The topic is word clouds (tag clouds, Wordle). I think this new chart type (circa 2000s) has a lot of potential. Software like Wordle has made it extremely popular; Wordle is amazingly easy to use. It is most often used to summarize text documents, such as speeches. This usage has turned it mainstream but should not restrict our imagination of other use cases.
However, the developments since 2005 when I first wrote about word clouds have been disappointing. Wordle has made it easy to turn out chartjunk (crazy colors, "dis" orientation, etc.). In this presentation, I lay out a set of improvements that can help realize the potential of this chart type. They are all based on statistical principles, which is the underlying theme of the talk: that when we design charts, even the apparently artistic decisions can be made by appealing to logical or scientific or statistical concepts.
There were a lot of questions, and I couldn't get to all of them. Feel free to continue the conversation here in the comments.