Over the past year I’ve been working on a number of things. One exciting piece of news is that my book is finally coming out (yay!), another is that I’ve been working as a Postdoc at McMaster on a Digital Humanities project focused on Austin Clarke. I’ve been treating this postdoc as an opportunity to stumble around, think about, and experiment in digital humanities. From both technical & hermeneutic perspectives, in what ways does DH transform our notions of reading, writing and canonicity? Big questions…
Clarke provides a useful test case for thinking about DH not only because his archives are so extensive but also because he writes about humanities, humanism and who gets to count as human in a broad sense. Clarke writes the slave past within the multicultural present – he shows how the social constructs of the colony and the plantation continue to structure life for black people even in post-colonial (post-national?) Canada. In this sense his work brings to bear the postcolonial critique of the black Atlantic onto the discourse of digital humanities. Whose humanities? Whose humanism? This topic – of the latent forms of humanism that lie at the heart of DH – has been raised by Tara McPherson but requires far more discussion and I’m hoping that Clarke’s work will open new ways of thinking about the category of the human within DH.
At the same time Clarke’s work provides a compelling case for testing the technical methods of digital humanities. His archives span 23m of boxed material! How else could you ‘read’ these archives other than by using digital methods!?!?
One of my first challenges in working in Clarke’s archives was to engage in a topic modelling analysis of his work. This involves scanning in his manuscripts — drafts and final editions — and running them through a topic modeller. I actually ended up using MALLET as well as a custom topic modeller built in Python but I’ll save that discussion for another time.
While my topic modelling experiment isn’t done it has raised a number of interesting questions about Clarke’s writing and his place in the Canadian canon as well as about the technical dimension of topic modelling work. I’ve noticed in my preliminary work that Clarke’s topics leak into one another pretty substantially and produce a number of Chimera topics. Instead of neat, discrete, and meaningful topics I am seeing a large number of blended or ‘leaky’ topics. Now I can tune some of the variables in MALLET and change my number of topics to try and generate clearer topics. In another sense though, perhaps this generation of chimera topics is pointing towards something else in Clarke’s writing.
I’ve long suspected that the inability of Canadian critics to place Clarke within the CanLit canon and the relative dearth of critical commentary on Clarke’s work is an effect of his creolization, his blending of the symbols, languages, iconography, and themes of Canada and the Caribbean. Particularly in the heady days of CanLit nationalism, Clarke’s work simply didn’t fit the bill of what constituted Canadian writing. This is partially the failure of critics relying on a narrow definition of Canadian themes and writing. Additionally, however, Clarke doesn’t write in a Caribbean idiom nor does he focus strictly on Caribbean or Canadian themes (whatever those might be) but instead practices a kind excessive creolization — a creolization that doesn’t just combine multiple symbolic registers but renders them inextricable from one another.
If Clarke practices creolization — what I think of as a kind of aesthetics of crossing — then perhaps that works against the intuition of topic modelling. Where topic modelling wants to ‘discover’ topics in a corpus, Clarke’s aesthetic project works in just the opposite way: by blending together seemingly discrete images, themes, languages, and discoruses. Perhaps the chimera topics that I’m discovering as I do this topic modelling are actually evidence of Clarke’s creolization. In this sense, I’m learning more about Clarke’s writing through the failure of topic modelling to find the discrete topics in Clarke’s writing than I am through a ‘successful’ topic modelling.
Sure, I could try and reduce these chimerical topics through a number of measures but I think doing so would be fixing the results of my experiment somewhat. I think this raises the question of how we link distance reading or data analysis to our close reading and our hermeneutic practices — to what degree are we just fixing the data to give us the results we seek? Yes discrete topics would give me intelligible data to work with but would it tell me something meaningful about Clarke’s writing? Would it be more useful to Clarke’s writing than my discovery of these chimerical topics?
These are the kinds of questions I’m continuing to wrestle with as I work in Clarke’s archives and think about topic modelling as a means to understanding his work and his place in the Canadian canon.