How to give a good talk, an accessible talk, and an interesting talk
This post is intended to prompt me, and hopefully you, to think about two things. The first is practical: what constitutes a good style for an academic talk? The second is more abstract: what do the advances in computing speed and power offer to psychology conceived in it’s original form as a science of subjectivity?
I think an excellent starting point for launching both of these discussion is a recent talk at the Institute for Advanced Studies by Christopher Manning of Stanford.
From a purely practical point, many of his slides were solely illustrations. Bullet points and printed repetitions of what he was saying were few and far between. There were plenty of words on his slides, but then his content was computational linguistics - hard to avoid words as illustrations in a talk on language. But when words did appear they were illustrations, not redundant echoes of what he was saying. Manning used a clean white background with little embellishment. Colors, fonts, and changes in arrangements were used sparingly and selectively to group and highlight. In short, I think the formatting of the graphical material in support of the talk was an excellent example of how things should be done. The point is not that we should grab the slide template for use in our own talks, but that we should consider what we are going to show with the same care that we decide what we are going to say, and that just as we use inflection and gesture to enrich the semantics of our speech, we should view the visual opportunities of modern graphics as providing us the same opportunities in the selection and formatting of our visual materials.
Another exemplary feature of this talk is the structure of how information was unfurled. He did not tell us everything interesting he knows or has done, but he edited himself scrupulously to present a single coherent thread from beginning to end. We began at the beginning, and we were led on an apparently seamless trail to the current state of the art. Lots of potential detours were omitted. Not because they were not interesting or important, but because they were not in service of getting us smoothly to our information destination. Thus, this talk is also a good example in plotting a presentation.
Now what about the second goal: telling us how psychological science can benefit from the technical progress in computing speed and power? There things fall short, but that may not be Manning’s fault. There are limits of time, attention, and topic. The goal here was to share some of the accomplishments and potential of computational linguistics, and that is what the talk did. But, given the stylistic and organizational talent, combined with Manning’s position, history, and perspective on the field, one can only hope that he gives the continuation of his talk where he evaluates the contribution of all these practical successes to psychology as a science. It does indeed seem like magic that optimizing a set of vectors (of an arbitrary and comparatively small (100) dimensionality) for the components that probability that they precede and follow other words yields such a high quality result when applied to machine translation. No semantics necessary it seems to say. No knowledge of the world required. Does this empirical result, existentially dependent on computing power, tell us something about the human mind or not? If not, is that just an artifact of the specific application or a more general fault that limits our ability to gain insight into the language of thought by post-hoc reasoning from incremental improvements in Siri, Echo, or Alexa? Hopefully, some day soon Manning will use his rhetorical and graphical talents to shine some light on this knotty problem.