How to Do Things with Networks: A Response to Franco Moretti

The first post in the Magazine Modernisms Essay Club reading of Franco Moretti’s “Network Theory, Plot Analysis” comes from James Murphy, editor of this blog.  He is a recent Ph.D. from UC Berkeley’s English Department, where he wrote a dissertation on the theory and practice of modernist revision, and then joined the Committee on Degrees in History and Literature at Harvard University. He has published essays on Robert Frost and meter, Henry James and revision, editing and electronic archives, and teaching serialized novels.  He is currently working on a book entitled Modernist Economy, but don’t hold him to that title.

The publication of Franco Moretti’s essay  “Network Theory, Plot Analysis” as the Stanford Literary Lab’s second pamphlet is an exciting event for those of us interested in new approaches to the study of literature made possible and provoked by the digital.  Digitization hasn’t only provided unprecedented access to our literary heritage–it’s begun to redefine what we think that heritage is and asked us to find new ways and new tools to conceptualize and curate this growing horde of materials.  It’s on these last two fronts that network theory and social network analysis (SNA) come in, but more on them later.

It’s a real pleasure to respond to “Network Theory, Plot Analysis” and to look at how its insights might be expanded to modernism and periodical studies because Moretti has been an inspirational and validating figure in the relatively young fields of quantitative literary analysis and digital literary history since the publication of “Conjectures on World Literature” in 2000.  His Graphs, Maps, and Trees:  Abstract Models for a Literary History was a gateway for this reader and many others into thinking about (non-belletristic) literary history rather than literary criticism and the difference between the two.  He also argued more cogently than anyone else out there (at a moment when it wasn’t clear to the non-initiated that there was anyone else out there) that there could be real power in looking at large numbers of books, at titles, at production numbers, and other kinds of data, rather than, as I’d been trained, at the content of particular texts.

Subsequently, I learned that Moretti was not the first literary scholar to make this argument, but I have continued to feel the rush of reading Moretti for the past decade and a half or so in part because he’s so persistently reminded us that we literary critics do indeed actually have data and we need to learn how to dig into it.  Literary scholars are sometimes reluctant to acknowledge this fact because our métier is placing words like “data” and “facts” into question; we are awfully good at problematizing, but sometimes problematizing can be a problem in itself.  While skepticism is one of our discipline’s strengths, it comes at a price.  We unintentionally encourage the not infrequent charges of subjectivism and relativism made against literary scholarship if we deny that there are objective claims to be made about literature and literary texts, in addition to interpretations based on close reading.  We neglect salient aspects of literary history and literary texts that might not emerge through exegesis.  And, most important, we risk the  complacency of skepticism, every bit as dangerous as the complacency of certainty.  It’s probably better to work in an academic culture in which no one is ever 100% certain than to work in one in which everyone always is. But only probably, and probably not that much better.  Moretti helps us to remember to be skeptical about skepticism.

I should be clear here that I don’t think of close reading and distant reading (Moretti’s term for his approach) or literary criticism and literary history as an either/or.  I think that interpretation and close reading remain, will likely remain, and should remain for a long time the defining practice of literature departments.  It’s in the “should” that I differ from Moretti, who, perhaps in order to provoke, has been quite vocal about his disdain for close reading (see “Conjectures World Literature”), although he’s clarified his position recently.  There’s no need to choose between these approaches, not, at least, as a discipline, and the debate is an old one (At the risk of sounding antediluvian, I think Crane’s “History Versus Criticism in the Study of Literature.”  English Journal 24 (1935) is pretty good on the topic), but Moretti’s stridency has done more good than harm in that it has pushed some scholars to think about the limitations inherent in reading and explicating a fairly miniscule percentage of the poetry, drama, and fiction produced in the past 1000 years.

All of which is why I was excited to hear that Moretti had published an essay employing network theory, but rather surprised to find out that a guy known for studying Victorian and modernist literature, who declared that Victorian fiction “cannot be understood by stitching together separate bits of knowledge about individual cases” because the field is “a collective system, that should be grasped as such, as a whole,” was using it to study character and plot in Hamlet.  Why not continue the genre analysis of the Literary Lab’s first pamphlet?

I wonder whether Moretti felt obliged by the conventions of social network analysis (which is more what Moretti employs here, rather than network theory per se).  SNA was developed by sociologists in the fifties and sixties as a way to model the structure of social ties that exist among individuals (or individual units) and reveal the ways in which these social structures determine the identity and opportunities of individuals within that network. As digital tools have increased the efficiency and incisiveness of SNA, it has proven to be a powerful tool for determining how diseases spread, how people vote, and whether individuals are likely to become criminals.  A key component of this analysis has been its focus on large networks formed by hundreds or thousands of connections (edges in the terminology of SNA) among hundreds or thousands of agents (nodes).  It is, in other words, a natural fit for the kind of work Moretti has been pushing, and the growing number of SNA programs (I use Gephi, which is still in Beta mode) make it possible for humanists lacking mathematical and social science chops to get in there and try their hand at modeling  and visualizing large relational structures.  As modernist studies has moved away from the author-centered model of modernism promulgated by the New Critics and subsequent generations of scholars, and looked for new ways of thinking about context, consumption, and circulation, SNA and digital visualization present tantalizing possibilities for rethinking both the contexts and texts of modernism, including, I’ll suggest later, modes of publication (e.g., anthologies, magazines) relatively neglected by scholarship.

But back to Moretti . . . Why Hamlet?  Moretti himself suggests that the his lack of knowledge of statistics and his desire to work out the network by hand (Could we call this artisanal SNA?  Small batch rather than industrial production?) drove him to a small network.  That modesty is, of course, balanced by the bravado in tackling one of the most (over-)analyzed texts in world literature as the occasion for seeing whether SNA can tell us anything new about literature.  If SNA can provide insights into a play about which seemingly nothing original was left to be said, then what might it do for the rest of literature?  I suspect that the choice of Hamlet might be further motivated by Hamlet’s (speaking now of the character) importance in the mythology of the individual.  Even if Shakespeare didn’t intend him to be a proto-Romantic hero, he’s repeatedly been claimed as such (see Harold Bloom as the most yawn-inducing exemplar on this point).  SNA is one way to expose the limits and falsehoods of the concept of the autonomous individual.  SN graphs let us see the ties that constitute and constrain the kinds of individuals agents might be. And about Hamlet the text, can’t we say the same thing as we have about the character?  That the heroes of literary study–the works at the center of our canon–are also embedded in networks, that the identities and prestige of texts are also always relational, and that it is also these relations that determine what place texts have in literary history and study.

Hamlet is right at the center of  Moretti’s diagrams, just as we’d expect, but that might be just the point:  Hamlet is the center, yes, and that’s not all that interesting.  No graph was necessary to tell us what a title already has, but it’s the graphs that reveal what’s interesting and what we might not bother to think about precisely because Hamlet is the center of Hamlet. What’s interesting is how the other characters in the play fit into that network; what’s interesting is that Horatio is almost as central to the social network and yet not  central to our imagination of the play.  Who goes to Hamlet waiting to see how Horatio will be played?

Centrality is a fairly specific term in SNA and is typically tied to 3 aspects of nodes in networks:  degree, closeness, and betweenness.  Degree centrality measures how many edges a node shares other nodes in the network.  Think of it as how many people you know.  Hamlet is directly connected to 16 other nodes, while Horatio is linked to 12.  Closeness centrality is a measure of how efficiently one node can connect, directly or indirectly, to any other node in the circuit.  Think of it as how directly you can get in touch with other people in a network directly or through your direct contacts; the fewer intermediaries, the greater your closeness centrality.  In Moretti’s graph, Hamlet is separated by one-degree from 16 other nodes, and by two degrees from the other 13 nodes.  Horatio has 12 first-degree connections, 9 second-degree connections, and 8 third-degree connections.  Betweenness centrality is kind of like the other side of closeness.  It measures how many of the shortest paths between any two nodes pass through a particular node.  Think of it as how many people must pass through you to get in contact with other people.  Working with these parameters on Gephi, I generated a graph that looks a good deal like Moretti’s.

What Moretti shows in his diagrams is that network centrality ≠ narrative centrality.  He proposes that Horatio’s position in the Hamlet network “announces what will soon be called, not Court, but State,” since in his (and the Gephi diagram I made) visualization Horatio is linked to characters situated beyond Elsinore, and his flat style is a reflection of the discourse of bureaucracy.  Intriguing as this proposal is, I think that Horatio serves a different function, one with something of a tradition in drama, film, and fiction:  he is the bystander and thus a proxy for the reader.  Like the Greek chorus, Zelig, and Nick Carraway, he is there, standing witness to most of the important events in the narrative, even knitting those events together at times, often merely by witnessing and corroborating them, but the bystander can’t do or at least doesn’t do anything to affect those events.  Who is this bystander but a stand-in for the reader?  Are we not all Horatios, linking together the disparate elements of narratives, standing at the center of the story, watching as events more marvelous (and horrible) than anything we can hope (or fear) to experience take place before our very eyes?

So what can this kind of analysis tell us about modernism and magazines in particular?  The most obvious application of Moretti’s work with Hamlet would be to apply it to modernist texts.  What would the SN diagram of character ties in Ulysses look like? Or Mrs. Dalloway? À la recherche du temps perdu?!?  And what would these diagrams reveal? At the very least, this work would likely take a good deal effort more than tracking the 30 characters in Hamlet; it’s an occasion for some facility with text mining.

We needn’t limit ourselves to characters in novels, however.  What about words and phrases, ideas, scenarios, places, time periods, etc.?  Jeff Drouin’s done some intriguing work with topic modeling, SNA, and Proust over at the Ecclesiastical Proust Archive.  And, of course, we needn’t limit ourselves to imaginary people or the insides of books either.  Mapping the literary sphere beyond the content of solitary works–the kind of work I’d expect Moretti to be doing–avoids the problem of intentionality that structural criticism of single works always comes in for because it looks for the structures that are coincident with and determinative of networks that exist beyond the reach of any author’s intention.  It replaces intention with emergence as the principle underlying form.

I work with SNA as a way to examine author networks as they are formed through anthologies and magazines.  I’m not exactly interested in coterie publication, if by that we mean the ways that authors created cliques within print culture.  Worthy as that topic is, if for no other reason than to test the claims of those who insist on the pursuit of exclusivity/elitism by, say, the Men of 1914 or Bloomsbury, I’m much more curious about the ways networks worked as ways for readers to identify writers within the literary field.  My hypothesis is that knowing how to recognize a modernist when you saw one depended, in part, on who he or she was published alongside, as in this diagram of D. H. Lawrence’s appearances in fiction anthologies.  Notice all the major figures of modernism there?  A. E. Coppard?  Oliver Onions?  Could they have something to do with DHL’s precarious place within the modernist canon?  SNA is a great way to begin answering that question.