How to give a talk
Ulle Endriss — created: 7 December 2010 — last updated: 30 May 2017
Below I'm offering a few opinions about how to give a good talk. This is about scientific or scholarly talks presented in an academic setting. The typical talk I have in mind is a contribution of, say, 20-30 minutes at a conference, but much of what I have to say also applies to longer talks, e.g., the kind of talk you might deliver at a departmental seminar. My own field is AI and my suggestions apply, in particular, to talks in AI that combine contributions of a conceptual and a mathematical nature. But from what I have seen, things are not all that different in many other disciplines.
These notes have grown out of a set of guidelines originally written for the students in my course on computational social choice in 2010.
There's lots of information out there, with many other people also offering their opinions about how to give a talk. You should have a look at all of that stuff as well, particularly by people who work in a similar field as yourself.
For any of the guidelines I formulate here, feel free to not follow them. All I ask for is that you think about these issues and, for any particular piece of advice, make a conscious decision about whether or not to follow it. This will make attending talks a lot more enjoyable and productive for all of us.
Types of Talks
There are different types of talks. You might use slides (common in Computer Science), give a blackboard presentation (common in Mathematics, at least for longer talks), use a handout (common in Linguistics), or read out a carefully crafted paper (common in parts of Philosophy). I'm only going to talk about the first option here, as it is the only one I know how to do well myself.
I think using slides is an excellent way of giving a talk. It's main—maybe even only—disadvantage with respect to all the other types of talks mentioned above is that it allows the speaker to move at a much faster pace than the audience ever could. As long as you are aware of this pitfall and try to avoid it, you can safely work with slides.
The Objective of the Talk
Think about why you want to give your talk before you start preparing for it.
If you are giving a short talk at a conference about your own research, then your objective should be to convince the audience that the problems you are working on are important and interesting, and that the paper with your latest results is worth reading. You don't have to list all your results and you should not be under the illusion that it will be feasible to fully explain your results in the time available. Instead, by carefully choosing bits and pieces that you can explain in the context of the talk, you should try to convince the audience that you are serious and that your judgment on what is significant and what is technically correct can be trusted.
For a longer talk, such as a seminar talk given in front of a small audience of—mostly—specialists, your objectives may be somewhat different and you may really want to, say, go through a proof in some detail. But even then, your main goal should be to get people interested in the topic and to read your paper.
The Structure of the Talk
A good talk will have a beginning, a middle, and an end.
So, don't forget about the beginning, i.e., don't just jump right into the technical contribution of your research without having motivated it first.
Also don't forget about the middle part either, i.e., don't talk only about motivation or background material or related work, but make sure you use most of the time you have to talk about the real thing: your contribution.
Finally, don't forget the end, i.e., don't just stop after you have stated your final result, but spend some time on a proper conclusion in which you make sure your take-home message really does hit home.
Here is a suggestion for a set-up that will work well in most cases:
Start with a very short high-level explanation of what the problem is
you are going to discuss. Maybe you can give this explanation while
your title slide is up, or maybe you want to prepare a very short
slide (definitely not more than one!) specifically for this purpose.
Maybe that short slide could involve a simple example.
It is common practice to follow this up with a "talk outline" slide, where you sketch, in a few bullet points, what your will talk about. It is easy to do this badly. At one end of the spectrum of ways in which to get it wrong, don't just read out a list of difficult words your audience won't understand, as you haven't defined them yet. At the other end of that spectrum, don't just use generic words, i.e., don't just announce that you will first cover some preliminaries, then present your results, and then conclude. The purpose of presenting the outline is that it will help the audience to orient themselves later on, when they try to build a mental picture of the overall (presumably nonlinear) story you are telling them, whilst processing your (necessarily linear) input.
For a very short talk (of, say, 10 minutes or less), you may want to omit an explicit presentation of the outline.
For the middle part of your talk, it is much more difficult to give general advice on structure. But often it makes sense to first cover some preliminaries, e.g., a slide or two leading up to the central definitions of your framework, and to then present a small number of representative results. But it usually is not a good idea to first present all your definitions, and then all your results. People won't remember all of those definitions for long enough. So present minor definitions locally, in the context of the results in which they are needed. (This is one of the differences between how to write a paper and how to give a talk, as the reader of a paper has the freedom to move back and forth between definitions and results, which a member of the audience in a talk cannot do.)
At the end of your talk, you should summarise again what you have done,
offer some kind of evaluation of the results presented, and make suggestions
for interesting and potentially fruitful directions for future work.
Make sure you use your final slide to (once more) get across your main message.
A good conclusion of the kind sketched above, presented on a single slide, will do the job.
Thus, your final slide should most definitely not just be saying "Thank you!" or "Questions?", it should not show a cartoon (however funny it may be) or a picture of your favourite pet or child (however cute they may be), and it also should not be a list of all the references you cited earlier on in the talk.
Why? Due to the question period, your last slide will typically be visible for much longer than any of the other slides. That's an opportunity not to be missed! So make sure it optimally serves your objective. A well-done final slide also helps the audience to ask relevant questions.
Preparing the Slides
Here are some basic issues to keep in mind as you are preparing your slides:
- Font size.
Use a font that's large enough for everyone to be able to read your slides comfortably.
- Number of slides.
Choose a realistic number of slides. A very common mistake (not only of beginners) is to bring far too much material to a talk.
Personally, for a short conference talk of 20 minutes I usually aim for around 12 slides, for a 45-minute seminar talk I usually aim for something just under 20 slides, and when I teach a 90-minute class (and don't need a substantial amount of time for other things, such as discussing homework), I know I can manage around 24 slides without having to rush things.
- Density of slides.
Do not put too much information on a single slide.
At the same time, if you put too little information on a single slide,
you may be tempted to switch to the next one too quickly.
Personally, I prefer somewhat more populated slides on which I then spend several minutes each.
This gives the audience a bit more freedom to adapt to my pace, and to occasionally look either back or ahead.
- Content of slides.
Only put material on the slides that you actually want people to read
(of course, there are some rare cases where you might want to deviate from this rule).
For a typical talk at a conference, your slides should be optimised for the
audience experience during the talk. This is different from slides
used for teaching, where you probably also want the slides to be useful for
your students afterwards, to look things up. That is, good teaching
slides might look subtly different from good conference slides.
Don't use too many transitions within a single slide (personally, I usually
don't use any). There are several reasons for this. First, it can come
across as patronising if you are trying to dictate in too much minute
detail what your audience should focus on at every single moment in
time. So, even if you do make use of a modest number of transitions,
definitely do not grey out again the text you have gone over already
(it is extremely annoying when a speaker does that). Second, having
too many transitions gives you less flexibility to adapt your speed
in case you have to go a bit faster during the second part of your
talk. Third, if during question time a member of the audience asks
you to return to a specific slide, then that can be more difficult if you
have too many virtual slides making up one proper slide.
It often makes sense to mention a few carefully selected references to prior work on your slides. When you do this, make sure the audience has a chance of understanding what papers you are talking about (and to look them up, if they want or need to). At the very least, include the names of the authors and the year of publication (i.e., a numerical reference to some invisible list of references somewhere else on your slides is completely useless). Even better if you can also include the journal name or conference acronym. Personally, I almost always put the full reference at the bottom of the slide: people familiar with the papers will be able to recognise them in an instant, and newcomers who may look through my slides after the talk will have all the information they need to locate those papers.
Delivering the Talk
Next, a few tips regarding things to keep in mind as you are delivering your talk. If you know that you usually tend to get a couple of them wrong, particularly when you are nervous about presenting in front of people, make it a habit to remind yourself of those things directly before your talk (for me, this is: "speak slowly!").
Speak loudly and clearly and not too fast. Face your audience while speaking.
- Audience awareness.
Try to actually look at the people in the audience—this is surprisingly difficult and requires a lot of practice!
You should notice when someone raises their hand to ask a question (it is very common that everyone in the room sees that hand up there, except for the speaker). You should also notice when the chair is trying to give you signals about your time running out. You should have some idea of who's (pretending they are) following your explanations and who's almost asleep. You should not constantly stare at that same person sitting in the front row.
Make sure the audience looks at you every now and then, not just at your slides. This makes for a more engaging listening experience for them. So move around, use your hands, vary the way in which you speak.
Keep in mind that what you present is difficult and that most members of
your audience probably will not know much about the topic beforehand.
Say what you will say, then say it, and finally say that you said it.
This is good advise at all levels of a talk. At the global level,
for instance, say what you will say as your outline slide is on,
then give your talk, and finally say what you said with your conclusion slide.
But the same advise also applies to a single slide (at least
when it is one of the more difficult ones): first say what the next
point will be, then make that point, and then briefly summarise what
you have just done.
This may not actually be a word, but here's what I mean.
Keep in mind that most people in the audience will not pay 100%
attention all of the time. Try to organise your talk in a way that
allows them make up for what they may have missed and to reconnect again to your talk.
For instance, to give an overly simplistic example, you may have defined that N is the
set of agents on slide 3, but it does not do any harm to occasionally
remind people by writing (and saying) "the set of agents N" rather than just "N"
also on later slides (unless those slides are very crowded already). Also, do
occasionally remind people where in the talk you are right now
("we have just seen that X is not possible, so now we will try Y").
Use familiar terminology and notation wherever possible, to make it easier for people to relate what you do to things they know about already.
Prepare well, but not too well. In my experience, it pays off to know
exactly what you want to say during the first two minutes or so (when
you'll still be a little nervous), but for the rest it is better not
to have everything planned out word for word. The reasons are that if
you are over-prepared, then you will tend to speak too fast, you will
sound boring, you will not think yourself while you are presenting and
thus lose the feel for what's difficult and what's routine (thereby
making life hard for your audience). You also risk getting totally
derailed by a simple clarification question or some other unexpected event.
Don't be afraid of questions.
The worst that can happen to you as a speaker is when nobody has any questions. If they ask question that are on topic, you surely will have something to say. If it's a difficult technical question, nobody expects to get a perfect answer right away. And if it is about related work you are not familiar with, it's fine to just say so and to thank the person asking for the pointer.
As a member of the audience, do your part and contribute to scientific progress by asking good questions. During the talk itself, if it is a relatively short talk with strict time constraints, it's usually best to avoid questions or to restrict them to very short clarification questions that you feel really need an answer, right now and not just for your own sake. For longer talks, on the other hand, a few questions throughout the talk make the whole thing more interesting and fruitful for everyone involved. After the talk, the audience is expected to come up with at least a couple of questions for the speaker. If you are chairing the session, you must ask something if nobody else does.
Besides questions, other types of feedback can also be useful. But make sure you are not coming across as lecturing the speaker. And always resist temptation to talk too much about your own work.
If the speaker is a relatively junior member of your research community, and maybe their senior coauthor or supervisor is also present, when asking your question or giving your comment, resist temptation to address that more senior person you are trying to impress rather than the speaker standing in front of you. (And if you are that senior coauthor or supervisor, just sit tight and don't talk during any of this.)
Before your talk, familiarise yourself with the conventions of the research community you are talking to (and possibly also the conventions associated with the specific event where you are presenting).
For example, in my own community (in AI), there will be a someone chairing the session. You are expected to introduce yourself to them a few minutes before the talk. They are expected to introduce you to the audience before your talk. They will usually mention your name and the title of your paper. If you have coauthors, you are expected to mention them at the start of your presentation. If any of them are also present, it is nice if you point this out explicitly.
At the end of your talk, people are supposed to clap, and then the chair will invite the audience to ask questions. So don't invite the audience to ask questions yourself. First, that would be (a little) rude towards the chair. And second, it will confuse the audience, as they won't know anymore when to clap. During the question period, the chair also gets to point out who is next to ask their question and the chair decides when the question period needs to end. Of course, every now and then you will run into a chair who hasn't read the Big Book of Rules themselves, and you may simply have to take over some of their tasks.