How (not to) write a paper
Recently, one of the professors at UBC gave a talk on what to (and not to) do when writing a research paper. A lot of good points were raised concerning who is reading your paper, what they want to know, and how best to make your paper best fill those needs.
There’s a number of technical-/academic- specific writing points, but much applies in general. Know your audience, give them what they want. The majority can be summed up by the last point of Kurt Vonnegut’s How to Write With Style:
PITY THE READERS
They have to identify thousands of little marks on paper, and make sense of them immediately. They have to read, an art so difficult that most people don’t really master it even after having studied it all through grade school and high school — twelve long years.
So this discussion must finally acknowledge that our stylistic options as writers are neither numerous nor glamorous, since our readers are bound to be such imperfect artists. Our audience requires us to be sympathetic and patient readers, ever willing to simplify and clarify — whereas we would rather soar high above the crowd, singing like nightingales.
My somewhat erratic notes of the talk follow:
who are you writing a paper for?
-
supervisor:
show them your hard work, knows everything
writeup / project report
-
other researchers:
give something to the research community at large
list facts + practical details
no need for motivation / justification
-
reviewers:
how is an opinion of a paper formed?
motivation / results come before algorithm
why are you telling the story chronologically?
that’s not the order that the reader cares about the information you present
What not to do: Guess my contributions game
reader will understand contribution from description
be explicit, consider carefully
often different from original goals
have a short “take home message” — repeat
motivation and contribution at every level — heirarchical
What not to do: I am so unique
Ignore previous work
Similar problems or solutions
Enumeration
“X did Y” not enough
Different not enough
- Must say why previous work doesn’t solve your problem
- What limitation of theirs do you fix?
What not to do: Trust me
Why did I include this example?
Every example should have a story.
What not to do: What had I done this summer?
We did this, we did that, we did that other thing
- How allowed only after what and why
- Motivate every choice you make
- Compare alternatives
- Cite/contrast sources
- Highlight cool ideas
- Can this be applied elsewhere?
What not to do: Deadly detail dump
You spent 3 months optimizing this part of the code
- The world needs to know about it!
Prioritize ideas and focus on the major ones
- Motivation: why should I care
- Overview: what did you do
- Details: how did you do it
What not to do: Hiding bad writing - Jargon attack
Notations, notations, notations
- Use every letter of Greek alphabet
- Change notations throughout the paper
- Unique notation system (x function and f value)
Ab(use) pseudo-code
Invent new terminology
- Bonus points 1: re-invent terminology
- Bonus points 2: use acronym
Use plain English
Avoid buzzwords
What not to do: Oversell
Insult previous work
- “We do not show the pitiful results of our competitors”
Make unproven claims
- “Our new method is proved to achieve dramatic speedup compared to
old method”
- Reality: on only one example we showe it is 1% faster
It is about you not “them”
- Use “augment,” “improve,” compare results
- Have a short list of demonstrated claims
What to do: Paper’s message
- State contribution
- Usually combination of problem and solution
- Keep it compact but clear
- Key idea - take home message
- What is the major feature of my solution
- Convincing results
What is the elevator pitch for your paper?
Consider actually giving a talk on it before you write to clarify you ideas
What to do: define your contribution
- Often not obvious
- Frequently diverges from original goals
- Introduce / define problem
- Why is my solution good?
- Why better than others?
- What can we do that wasn’t possible before?
- How can we do something better than before?
- What do we know that was unknown/unclear before?
What to do: contribution determines everything
- From high-level message to which details to include
- State explicitly and clearly in introduction
- Don’t hope that the reviewer or reader will fill in for you
- Goal is clarity, not overselling
Paper parts: title + abstract
- Determine who will review/read
- Think which community you want to review this
- Ideal title: self-explanatory + catchy
- Abstract: contributions go here
- Keywords/Categories - again imports reviewer community
- Think well who you want to target
Consider choosing your bibliographic citations to guide the choosing potential reviewer process.
Paper parts: intro
- Most important section of your paper
- Determines >50% of reviewer opinion
- Motivation + overview
- Sometimes include previous work (or part of)
- Motivation = problem + contribution
- Overview
- High-level view of the technique
- Key algorithmic contributions
- Insights
-
NOT a table of contents
Figures
- Figures + captions = paper summary
- First thing reviewers/readers look at
- Self-explanatory (with caption help)
- Convey main points
- AESTHETICS MATTER
- Main tool for judging your method
- Examples of what?
- If you want reviewers to notice something highlight it!
-
Rendering critical
- View, lighting, texture, resolution, …
Other references
![“Bruce had me up to three miles a day, really at a good pace. We’d run the three miles in twenty-one or twenty-two minutes. Just under eight minutes a mile [Note: when running on his own in 1968, Lee would get his time down to six-and-a half minutes per mile]. So this morning he said to me “We’re going to go five.” I said, “Bruce, I can’t go five. I’m a helluva lot older than you are, and I can’t do five.” He said, “When we get to three, we’ll shift gears and it’s only two more and you’ll do it.” I said “Okay, hell, I’ll go for it.” So we get to three, we go into the fourth mile and I’m okay for three or four minutes, and then I really begin to give out. I’m tired, my heart’s pounding, I can’t go any more and so I say to him, “Bruce if I run any more,” — and we’re still running — “if I run any more I’m liable to have a heart attack and die.” He said, “Then die.” It made me so mad that I went the full five miles. Afterward I went to the shower and then I wanted to talk to him about it. I said, you know, “Why did you say that?” He said, “Because you might as well be dead. Seriously, if you always put limits on what you can do, physical or anything else, it’ll spread over into the rest of your life. It’ll spread into your work, into your morality, into your entire being. There are no limits. There are plateaus, but you must not stay there, you must go beyond them. If it kills you, it kills you. A man must constantly exceed his level.”
— From The Art of Expressing the Human Body](http://25.media.tumblr.com/tumblr_l3gkjjNcKN1qz4cueo1_400.jpg)
