Luis Pedro Coelho
Most important question: What is the purpose of the talk?
Types of talks you might encounter
- Research talks: one research result (1-2 papers, 10-30 minute slot).
- Seminar talks: one research program (1 lab, over a few years, 45-60 minute slot).
- Keynote talks: one research field (many labs, over decades, 60-90 minute slot).
- Outreach talks (to lay people, e.g., TED talks): many formats.
- Lectures (like this one!)
Many lessons apply to all these types of talk, but there are specificities too. We are focusing here on the research talk context: a presentation of less than 30 minutes of novel scientific work to an audience of peers.
This lecture has too much text to be a good research talk!
To whom are you presenting?
What do they know? What should they know after you present?
Scientists at a conference are (generally):
- knowledgeable and smart.
- very busy and will see 20 presentation + 50 posters that day.
Many structures are possible. A generic one is
- Background
- Specific problem
- Your results
- Conclusions
Often, this is called the introduction, but I find that word overly broad.
What you need to do is take the audience from where they are to where you are. The purpose is to make the audience care about the specific question that you studied.
Explain/discuss:
- Broader context (scientific/societal)
- Current/previous solutions (limitations, trade-offs)
This is where you need to think "what does your audience know? What do they care about?"
Avoid clichés or statements that are obvious to your audience, no need to say Studying the brain is important for understanding cognition in a neuroscience meeting!
- This part should at least 50% of the time.
- It is your work, it can be plural you (your group's work), but it is should be new material (for the audience).
- Do not present more background than necessary.
- Even if you are presenting a paper, you do not need to present the whole paper. It can just be an advertisement to get people to read/cite it.
This is thinking of a research talk in a field-focused conference, an invited talk at a department to a broader audience can be different.
Tell a story. This does not mean the real story of how you got to the results. That can be interesting and is one way to do it, but not always the best. What is meant by story is a scientific story, namely it means that there are natural, logical, transitions between slides and not just one thing after another.
- Repeat the key points.
- Not too many.
Tip: when preparing start with the summary slide. In 10 minutes, you will be able to transmit 2-3 points, every slide should serve to reinforce these points (if they don't, either toss them out or revise the narrative).
- Often, it is appropriate (even required) to acknowledge co-authors/funders/...
- Make sure people know when you are done, do not trail off.
- Some people say I will take questions, but let the chair do that (if there is one).
Assuming you have 10 minutes:
- Background (2 min)
- Specific problem (2 min)
- Your results (5 min)
- Conclusions (1 min)
- Keep to the time limit (including leaving time for questions!). It is better to be short than long, but not too short either (take 90-100% of the time you are given).
- Briefly acknowledge the invitation and the audience.
- Slides are for support.
- You can use material from manuscripts &c, but slides are their own format.
- Reuse is good, but each talk is different.
- Less is more (bigger fonts: many projectors can only do low quality projections, even today).
- Content-full animations and pre-prepared callouts are excellent, but do not over-use
- Be polished: no spelling mistakes, consistent use of colour/style,...
- Have text. Not too much. Don't read it.
Disease X is a very important disease, with increasing prevalence in the population. For example, in the US, an estimated 30% of the population will be diagnosed with X in their lifetimes (compared to only 8% in 1999). In China, the rate is still much lower, 12%, but expected to grow as population increasingly adopts a sedentary lifestyle.
Just the main points (with references) and the narrative can be added orally:
Prevalence of X (lifetime):
- US 1999: 8% (Author et al., Journal of X, 2004)
- US 2020: 30% (Author et al., Journal of X, 2018)
- China 2019: 12% (SomeoneElse et al., Journal of X, 2017)
- You need to explain what you are showing. Explain the axes on plots.
- You should not be reading.
- Look to the audience.
- Watch your body language.
When you practice, there will be things that your test audience will point out or ask. Remember, the goal of practice is to improve the presentation. If there is a question in practice, you should:
- improve the slides/delivery to make the question go away.
- prepare supplemental material if that question comes up live.
- be prepared to answer the question if it comes up, but realize that there is never a perfect presentation for everyone.
Normally, the first practice will involve a lot of #1, then if you polish more and more, you have feedback that falls into group #2/#3.
People will (hopefully) ask you questions!
If you know the answer immediately:
- answer it, but keep the answer short. You can use formulations such as [SHORT SUMMARY], and I can go into more detail offline/during the coffee break
- clearly mark where you are speculating, I do not have the data for it, but my intution/mental model is ... or one thing that is consistent is that...
- engage with the person.
If you do not:
- deflect, while I do not have an answer to that exact question, we did do analysis X (obviously, only if analysis X is somewhat related). You can then transition smoothly into speculation if appropriate.
- say I don't know or that was outside the scope of this project.
Almost always, there will be a small set of questions that keep coming up (which is why preparation is important), for those you can pre-prepare additional material.
You can follow-up during the break (network)!
- Rules of thumb are mostly useless: 30 seconds per slide, one minute per slide, two minutes per slide, ... Really depends on your style and some slides may require 10x more time than others.
- Do not be faster than if you were writing on the whiteboard.
- If you have issues with English, slow down.
- Prepare Breadcrumbs ("after 3 minutes, I should be finishing slide 4").
- Practice the talk.
- Not just writing the slides, but the talk as a whole.
- Practice sections: a particular group of slides, even a single slide.
- Practice the end!
- Leave it natural.
You can have notes, but do not read too extensively:
- I often have 1 small note with breadcrumbs.
If you work across disciplines, it will often happen that you may be presenting to an audience that is not 100% in your field. You need to adapt, but still present your research.
You will need to
- Pay attention to jargon (avoid/explain)
- Avoid references to work that "everyone knows" in your field (maybe it is not so well known) or inside references/jokes
You may end up present less material in the time allotted: for example, if you are a computational person presenting to an applied audience, the best use of your time may be to present the intuition behind your algorithm and some empirical evidence, leaving out any formal analyses altogether. This is perfectly fine.
Do not try to present something inferior to please the audience (e.g., some speculations from your work which are only weakly supported, but fit what you think the audience wants).
- I am going to skip this: why is it there?
- Ignore this side of the plot: why is it there?
- Since I am running out of time, let me skip this section: why were you not paying attention to time before?
This is just bad preparation.
- Avoid laundry lists: "this is X, this is Y, this is Z, ...". Either you need more lack narrative structure (instead of "these are the results of X analysis", say "X analysis confirmed/disproved the hypothesis") or need to cut things out.
- be precise, present numbers and context
This is common in junior researchers.
Often, it is just a matter of re-framing. Example:
- if it your method works in a limited type of data (or your results are confined to a single experimental system), you can say our method is the best for datasets with fewer than 5,000 datapoints instead of we do not work well on large datasets.
For example, having mathematical formulas which are tangential to the point normally adds no information for the people who are know what you mean but confuses those who do not.
Instead of y* = X · argminβ { |y - Xβ|² + λ₁|β|₁}, write "using a linear model".
This does not mean, avoid formulas. If they are relevant and you take the time to explain them and are presenting to a mathematically-literate audience, your whole presentation can be a series of formulas.
- If you use material from others, you must cite it.
- If you can support specific statements with citations, maybe do so.
- It should never be unclear what is your work and what is other people's work!
- (Your work may be the plural you, not necessarily yours individually, but, in that case, name those individuals as authors).
- Famous athletes get nervous before coming on the field!
- A bit of stress is good, too much is not.
- Learn how to manage/decrease your stress (personal).
- Watch out if you speed up/slow down when nervous.
- Rely on your preparation (breadcrumbs, initial statements, ...).
From >100s of peer feedback on student, presentation, by a large margin, the most common piece of feedback was: