It is easy to make the mistake that what is obvious to you (the author) is obvious to everyone. It is also easy to just describe what you did, but neglect to say why. You'd be amazed how many papers get rejected just because of that, even if the research is otherwise perfectly good. Writing a paper is actually surprisingly simple. There is a small set of questions you need to address; if you do that, the readers will get what they want. Like this: The above list is a useful guideline at three levels: First, it is a detailed outline of the introduction for your paper. You can follow it pretty much as it is. Second, it also serves as a guideline for the abstract, although you have to be much more brief in the abstract. Third, it serves as the list of topics you will discuss in the later sections.


A typical paper consists of the following sections: The abstract is intended to be a short summary of the paper that a person can read and understand without reading the whole paper. It should follow the same structure as the intro (see the first list above). Do not forget to say how the results turned out and what they mean, i.e. what new knowledge you have established with the paper. Be specific and give content, not just a list of the main sections.

Each of sections 2-5 should do the following:

In Discussion and Future Work you usually give a brief review of what your successes were, and how it advances the state of the art. Then you can bring up limitations, i.e. something that didn't work, or you could have done differently. Limitations lead naturally to future work (which is sometimes part of Discussion, sometimes a separate section, depending on the emphasis of the paper). Do not be shy or protective of your ideas: if someone else does what you proposed, that's great (as long as they cite you for it).

Conclusion should really conclude, i.e. you should step back and evaluate the main results in the context of the big picture: what your results were, how they contribute to the big issue, and what can be done in the future building on this work. It is not a summary.

You should also use subsections; that way it is easier for the reader to put it all together. Each subsection should cover one coherent topic, typically about a 1/3-1/2 page in length (in a double column format). If you have subsections, you should also have an intro paragraph in the main section saying what they are. But don't go too deep and chop your paper into many short subsubsectdions or even subsubsubsections. The paper should present a narrative, not a manual.

The process of writing

Writing papers is a bit of work, but you get better with practice. There are two principles that allow you to write the paper with the least amount of effort and iteration: the "iterative deepening" principle, and the "outward expansion" principle:

In "iterative deepening", you'll start by putting together a high-level outline consisting of the above main sections. Then you put in each of the above main questions: they should be your outline for the abstract and intro, and each question should also be included in the outline for the main text. Then you start expanding the outline, gradually adding more and more detail (using the "outward expansion" principle below). Once you have a note for each paragraph, you can start writing out the paper in actual words.

In "outward expansion", you'll start the writing from the middle of the paper and expand towards the beginning and the end. That is, you first write the description of the architecture/model/approach, experiments, and results. Those are easy to write because they are very concrete. After that, you write the background/related work/research base, once you know what is relevant. Then you go on to discussion and future work, and then to the intro and the conclusion, once you know what the paper is all about. The last thing you write is the abstract, after you are done with everything else. As you write, you may come up with ideas to be included in the other sections, and you should add them to the outline, to be expanded later.

Typically, you should utilize both principles: you'll start with a complete high-level outline, and in every iteration, you'll deepen it by expanding out from the middle. When you first write it out with actual words, don't worry too much about expressing everything in perfect form: it is more important to just get the ideas on paper. After the whole paper is done, you'll read and edit it a few times and polish it up.

Polishing up the expression is important. Above all, write in plain English; the reviewers have to read 10 papers in 2 hours (or something like that), and they are not going to be impressed by long, complicated sentences; they will simply not read them. Use strong verbs, avoid complex noun groups. Avoid using "we", "us", "our"---telling the story from your perspective does make your writing look pedestrian (like "what I did this summer" instead of scientific (which focuses on the science rather than the scientist). It is hard, but with practice becomes a second nature and more effective.

Use a lot of figures and tables to illustrate your ideas. Explain the main points of each figure/table in the text, and the details (that are necessary to actually understand the figure) in the caption. Each caption should have three elements: (1) A standalone title that specifies what it is about, and forms a consistent story throughout the figures; (2) explanation of what is in the figure, enough to understand it; and (3) the point, i.e. the reason why the figure is there: what is it trying to communicate? It should be possible to just look at the figures and get an idea what the paper is about.

Each of the main ideas that is not your own should be referenced, sometimes with a couple of references. It serves to show that you know the area, and also that your own ideas are based on existing, well-known work, which makes them more believable. Clearly indicate in the paper, however, which of the ideas are yours.

If you have trouble getting started, don't panic---it is competely normal. There is a trick that may work for you: set aside e.g. 30mins at the beginning of every day to write. When the time is up, you stop, no matter how much or how little writing you actually got done. In a couple of days, you actually get into the habit of writing, and make progress in non-scary baby steps. You may then increase the time e.g. to one hour, or two, but you still stick to some time window, and still do it every day. This way you can make a lot of progress in e.g. a couple of weeks.


At the moment the best tool for collaborative writing is Overleaf. It provides the full power of LaTex, makes it easy to see the formatted result, and allows tracking changes and comments easily. It is possible for several people to work on the same paper at the same time. And the output is superb compared to e.g. Word or Google Docs (especially for math). If your account on Overleaf does not support change tracking, ask Risto to make a copy of the project that does.

Make sure you use the right template for whatever you are writing, e.g. a thesis or a dissertation, or a particular conference or a journal. Also make sure you adhere to the Call for Papers, especially the page limits and anonymity requirements. Note that we have a large collection of references in cs:/u/nn/bibs, but many conference and journal web pages provide bibtex refs as well. So most of the time you don't need to type them up to overleaf by hand, but collect from various sources.

Further Reading

There are several good books on writing style and how to put together scholarly articles. Try for example Strunk and White: The Elements of Style, or van Leunen: "A Handbook for Scholars". You may also want to check out the suggestions on Bill Starbuck's collection of writing tips or Mark Leone's collection of advice on research and writing, Jim Bednar's collection of tips gleaned from grading the CS technical writing course.

Last modified: Sun Sep 25 16:14:02 PDT 2022