sentence here and there, or perhaps shortening some wordy phrases and dropping "filler" like "with this in mind, we proceeded to perform the operations required to complete the" . This sort of work can be very frustrating, so don't start until you've got roughly the right number of sections, pages, and paragraphs. Only then, refine the word count.

2

Several other answers make excellent points already, so I won’t repeat them, but to add one point they don’t cover: Outline extra-carefully, with the word/page count in mind. (I’ll use page limits here since that’s more common in my field.) This also has a lot in common with planning slides for a talk, where you have a fixed time limit which (for planning) can often roughly be translated to numbers of slides.

For a tight page count, I’d start by making an outline including estimated page counts, something like

  • Title, abstract: c 1/2 page
  • Introduction: 2–3 pages
  • Technical background: 1–2 pages
  • Definitions + basic properties of reticulated widgets: 2–3 pages
  • Main proof of frobnicating theorem: 1–2 pages

so I can make sure my outline looks well-balanced and feasible. Then I start drafting based on that outline; as I draft, I keep the outline in mind, and both shape the draft to fit the outline, and adjust the outline as necessary to fit the growing draft — when (inevitably) some sections need more space than I’d expected, and (hopefully) others turn out to work more concisely. If too many sections are demanding more space, and all are feeling as concise as I can get them, then that’s the time to cut more concisely.

What I do not recommend (at least, it doesn’t work for me) is just writing first without thinking about the word/page/time limit, and then trying to cut/compress it afterwards. One of the most frustrating writing tasks I’ve ever attempted was cutting a c.24-page draft I was very happy with down to 20 pages — no paragraph could be cut without weakening the whole structure. In the end I think we just submitted as 24 pages and trusted to the journal’s generosity. Different space constraints are often best suited to a different outline and selection of material entirely — e.g. a longer version gives space to present technical details; in a short version, rather than presenting them unreadably densely, it may be better to cut them entirely and set up a black-box abstraction of what they provide, which can then push towards a different organisation of the surrounding material. So keeping the length constraints in mind from the start helps to find an exposition well-suited to them, and avoid too much backtracking later.

0

I would recommend a mixture of both methods, and add a third component:

(3) Prioritize the parts of the argument that are really important. Not everything in your work must go into the paper. It's always possible to put an extended version on arxiv or similar.

I can't stress that last point enough. I find it not uncommon when reviewing papers to encounter the case where the authors have worked hard to cram everything they found out about their scientific question into the page limit of the conference/journal they submitted to, only to end up with a paper where half of the results were weak or partly wrong or at least insufficiently supported by evidence or not fully worked out.

These hurt their chances of getting accepted, even if the core contribution is good. I would in those cases much rather have reviewed a 15-page paper containing only the core contributions and referring the interested reader to the 42-page extended version, than the 30-page thing the authors submitted because they were afraid to leave something out, and maybe that the short version would insufficiently dazzle the reviewers.

Not the answer you're looking for? Browse other questions tagged .