I was asked by a young researcher, already a talented and engaging scientific communicator, what principles I keep in mind when preparing a talk. He specifically asked about the idea of telling ``short stories" about ongoing work. After citing the above essays by Geroch and Mermin, here's (a lightly edited version of) what I said:
The main lesson from both of the essays above is about having mercy on the audience. This is closely related to the central difficulty of teaching, which is:
A related principle is to think about what I would want to hear the speaker say and see on the slides if I were listening.
In that regard, one thing that's important to me when listening to a talk is the logical flow from one idea to the next. There should be some inevitability of the logical progression (and enough foreshadowing) that someone who is really following closely can guess what you are going to say next. I think this kind of talk is the most fun to listen to. Such a logical narrative flow also makes the talk much easier to give, i.e. it makes it easy to remember what you should say next. (The opposite extreme is a list of disconnected results.)
In your message, you mentioned the word `story' and I think this is exactly the right idea. The goal should be to tell a coherent story, and each thing you say and each slide you show should have a purpose toward telling that story. The bigger story can have sub-plots (like when advertising work in progress). Generally, in figuring out what is the story that needs to be told about some piece of work and how to tell it, I find that the most reliable strategy is just to tell the story of why I actually did the work in the first place, i.e. what led us to ask the question in the first place, even if it is a somewhat circuitous logical path. The honest motivation is often the best one.
Try not to put anything on the slides that you are not going to talk about, or that is not directly helpful for the story you want to tell. I don't always succeed at this goal, usually because there are some equations I feel like I need to show, even though I shouldn't. This can be OK because of the concept of ``pictures of equations". That is: the point being conveyed by showing the equations is not to explain what they say, but just to show that they exist, or maybe a little bit about their general form. I like to think that this is analogous to the fact that when experimenters give talks, they show pictures of their equipment. What does this accomplish for the large part of the audience that is not intimately familiar with such equipment? It shows that the equipment exists, how big it is, what kinds of things are involved.
Try to have a consistent color scheme. For example, the tex macro I use to put references always makes them green. I try to reserve red text for an apparent contradiction or for pointing out a problem.
Put things in boxes (both literal and figurative) and give them names whenever you can. The human brain is really good at compartmentalizing.
It's true that I try hard to start from the beginning when possible. It can be really useful, even for experts in a field, to reconsider the starting point of some subject. Often this requires simplifying the story, e.g. not talking about the most general case. If you can convey well the simplest example of an idea, the audience will always give you the benefit of the doubt about the general story. Doing this well requires a balance between the goal of complete honesty and full disclosure, and the goal of efficiency. Even the slightest extra information comes at a cost in the listener's attention. But of course one should never leave out information in a misleading way. [Here is a case study: in tomorrow's lecture I'm talking about density matrices in general (and discussing the von Neumann entropy). At some point I want to say that they are positive and therefore hermitian. In the general case, where the Hilbert space is infinite dimensional, this conclusion actually requires the operator to be bounded. There is some temptation to mention this and spend time and attention explaining this assumption. But I know that for many of the students it's much more urgent that I explain what we are going to do with a density matrix in the first place. So I'm not going to say anything about this assumption. To the more advanced students who may worry about this it will become clear what I'm assuming when I write down the spectral decomposition.] The key is to give the right idea, and it requires judgement to decide which are distracting technicalities and which are crucial assumptions. In a talk, one should default to concluding that something is a distracting technicality, much more so than when writing a paper.
I find it useful to write down beforehand what I want to say at the very beginning. That's a time when it's difficult to think clearly, but it's also the moment when the audience is paying the most attention, and deciding whether to continue paying attention. Many speakers waste these valuable moments thanking the organizers or saying something irrelevant. I think it's very important to start by stating as clearly as possible what the audience can hope to get from the talk. (Another important goal at the beginning is to acknowledge collaborators, but this can be done second.)
Another goal that's important to me is to, as much as possible, speak in complete sentences. It makes it much easier for the listener to process the information. At least this is true of me as a listener -- it makes me crazy when the speaker doesn't finish their sentence before starting the next one. Another goal is to not say unnecessary words or make unnecessary sounds, like "umm". These can be very distracting for the listener.
An important influence on my thinking about giving talks was Steve Shenker. When I first arrived as a postdoc at Stanford I gave a talk (about this paper). It was way too technical (even for a high energy theory audience) and Steve was kind enough to explain to me what I did wrong.