Jim Kurose, Department of Computer Science, University of Massachusetts
-
Every paper tells a story.
- what is the “elevator pitch” of your story? (elevator pitch = summary that is short enough to give during an elevator ride)
- the story is not what you did, but rather: (i) what you show, new ideas, new insights, and (ii) why interesting, important?
- why is the story of interest to others? - universal truths, hot topic, surprises or unexpected results?
- know your story!
-
Write top down
- scientists (and most human beings) think this way!
- state broad themes/ideas first, then go into detail (context, context, context)
- even when going into detail … write top down!
-
Introduction: crucial, formulaic
-
if reader not excited by intro, paper is lost
-
Structure
-
para. 1: motivation: broadly, what is the problem area, why is it important?
-
para. 2: narrow down: what is problem you specifically consider
-
para. 3: “In the paper, we ….”: most crucial paragraph, tell your elevator pitch
-
para. 4: how different/better/relates to other work
-
para. 5: “The remainder of this paper is structured as follows”
-
-
-
Master the basics of organized writing
- paragraph = ordered set of topically-related sentences
- lead sentence sets context for paragraph and might tie to previous paragraph
- sentences in paragraph should have logical narrative flow, relating to theme/topic
- don’t mix tenses in descriptive text
- one sentence paragraph: warning!
-
Put yourself in place of the reader
-
take the time to write less. less is more:
- I would have sent you less if I had had time”
-
readers shouldn’t have to work
- won’t “dig” to get story, understand context, results
- need textual signposts to know where ‘story” is going, context to know where they are
- good: “e.g., Having seen that … let us next develop a model for …. Let Z be ….”
- bad: “Let Z be”
-
what does reader know/not know, want/not want?
- write for reader, not for yourself
-
page upon page of dense text is no fun to read
- avoid cramped feeling of tiny fonts, small margins
- create openess with white space: figures, lists
-
enough context/information for reader to understand what you write?
- no one has as much background/content as you
- no one can read your mind
-
all terms/notation defined?
-
-
No one (not even your mother) is as interested in this topic as you
- so you had better be (or appear) interested
- tell readers why they should be interested in your “story”
- don’t overload reader with 40 graphs:
- think about main points you want to convey with graphs
- can’t explore entire parameter space
- don’t overload reader with pages of equations
- put long derivations/proofs in appendix, provide sketch in body of paper
-
State the results carefully
- clearly state assumptions (see overstate/understate your results)
- experiment/simulation description: enough info to nearly recreate experiment/description
- simulation/measurements: statistical properties of your results (e.g., confidence intervals)
- are results presented representative? or just a corner case that makes the point you want to make
-
Don’t overstate/understate your results
-
overstatement mistake:
- “We show that X is prevalent in the Internet”
- “We show that X is better than Y”
when only actually shown for one/small/limited cases
-
understatement mistake: fail to consider broader implications of your work
- if your result is small, interest will be small
- “rock the world”
-
-
Study the art of writing
- writing well gives you an “unfair advantage”
- writing well matters in getting your work published in top venues
- highly recommended:
- The Elements of Style, W. Strunk, E.B. White, Macmillan Publishing, 1979
- Writing for Computer Science: The Art of Effective Communication, Justin Sobel, Springer 1997.
- who do you think are the best writers in your area: study their style
-
Good writing takes times
- give yourself time to reflect, write, review, refine
- give others a chance to read/review and provide feedback
- get a reader’s point of view
- find a good writer/editor to critique your writing
- starting a paper three days before the deadline, while results are still being generated, is a non-starter