This post is a collection of paragraphs without strong dependencies.
Writing is much more visual than we care to admit. Academic writing is even more so. To see this, think how we read a paper; what elements are most prominent?—title, abstract, conclusions, figures, section headers, algorithms, tables, equations, other text. Even text, the least visual element, is visual to some extent—long paragraphs can be surprising unappealing (where should I start, what is important).
Think about a paper that disregards the most visual elements in favor of large amounts of text. How excited are you about reading it? Unless it is really good exposition (which often isn’t), it is likely really boring. A paper needs clear visual markers that guide the reader’s attention.
Good writing presupposes understanding of what makes good reading.
Reading is often done with a purpose. Most readers will not read your paper linearly from start to finish. Your writing should support targeted reading.
Papers that fail to guide the attention of the reader feel like laundry lists. They wander aimlessly without much thought to what points should be driven. Eventually, the reader gets lost, loses interest, and gives up. It is your job to guide the reader through the interesting things that you have to say.
Structure first, then execute. There is no point to start writing if you haven’t thought about the message of the paper. There is no point to start editing if you haven’t written the complete message yet. First nail down the high level aspects of the message, then progressively flesh them out. Always be wary of working at multiple levels simultaneously. For nailing the structure, thinking about structuring a presentation around your work often works well. Then write the section headers. Then write placeholders for the figures, algorithms, and tables that you want to have. Then write the first sentence of each paragraph. Then write the rest of the paragraph. This it is just like “connect the dots”; all the high-level structure is in place.
Take a real estate perspective. The most expensive real estate is your title, followed by the abstract, the introduction, and the first sentences of each paragraph. Use these wisely. The way you use them signals to the reader the quality of the content. If you waste precious real estate on trivialities or simply on not driving your story forward, the reader will lose interest and think that you might not have much to say or that the contributions are uninteresting.
Fight the urge to be vague. One idea that stuck with me is that for you to have a paper, you must have a falsifiable contribution, i.e., you must be clear about the problem that you are trying to solve. It is then up to the reader to validate that you have solved the problem. There is often the risk to state your contributions too vaguely. For example, claiming that you are going to study Y. This is not falsifiable. You must single out aspects of the problem or approach that are novel. Until these has been established, you don’t have a paper.
Concrete examples first, abstraction later. It is easier to go from concrete to abstract, than to present the abstract and then the concrete. A concrete example allows the reader to see the overall structure of the object of interest and it much more digestible than abstract formalism. Concrete examples scaffold the presentation.
Establish the contribution early. The reader shouldn’t have to wade through the whole first three pages to understand what the paper is trying to establish. This should clear from the abstract, and then perhaps with more detail from the introduction.
How not to write papers. Distract yourself enough to start writing aimlessly. Do this for one hour. Spend the next three hours trying to figure out where do you want to take the paper and how to edit the text you’ve produced.
Figures serve two purposes. First, they signal that specific information was important enough to merit a figure, and therefore it is important to understand it carefully. Second, they communicate an important idea that might be more effectively communicated visually than through words.
If you don’t have all the figures yet, make placeholder figures. The closer to the format of the final desired figure, the better. This will show what data you need to produce to write the final version. Placeholders will allow you to construct the narrative in the meantime.
Write the paper at different scales. Write strong paragraph openings—it should be possible to read only the first sentence of each paragraph and still get a decent idea of where the paper is going and what it is covering. Try it. Once you have a first sentence for a paragraph, the rest of the paragraph should fall into place.
There are two places that you should reserve dearly: the beginning of the paragraph and the end of the paragraph. The beginning of the paragraph should be strong. The end of the paragraph should be memorable.
Don’t open a paragraph with a generality. Don’t end a paragraph with a generality. Better yet, don’t write generalities. It is an incredibly boring way of opening a paragraph or a section. You will lose the reader’s attention.
Make your sentences crisp. If you are six words deep into a sentence and you still don’t know what it is about, that is a problem. This is doubly true for the first sentence of a paragraph. Each paragraph that is not opened strongly is a lost opportunity.
The verb is often the highlight of the sentence. While reading a sentence, we feel kind of lost until getting to its verb. I think that this is one of the reasons why “Note that” is so aggravating: it adds extra words before the verb without contributing any information. Rewriting the sentence to bring the verb closer its beginning often improves it.
Constraints are forcing functions to write. If you have two weeks until the deadline, you will find a way to fill them with unimportant work. If you have only two days, you will take what you have and write the paper and it will often look just as polished.
I like to write liberally and then read the complete draft and mark it up with what is missing and what needs to be removed. If there is a sentence that needs to be written, I write it right away. If there is a paragraph that needs to be written, I write its first sentence. After I’ve finished giving a pass over the paper, I implement the edits and repeat. Marking it up splits work in two phases: deciding which edits to do and doing those edits. This saves you the back and forth between reading and editing which I find cognitively inefficient.
Editing is best done after letting the draft rest for a while to let the emotional attachment to one’s own writing wane. After that, it is easier to ask yourself “do I really think that this sentence adds anything to the narrative? This is best done for drafts that are already in a reasonable state in terms of structure and content. In the early stages of writing there is still a lot to improve, so there is often not much emotional attachment to it.
A writing checklist:
- Are sections sequenced naturally?
- Do all paragraphs start with a strong sentence that opens up the topic of the paragraph?
- Are paragraphs coherent, with a single idea developed?
- Are paragraphs sequenced in a way that supports the topic of the section or subsection?
- Do figures strongly support the discussion of the paper?
- Are all figures and tables mentioned and discussed in the main body of the paper?
- Is the work well contextualized with existing work?
- Are all symbols clear or explicitly defined?
There are two types of sentences: those that move the narrative forward, and those that don’t. Sign posting is almost always of the second kind. “In section 1, we do X; In section 2, we do Y; In section 3, we contrast X with Y”. This is boring to read. If you have done a good job structuring your paper, it should be obvious what is presented in each section and how the sections support the narrative.
As important as deciding what to include is deciding what not to include. Is it really that important for the message of the paper to tell the reader that we are dealing with a special case of paper X and Y? Save side remarks for the appendix.
When in doubt, leave it out. If you are editing a draft and are strapped for space, if you are unsure about whether a particular sentence should be kept or not, leave it out. It is unusual to regret the decision after it is taken.
Loosely speaking, sentences with fewer words are easier to read than sentences with more words. Paragraphs with few fewer sentences are easier to read than paragraph with more sentences. Sections with fewer paragraphs are easier to read than sections with more paragraphs.
Style improvements are best left to the editing stage:
- Avoid excessive nominalization.
- Pick strong active verbs.
- Avoid the passive tense.
- Avoid excessive hedging (e.g., maybe, potentially, it might be, it seems.). Think what you want to say and say it.
- Avoid excessive negatives. Is something not fast? It is slow then.
“Knowing is not enough, we must apply. Willing is not enough, we must do.” - Bruce Lee
When it comes to improving, nothing replaces experience. Tips and references can only provide guidelines to incorporate in your work. You will get better as you try to apply them.
Simon Peyton Jones has a really good talk titled How to Write a Great Research Paper. Well worth your time. There is also a course on Coursera titled Writing in the Sciences that has some good content. Videos are available on Youtube. Unit 4 was one of my favorites (I’ve recommended it to other people a few times), but overall I think that it’s going through the whole thing (maybe at 1.5x or 2x speed).
In regards to reading material, the standard reference is Elements of Style. If you only read one, read this one. Writing in Computer Science is also pretty good and goes into specifics for computer science research. On Writing Well, Style: Toward Clarity and Grace, Sense of Style are also relevant. Sense of Style was one of the first ones I read and enjoyed. I also very much enjoy How to Write Mathematics by Paul Halmos and some lecture notes on mathematical writing by Donald Knuth. The one by Paul Halmos one is a short essay. Read that one for sure.
I’ve written a script to catch and revisit redundant or verbose phrases. This is mainly useful in the later stages of writing when you are optimizing individual sentences and paragraphs.