The convergence of the digital and the literary in research: an example
Over the last couple of days, I went to three conferences: two were very clearly on the humanities side of my interests, the first taking place in Berlin and concerning the relations between texts and images, the second one taking place in Mainz, and concerning Rousseau and his literary and intellectual consequences. The third conference was approaching the humanities from a clearly digital perspective, taking place in Leipzig and dealing with the construction of a digital infrastructure for language resources, tools and services.
It was a very exciting and intense week, and I was attending each of the conferences with a different set of mental questions which, however, all gravitated around the central question of how digital approaches and methods one the one hand, and typical research questions from literary studies on the other hand, could converge. Of course, this is a pretty fundamental question, but the clues I picked up at these three conferences are quite interesting, I thought.
First of all, at neither one of the “humanities” conferences did the digital play even a marginal role, while at the digital infrastructure conference, the actual use of resources and services for genuine research questions from the humanities, and the development of those services for such questions, was clearly and centrally articulated many times over. Infrastructure projects like CLARIN and DARIAH clearly see themselves, and rightly so, as places of convergence between the digital and the humanities, while in the humanities (and largely, to be fair, in information science), the interest of such a convergence is sometimes perceived but rarely acted upon. I’ll leave the question of how more specifically linguistic resources and tools can be relevant to literary research to another post. Here, I would like to comment on one area where the convergence of digital methods and literary research questions can be quite concretely imagined.
In Mainz, there was one type of question which came up again and again, in different forms: that of how phrases, ideas, the “spirit” or specific writing strategies developed or practised by Rousseau persisted, or were on the contrary opposed or subverted, by later literary figures, and how this relates to their respective historical, political and literary contexts. What relation is there between Rousseau’s massive autobiographical statement, the Confessions, and later autobiographical writings by Stendhal or George Sand in the age of industrialization and the realist novel? (Béatrice Didier spoke on this.) How does the relation between eros and polis, as Rousseau constructs it, resurface in different forms both in Latin American fiction and in French fiction, after the French Revolution? (this was Wolfgang Matzat‘s topic.) Which forms of reception of Rousseau can be distinguished in writings by Chateaubriand, at various stages of his literary career, and what do they signify? (Jean-Louis Haquette made very subtle comments on this.)
Of course, this type of research question is very widespread; beyond the case of Rousseau, research into the sources of the Encyclopédie or into the literary permanence of the Bible are just two further, massive examples. What seems to be common to such research questions seems to be that they imply at least three aspects or steps, each of them requiring a very different operationalization if taken up by digital approaches.
- The first aspect is that of intertextual relations between the initial reference corpus and the later corpus under investigation: given both corpora are available in suitable digital form (an issue in itself), literal uptake is (relatively) easy to spot, but allusions, paraphrases, implicit references, or the reuse of stylistic devices and the like, are much harder to find. Approaches like that of the eTraces project discussed at the CLARIN meeting in Leipzig are being developed for some of these questions, which I think is a good sign.
- The second aspect is that of the attitude with which such references are made: is Rousseau (or the Bible) admired, refuted, subverted, parodied, or misunderstood in such references? Here, approaches using methods like the identification of axiomatic collocations with key reference terms may yield results, but methodological work on this seems to be sparse (I have heard about some bibliometric work on such aspects, and I would be very glad if someone would point out some references to me).
- The third aspect is that of the historical, literary and/or biographical contexts which are used to construct a significance for the findings in the first two aspects. This activity, synthetic and hermeneutic by nature, and in practice ususally based on selection or even speculation, is without doubt the most challenging to support with current or even future computational approaches.
For a large part of the first two aspects, it seems that current computational approaches present some promising starts but have clear limits in what they can achieve. What this means, it seems to me, is that we need to acknowledge such limits, be they due to the current state of the art or more fundamental in nature. But we also need to conceptualize tools or services that best support our needs; for example, it would be really cool if we had tools which can be manipulated by the researchers in a way that would make them recursive, heuristic devices: in this way, a set of results to a query could be inspected very closely, each hit being evaluated as to its pertinence to the query, and this evaluation be used for resubmitting the query and create better results.
Another point is that there needs to be a dialectics between humanities research question and digitally-supported methods of analysis. For example, the eminent experts coming together in Mainz were working with an impressive breadth of references and immense insight into the history of ideas and of literature. However, their argumentative strategies necessarily rely either on close reading of selected passages, or on a synthetic or generalizing statements whose basis is sometimes hard to establish; computational approaches certainly change this, because minute computational evidence only gains argumentative strength when it forms larger patterns which can be visualized and given significance.
I’m pretty sure I will come back to quite a few of the issues I just quickly raised here; if you would like to share your point of view, in the meantime, or alert me to any research which you think is relevant, don’t hesitate to write back in the comments!
But let’s not outlaw either close readings or synthetic arguments. They are essentially humanistic and absolutely necessary. Language is too complex to be reduced merely to computation. Numbers are easy because their semantics is so straight-forward: no polysemy, no shades of meaning, no diachronic perspective. Words, on the other hand, are hard because they are more malleable than numbers, more elusive and certainly more seductive.
One way of acknowledging the limitations of computational methods is to say: they cannot and should not replace traditional readings of literature. They can offer different insights and, as you say, help us detect patterns, especially when dealing with texts on a large scale, but they shouldn’t be seen as a framework for proving or disproving close readings or traditional intellectual history. In fact, patterns don’t prove anything. They are merely patterns. Once you detect them, you still need to interpret them. That’s why computational methods are not the end of the game, but really just the beginning.
Thanks, Toma, for this comment. It is interesting to see that this issue of the limitations strikes a chord even with those of us who have high hopes concerning computational text analysis or digital humanities more generally. This certainly goes to show, as Fotis Jannidis recently said, that this current wave of digital humanities is, unlike an earlier one, very pragmatic and realistic in its expectations.
I do think that there could be a productive way also of connecting computational approaches and close reading practices: close reading may raise questions we then reformulate to operationalize them in a quantitative way, and the (preliminary) results we get very well bring us new questions that we will test by going back to individual pages and read them very closely; or, of course, we need to go back to apparently “interesting” passages which the pattering of our quantitative results pointed out, and see what is going on there, even simply to start making some sense of the pattern.
I like the way you formulate this, toward the end, as a “dialectics between humanities research questions and digitally-supported methods of analysis.”
Implicit in the idea of a “dialectic” is two-way exchange, and I think that’s vital here. As you acknowledge, a lot of existing research questions in the humanities may not be supported all that well by text analysis — unless we include keyword search itself as a form of “text analysis.”
On the other hand, the strengths of digital analysis may suggest ways of reframing inquiry that wouldn’t otherwise occur to us. In the case of the Rousseau example, for instance, our habit is to approach this as a question of “influence” or “parallel,” so we might be tempted to construct special tools that compare Rousseau’s autobiographical texts to other texts. And yes, that approach might work.
But another approach, perhaps more “idiomatic” to the digital domain, would simply be to map biographical discourse as a whole between 1750 and 1850 — using clustering or topic modeling or some other technique — and then ask what ends up being grouped with Rousseau. It’s tricky to do that across linguistic boundaries, of course … but I’m just speculating idly here. My broader point is that we need to do both things: adapt digital tools to existing humanistic questions, and think imaginatively about new, unasked questions that digital tools make it possible to ask.
Many thanks for that comment, I couldn’t agree more. You spell out what I failed to make explicit. I do think that the research questions will have to change with the methodologies, so that neither the literary part just gets reformulated in digital terms, nor the digital part hampered by literary-style questions, in a dialectical convergence thought of as a productive process oriented towards the future.
Another “transitional” phenomenon a colleague in Mainz pointed me to was his idea of using clues from existing research to identify references from one author to the other. I don’t actually think this will have a particularly good ratio of effort to result. However, what this idea actually points to, and this would again be an approach closer to the digital (even if less in a computational than in a web 2.0 sense), is to continually collect comments on relations between authors and somehow bring all those crowd-sourced annotations together. The question being, in this case, whether those of us interested in the aftermath of Rousseau actually form a sufficiently large “crowd”…