This analysis will focus on “Quiet Transformations: A Topic Model of Literary Studies Journals.” It is a project undertaken by Andrew Goldstone and Ted Underwood in an effort to provide a new perspective on literary history. From the project website: “A topic model is a novel form of evidence for literary history; exploration is, we think, a good way to develop a sense for what kind of evidence it is and what sort of interpretation it permits.” Topic modelling is the extension of an associative algorithm into a body of text for the purposes of marking common and probabilistic associations within the text. “Quiet Transformations” extends this method over “seven literary studies journals over the 1889–2013 period into 150 “topics.”’ The interface is broken down by year and can be made to show articles from the entire spectrum of the model or from single years chosen by the user. The front page of the site also offers suggested topics that appear to be particularly meaningful to the project itself. For example, the following link leads to topic 80, which is marked as associated with the words “power violence fear blood”:
I offer this sample topic as a way into understanding the interpretive powers and choices that go into and come out of this project. As one can see by manipulating the interface in the link above, simply picking a year to filter is an acknowledgement of expectation. One might look at 1985 or 2001 in anticipation of tracts about violence in the Middle East or the politics of domination and hegemonic superpowers. One could also select 1929 or 2008 to see the ways in which power discourses bore on the financial crises of those years. That one can do these things reveals both the tendency to historicize by exceptional event (at least that was where my mind went first) and the ability of the topic model to disappoint that urge. One might point to the modelled texts and question whether journals of literary criticism would be concerned with these cultural events. This, too, would be to make an interpretive choice and to be the subject of such a choice on the part of “Quiet Transformations.” To elaborate, by selecting for these journals, Underwood and Goldstone propose to view literary history in a different light than is usually applied. Singular events and texts are often used to historicize the progression of literary studies in western culture. Exemplary and polarizing nodes are generally thought to guide the way through the web of history. Topic modelling is helpful because it is a method unburdened by the assumptions that cause me to select for singular years and expect something to happen. It is completely random and unbiased (as a method). It is also not definitive or an active force in making claims about the nature of the modelled texts. It is merely a new way to view the information available.
The traditional way of understanding documents like these is to trace the discourses they are a part of. As is the case with many digital humanities projects, it would be impossible for a person to actually read and understand all of the texts in the model. What one might do, however, is come to be familiar with the works of a particular critic and the citation network that critic draws upon. For example, one might search a journal for posthuman scholars and find that N. Katherine Hayles is a prolific and well-cited author in that discipline. One could then read her contributions and note the scholars she cites and the scholars that cite her. It is not difficult to see how centralizing and limited such a method can be. That is not to disparage traditional research practices (I make great use of them myself) or to say that topic modelling is a superior way to do research. I merely want to point out that what Underwood and Goldstone contribute in “Quiet Transformations” is the means to see texts that one might never have read and to make new associations that were unavailable in a smaller and more focused research scheme. To be more specific, in topic 43, “game Cervantes play Quixote,” I looked at 2003. My expectation was to find a lot of work on Don Quixote and the critics that work on texts in and around that one. However, I instead saw this:
There are texts about subjects ranging from artificial life to medieval drama. In asking why such diverse and seemingly-tangential texts appear in this topic, one can begin to get a better understanding of the words modelled and the larger significance of their interplay for this particular year. This is even more effective if taken across years and different topics. Moving up one topic to 44, I searched the same year and began to see some common threads developing. Texts within these topics are concerned with confinement and artificiality more than anything else. Too, despite topic 43 not featuring any tokens concerning drama, articles about dramatic texts still abound. Here the topic model serves to jump start a research question (why are all these things so important in 2003 in literary studies) rather than make definitive conclusions about the texts. This is, for me, the most important function of “Quiet Transformations.” One can use the topics and the chronological ordering thereof to start asking questions about literary theory that are not being asked.
What this project is not is a silver bullet for humanities research. It does not solve previously established problems nor does it refute presently accepted theories on the historicity of texts. What it does do is challenge the theoretical apparatuses that centralize and focalize. There is a wide world of interpretive work to be done on this model simply because it provides a view that was unavailable before. Stephen Greenblatt (just an example) might look at this collection of topics and use it to organize the next Norton Anthology because it gets down to texts that are traditionally ignored or considered unimportant. “Quiet Transformations” could lead to a more robust education on humanities theories and practices. It has the potential to open up new research questions and expose new reasoning. But it is limited. Because really it would be literary theorists doing these things, not topic models. This is perhaps the most important thing to take away from this analysis. Thinkers are still a vital part of this equation.