Hundreds, Thousands or Millions of Words: Fitting Our Methods to the Question

Many humanists rarely used numbers before discovering algorithms (like topic modeling) that are suited to exploring enormous libraries. As a result, their conversations about method tend to be organized by a strong opposition between “close” and “distant” approaches, aimed at radically different scales. But interesting things can be done at every scale of analysis — from a single passage, to dozens or hundreds of texts, to millions of volumes. In this talk, Ted Underwood will briefly survey that range of scales before dwelling on some approaches suited to the middle of the spectrum. Here scholars rarely have open-ended exploratory goals. Instead they begin with a loosely formulated theme, traced in a few examples, and need a way to scale up the question. In a situation like this, it can be valuable to keep human readers in the loop: we are, after all, very good at reading with a specific end in view. But algorithmic models also have unique strengths, catching certain kinds of patterns we would miss. The talk will describe ways of pairing human expertise with algorithmic flexibility to address questions about genre, gender, and the representation of time in fiction.

This event is co-sponsored by Green College, Early Modern Conversions Project (McGill University) and the Cultural Evolution of Religion Consortium (UBC)