Giving a Good Research Talk

Some thoughts on giving a decent talk, in response to seeing some bad ones. This page is inspired by similar advice from Simon Peyton Jones, John Launchbury and John Hughes. Some of my points overlap theirs.


Why give a talk? A talk is an advertisement for the paper. If people are sufficiently interested, they will read the paper. To repeat excellent advice from Peyton Jones et al: what is the one thing listeners should remember from your talk? You should know this, and your audience should too -- tell them!


Don't try to present the entire paper in the talk. Some details will have to be omitted. Also, the structure of the talk may not match the structure of the paper. In coming up with a simple, clear structure for the talk, you may realise that the talk structure is clearer and better than the original paper's structure.


Slides accompanying a talk can serve up to three distinct purposes, each one for a different group of people:

  1. For the audience: they make the talk easier to follow and understand.
  2. For the speaker: they serve as speaking notes.
  3. For people not at the talk: they can be read later, to learn what the talk was about.

Purpose 1 is the most important. Audience understanding is helped most when you present information that is difficult to communicate by voice: pictures, graphs, tables, equations, small code snippets, web addresses, etc. It is also useful to show key phrases or definitions. Text beyond that should be sparing. (An interesting question: does every slide need a title?) Less is more.

Purpose 2 is less important, and it conflicts with purpose 1. Obviously, you don't want to forget important things, but you shouldn't expose the audience to the mechanics of how you achieve this. Presentation software like PowerPoint allows you to add notes to a presentation that are not visible when the talk is given. Use them. You can even script every word of the talk in these notes. (This really helps in making sure the talk goes for the same length every time.) But be careful: writing and speaking are different. Many sentences that are fine when written down sound stiff when spoken. You may need to rewrite the notes multiple times to match how you naturally speak.

Purpose 3 is also less important, and also conflicts with purpose 1. Again, audience-invisible notes can solve this problem. However, talks often do get distributed in a form (such as PDF) in which any additional notes are not visible. Therefore, you probably want to think about purpose 3 to some extent, and put the most important points in a form that is brief but intelligible to someone who didn't attend the talk. But don't go overboard -- if they want more information, they should read the accompanying paper.

It is amazingly hard to write slides that truly, primarily serve purpose 1, and not get sucked in by purposes 2 and 3. In particular, when you first write the slides it's natural to write lots of bullet points and whole sentences of text, because you write down everything that you want to say. Once you've done this, try to remove as much text as possible, by (a) converting text to other forms (and note that small animations can be invaluable for guiding an audience's attention around a picture), and (b) presenting the remaining text as concisely as possible. This is hard work -- text is faster to create than pictures, tables, etc -- but it results in a much better talk.


If you can, check beforehand that the connection to the projector works, to avoid embarrassing technical problems. Also, have backed-up versions of your talk in at least one other place: a website, a CD, a USB keychain, or on transparencies.

Get laptops ready as much as possible before you start to minimise fiddling time. This means starting PowerPoint (or whatever you're using) and having your first slide loaded ready to go. The worst thing you can do is plug your machine into the projector, so everyone can see the machine's display, then boot it up. It will take one minute but feel like ten.


Make it obvious when you have finished! Otherwise you will have those excruciating few seconds of silence while the audience decides whether to applaud. It mostly comes down to the tone of your voice. If you finish with a sentence that clearly wraps things up and then say "thank you" in the right tone, it's obvious you have finished and the audience will applaud. Practice beforehand, it's easy.

Don't just trail off. Don't mumble "and that's about it". If you have a moderator, don't ask for questions -- it's their job; if you do it, the audience will be unsure whether to applaud, which can be awkward. If you don't have a moderator, ask for questions after the applause.


Don't exceed your time limit. It's rude, and annoying for the listeners. If you don't time yourself before to get this right, you will almost certainly overrun.

How many slides do you have? Time per slide varies between speakers, and between talks. However, if you have 25 minutes and 35 full slides, that is too many.