I have come to see a comment on twitter: "If you define “good ML paper” as “ML paper accepted at NeurIPS/ICML/etc” then writing a good ML paper is probably much more about how skilled you are at storytelling and paper writing than about ML." This comment is about Machine Learning (ML) and the comment is someone's opinion. My question is more general, how to improve the writing skill for scientific papers?

It would be great if someone can answer in Machine learning and NeurIPS/ICML perspective.


Some personal perspectives to advise you (or others who read this question later). Please don't take the writing of this post as perfect as it is hard to proof things in comment box. Do as I say not as I do! (That said, I have gotten 6 first authored papers, all of my academic papers, published without any revision at ACS specialty journals...so there is some "I have been there".)

  1. Most of the skill (or problems depending on your point of view) is applicable across writing in general or at least science/engineering/math writing in general. You are savvy to look for ML specific issues, but 90%+ of how to write a good paper would be the same in physics, chemistry, statistics, medicine, or even business.

  2. I agree with the Buffy comments on "practice".

  3. One slight note for anyone who is non-native speaker is to work on your English. Your (personal) English comes across as quite good (you may be a native speaker). But I am really addressing others who may read this who are outside the Anglosphere, publishing in it, or who are foreign grad students. English is the "lingua franca" (sorry France) of academia. Don't let that hold you back. But plunge in, in all of your daily life, into use of English (languages get better from use...be the machine learner that just jumps in, not hesitant.) That said, when submitting papers, first do your absolute best (not slack), than have an English speaker review. It is VERY common for papers from certain countries to have poor English. This is frustrating for reviewers since the content itself is difficult AND may have science flaws. Weeding through that along with Chi...bad English is a hassle.

  4. Feedback: Agree with the Buff comment. But would just add you will get differing opinions at time. And experience differing levels of insight from people. Use your noodle to differentiate when there are conflicts. 90%+ of the time, advice will be similar, especially on basic topics. But there will be differences. Don't let it fluster you.

  5. You are right to want to write well. I had very middle of the road results in noteworthiness, but I wrote clearly. And my papers sailed through. Make it easy. Let this be a life lesson. Same thing applies in the work world.

  6. References: I don't know the "how to write a paper" books. I'm sure they are mostly OK. But I will share a few gems that helped me. Google them.

    a. Katzoff Clarity in Technical Reporting. Short NASA pamphlet. Underground gem that I encountered at Langley. Very practical and researcher to researcher in tone. Free on the web. In particular, read his section on "honesty".

    b. E. Bright Wilson Introduction to Scientific Research. Great advice on writing up incomplete work. It is not sleazy. It is your duty! Somebody paid for that stuff. Don't be a tree in the forest. [Ignore the issue that this book is from the 50s. The lessons are universal. And you should learn to be able to extract lessons in old books, not put off by anachronisms. At least the writing is not hard to read...just it will hurt you when he talks about ENIAC or the like.]

    c. Strunk and White Elements of Style. Everyone recommends this. To the extent that I'm sure there are reactions against it. But it gets recommended for good reasons. Short, sweet. Lots of practical word choice advice. And "omit needless words" is a great philosophy. Very cheap...buy a copy. Get exposed.

    d. Barbara Minto Pyramid Principle. There are a couple versions but they are very similar. Ignore where she tries to say she invented something or plays on the McKinsey association. But pay attention to the advice on lists and hierarchical structures.

    e. Little Red Schoolhouse (writing advice). This one may be hard to find...sorry don't have copy to hand...typing in a ba...cafe. And I got my copy by Xoxing one from a Uni Chicago B school student. But it is a gem, gem, GEM. Key point is to try not to use nominalizations (Google this word) as subjects and use clear actors instead. "Bartenders measure how much booze to pour carefully." Not "The practice of rigorous mensuration...". Capisce the difference?

    f. Warriner (Harcourt Brace) English Grammar and Composition. Just a simple reference for when to put a comma in or not (and such like). Very user friendly. Written for high school students. Easier than more complicated style guides or grammar texts.

  7. Practical advice (may duplicate some from references):

    a. Avoid long paragraphs. This shi...stuff is hard enough. Don't write monograph style walls of text. Keep the paras under 150 words. Seven sentences max (five if the sentences are very complex/long). This is not some mathematical "rule" and I'm sure someone will disagree in comments. But it is practical advice.

    Ideally the paras should have unified topics. But if it gets too long in one para find/devise/create a reason to break into two (or three!) paras.

    b. Separate chemical or math formulas with line breaks and centering. They are too hard to process in line in normal text.

    c. Figure captions are some of the most important text in a paper. I am a fan of longer captions. Hone them and don't use needless words. But tell as much of the story as you can in captions. People read graphs and captions MUCH more than rest of the paper. Captions are PRIME REAL ESTATE.

    d. References: Check every citation personally. If it is a journal article make a photocopy and keep it on file (helps for your research itself). It is very frustrating to look for a reference and find it is wrong. Or even that it doesn't pertain (this can happen when people copy errors forward in a lab group).

    e. Journal instructions: Get a copy of the notice to authors (usually in first calendar year issue of the journal). Look at every instruction and check it personally. Like a machine. Like a nuclear technician. You would be amazed how many people don't do this...and how much it helps make a paper slide by easier.

    For references, use whatever scheme is particular for the journal (they are all different). If you use a bibliography program, check the results personally. Errors can happen. In the grand scheme of things...it doesn't matter and/or hopefully the copy editor finds it. That said, it is something under your control. Get it done. That way people can concentrate on the science, not the format (since you are so gnat's ass perfect in format!)

    f. Use spell check. And when you make any edits use it again! This sounds silly. But you would be amazed how often people neglect it (even in the work world). Have a PAPER dictionary next to you to check on words that spellcheck highlights falsely. (Merriam Webster collegiate, hardbound, is a good choice). Of course use the net to check words still of concern (especially proper nouns) in addition.

    g. Getting traction. I had a quite shrewd professor advise me to write the figures first and the abstract last. This is sound advice in a fashion but I realized it is WRONG. Write in whatever fashion gets you MOVING. In the modern world of computers, it is very easy to edit later (as opposed to when people had typists). My advise is to put the title of the paper down and the major sections (with NO CONTENT) and then hit "save". Guess what!? You are started now. Then write whatever comes easiest. If "abstract" is easier to write now, write it now. You can revise it later. I have even written up most of a paper and then just had a couple gaps. And the momentum and such moved me to do the final regression (or even the final physical experiment!)

Good luck. Kick ass. Have fun. "No prisoners!" https://www.youtube.com/watch?v=1m7cSALYOBM


As with most skills, both physical and intellectual, the key is practice and feedback. To be a better swimmer, swim a lot. To excel at it, get a coach who can give you feedback.

The same is true of writing and many other things. To be a better writer, simply, write. But the feedback is also important so find a way to get that. You will get it from referees when you submit papers, but you can also get it less formally if you share drafts among colleagues. If you are a student, you probably get feedback from professors, but ask them to comment on your writing specifically, not just your ideas.

There are, of course, books and such to help you be a better writer. Their advice may be helpful: Keep the reader in mind; Be correct but not pedantic; the opening should entice the reader to continue...

But the books won't make you a writer (or a swimmer). Just write.

And get feedback.

  • 1
    Thanks. I got a paper rejected last year November and reviewer said the paper is poorly written (the venue is prestigious and I think my writing is still not up to the standard). That's all I got from the reviewer about writing, certainly, that is not very helpful feedback. If I was told why it is poorly written then it would make more sense to me and would help me improve writing.
    – ffl
    Jan 7 '19 at 11:07
  • 1
    My advisor was happy with the writing. That is why I was confused about that comment. I do understand that paper reviewing is a labour intensive and time-consuming task. I am not complaining about the review. I just want to get more robust feedback that will help me improve further.
    – ffl
    Jan 7 '19 at 11:21
  • 2
    Then you need to consider if your advisor's writing is up to scratch... There are several academics that I know, who have English as a second language and struggle to produce linguistically polished papers, while technically they are beyond reproach... So, is this question really about your rejected paper then?
    – Solar Mike
    Jan 7 '19 at 11:25
  • 1
    @SolarMike, this question is more about improving my writing. English is my second language and working every day to improve the writing skill. Getting expert advice always helps.
    – ffl
    Jan 7 '19 at 11:31
  • 2
    @FarhanaLiza Best suggestion I was given was to read books from those who write well... And one example has always stuck in my mind, from a book about writing technical reports, which compared a paragraph written by Winston Churchill and a paragraph written by a person in the Tax Office... One was so clear, with no wasted words and the other...
    – Solar Mike
    Jan 7 '19 at 11:56

Not the answer you're looking for? Browse other questions tagged or ask your own question.