How to make a paper stronger?

After reading this question and David's answer I thought a more general question asking for tips would be useful. So

how can authors make (and write) a paper stronger?

• stronger is a loaded term. I can make a paper stronger by piling on the results. that won't necessarily make it easier to read, or even a better piece of research. – Suresh Venkat May 29 '11 at 9:20
• This question seems hopelessly vague to me. I think you should at least specify what is given and what are the variables here, even if you cannot specify what is the objective function? (For instance, have you got a certain fixed set of results that you want to publish at a certain conference?) – Jukka Suomela May 29 '11 at 13:54
• For general paper quality, it might be useful to have a look at Alan Jay Smith's paper, “The Task of the Referee”: cs.utexas.edu/users/mckinley/notes/reviewing-smith.pdf – Magnus Lie Hetland May 29 '11 at 19:13
• @Jukka, I don't agree, it might be general but I don't think it is vague. It is not a formal question where I can specify the variables. I could say something similar about the conference question, what are the given and variables in that question? – Kaveh May 30 '11 at 22:55
• @Tsuyoshi, it is not, it is not difficult to make up question if I was posting for the sake of asking a question if I wanted to. I have explained why I have asked the question and I sincerely think the tips can be useful for me and also for others. (I was first going to ask the OP of the previous question to make the question more general but thought that posting a separate question might be better.) Are you suggesting that I am not honest about the motivation? – Kaveh May 30 '11 at 22:58

This is sort of an unserious answer in that it's not what I consciously do most of the time, but:

Think about what the next paper after yours might be — not the next paper that takes some small follow-on problem from your paper and solves that too, but the next paper that takes it to another level, finds tight upper and lower bounds, shows that it's an instance of a more general phenomenon, and generally makes your paper obsolete. Then write about the results in that paper instead of the ones you already have.

Illustrations.

• (Had to add a figure to pad the length of the post. Unfortunately, with conference papers, it is usually the other way around.) – Jukka Suomela May 29 '11 at 20:23

• And there's the T-shirt banner ! – Suresh Venkat May 29 '11 at 19:37
1. Improving the writing always helps.
2. Better, more intuitive examples. Are the ones you first thought of the best ones?
3. Adding/improving empirical validation, including an implementation with performance evaluation or user studies, if applicable. On the other hand, if the paper is empirical in nature, strengthening the theory will improve the story. (Maybe this isn't so relevant for TCS papers.)
4. Ensure that your notation is consistent, both internally and with the "standards" of the community.
• +1 for examples and +1 for consistent notations It always is a pain to have the same symbol used for two distinct meanings in the same paper (e.g. permutation $\sigma$ and alphabet size $\sigma$ later), or to have two symbols introduced for the same concept (e.g. one in previous work and one in real work), when one has only a limited, too short, time to spend on a paper. – Jeremy May 30 '11 at 14:34

Concerning technical writing:

Check spell your paper, review your notations (even though an index of notation cannot be included in a conference submission in general, doing it early helps for later, longer versions, and for the clarity of the early versions), have someone else than those involved in the writing to read it and check it is understandable.

Concerning content: Justify all your choices.

When you have several related results, find what they have in common, describe it as a question, and explore alternative answers to this questions that you did not consider before, eventually adding new (minor?) results or justification for discarding alternate answer.

When you have a single result, a single answer to one specific problem, describe what are the other possible answers to this problem, in the literature or based on futuristic results.

Do not limit your paper to "I did this, and look, it works!". Explain why it works and why or how other choices would work worse or would not work.

• As for getting someone else to read your paper: always do this. Once they're done, never ask them if the paper was understandable: no one will admit to not grasping your paper. Rather, ask them to recount the major points of the paper in some detail. – funkstar Jun 3 '11 at 20:26