Underwood says topic modeling is a way of inferring what discourses could have contributed to the creation of a set of texts. This idea relates to the notion of literacy (and literature) as a social practice (which is the perspective I’m using in my own research) and perhaps allows us to shift focus from the interpretation of individual standalone texts to the interpretation of various contexts, texts in context.
Underwood also mentions that the focus here is on probability (not “fact”) and that we should “reflect carefully about which assumptions we actually want to make.” Of course I’m all about examining assumptions, particularly the latent cultural and racial assumptions that we’re often oblivious too.
There was a moment, though, when I wished we’d (humanities, esp. Literature, folks) stop mythologizing the hard sciences. In my notes I quote Underwood: “This sort of thing is very much a critical judgement call; it’s not a science.” I then add the following commentary: [but science is also quite often a critical judgement call!!!].
After reading Underwood, I think Jockers offers a clear, interesting, and useful explanation and illustration of topic modeling, texts in context, and probability. Jockers actually highlights the probabilistic nature of topic modeling by giving us percentages of accuracy. I found this act of transparency important in a field with so much “hyperbolic rhetoric” about the saving powers vs. evils of DH. Having read multiple pieces from his book, I’m actually intrigued and want to follow his entire thread by reading the whole book.
A most interesting read, given my present projects, is Rhody’s piece about topic modeling with poetry. I think a crucial difference in how Rhody approached her poetic corpus with the way I approached mine is that I already knew “the topic.” I’m not really mining for topics, per se, but essentially looking at how poets choose to write about one particular topic (colorism). My question when using these various tools is How do poets represent X topic? What images are they using? What metaphors etc. do poets turn to when writing poetry on this predefined topic. So, I’m curious to see how the topic modeling will work given that I already know the topic I’m interested in and the topic that each of these poems is (supposedly) about, and that I’ll be reading the results with the question, What’s the relationship between these word clusters and the central theme that’s at the heart of the project? What do these “subtopics?” reveal about the overarching theme and how poets are pulling from various discourses (Underwood) to portray that theme?
Finally, to explain why I use the canary in a coal mine metaphor, this all resonates with the intersection of labor, value, risk, experimentation, and uncertainty involved in DH projects. Of course the stakes are exponentially lower when mining texts, but we have seen that text mining requires labor, that we hope for some value, and that the text mining process requires risk, experimentation, and uncertainty.
