With his work Graphs, Maps, Trees. Abstract Models for Literary History (2005) Franco Moretti became one of the most prolific scholars of quantitative literary studies. The idea to incorporate methods and processes of empirical and mathematical sciences into the analysis of literary texts is extended via the Literary Lab, founded in Stanford in 2010: It applies the functioning of the laboratory as a model for humanistic production. Consequently, the reproach of understanding literature in a positivistic approach arises. However, as this conversation with Moretti shows, graphs, maps and trees, which visualize the statistic data, are not the result of positivistic measuring. They are not the aim, not the end of analysis, but they depict that confusion, facilitating a new perspective of the rigid schematic order of the fleeting literary.
Fabian Goppelsröder: Franco, in 1998 you published a book called Atlas of the European Novel 1800–1900. Publishing an »atlas« is not precisely what you would expect from a literary critic. In your introduction you mention those who asked you why the hell you would like to focus your work on maps and, as if that was not enough, not on already-existing maps but on the making of new maps!
Thus you start the book with a short explanation of the idea behind your project:
»An atlas of the novel. Behind these words lies a very simple idea: that geography is not an inert container, is not a box, where cultural history ›happens‹, but an active force, that pervades the literary field and shapes it in depth. Making the connection between geography and literature explicit, then – mapping it: because a map is precisely that, a connection made visible – will allow us to see some significant relationships that have so far escaped us«.1
So, what exactly does a map show that would otherwise without it would pass unnoticed? And how does a map do that?
Franco Moretti: I have to begin with a few words about the prehistory to the idea of the Atlas. A few years before the actual project I was asked to write an essay on European literature for one of these huge histories of Europe appearing at Einaudi. And while working on that I remember reading or re-reading Fernand Braudel, and I found this passage in the book on the Mediterranean where he is talking about the Renaissance and the Baroque. He basically says that we always talk about the enormous influence the Renaissance had, while in fact it is the Baroque that spreads more widely. And then he adds in a footnote that this is just a hunch, because we don’t have atlases of art history. This made me think, and I eventually agreed that such maps might be useful. The idea for the Atlas came up. At first it was supposed to be a collective work. But then the development was such that I finally had to do it on my own. Be that as it may, my initial thoughts were to map literature’s moving in space. This proved to be extremely time-consuming, it was very difficult to gather the data, etc., and so, even though the third chapter of the book is still that, I had to redo the basic design of this work, shifting my perspective from literature in space to space in literature.
At this point I come to your question: What does a map show and how does it show something that we would not have seen otherwise? This is a question that back then I thought would have a very intuitive and immediate answer and that I realize now, almost twenty years later, is the crucial question that kept resurfacing about graphs, maps, trees, about the literary lap, etc.
First: How does a map show? Well, in order to show anything, a map has to show its units. You can have a map of Brandenburg without cities but with elevation, rivers, etc. Or you can have a map with cities but without roads or with roads but without railways, etc. The crucial thing is that in order to produce any map you have to choose your units, and you have to be very clear about this choice. The choice of the units in the Atlas was very empirical, which means I really learned by doing. They are not always the same units at all. The first few maps, the Austen maps, are rather narratological; they are about beginnings, middles, and endings. The moment you get to Balzac and Dickens, they are more about where characters live in a city. This means that also the choice of the unit varies from case to case. In part it varies because I had no template, no model for it. So maybe the variance is also a sign of my uncertainty. But in part the variance is really a sign of the difference between the various genres. In an Austen novel endings have a role that they don’t have in Balzac. So there is a formal reason to map one aspect of the novel rather than another.
But how does a map show? It shows by linking certain narrative aspects of a text or a genre to certain spaces. That’s how it goes. Again: which traits, which aspects of the text or genre, depends on the researcher. This initial choice is always a hypothesis. I thought that in Russian novels the thing I could map was the statements that characters make that connect a certain idea to a certain place – certain beliefs to Russia, certain science to Germany, etc. This can be a good map or not, but while it is plausible for novels of ideas, it would be absurd for Austen or Conrad. Thus the initial choice is subjective; then there is an objective moment. Once you choose that you are going to chart where characters live in Balzac, anyone, more or less, will reach the same results. Then comes the third moment, which is the moment where we have to ask ourselves: Does this really show something that we did not know before? And this is, of course, the big question hanging over all we do. In the literary lab at Stanford we have a new series called ›So what?‹ where people doing research in digital humanities are invited to come and present their work. And in the final part of the presentation they have to explicitly address the question of what difference their work actually makes. This question can be asked in a nice way, but it can also be asked in a very nasty way.
FG: Of course everybody has to ask himself this question, not only in the digital humanities. The use of interpretation is as questionable as the use of quantitative analysis.
FM: Sure. But usually a new method has to answer this question more stringently. You did not exist before. Why should you exist?
FG: Right, because you don’t ask for the use of what you are used to …
FM: So, what did the maps show that we did not see before? I think that a new method, unless it’s a complete Copernican revolution, will always simultaneously confirm part of what we know and add something else. If it did not confirm something that we knew, it would be tantamount to saying that we knew nothing. So the results are always a mix of the already known with something new. This new knowledge, however, is also not just coming out of nothing. It’s something that had been there before. Implicitly. Thus a method that brings about new knowledge is basically a technique to make something implicit explicit, something invisible visible. In the Atlas, I think the best example of this is the punch line of the second chapter. It’s a section entitled »Stories of the Third«, in which the visualization of city movements and of the city as a field-of-forces shows that nineteenth-century city novels constructed a narrative system with three poles. All of a sudden the point is no longer a confrontation of two realms (this was Propp’s great metaphor) – it’s really triangulation. Was this new? It seemed new considering how strong the binary models of narrative analysis had been. Did I completely develop the new model? No. The results of this new method, let’s call it ›mapping‹ for now, have an empirical component, the visual artifact which is produced. Its pay-off, however, lies in the interaction between the empirical component and the already-existing concepts.
I may misremember this, but I think that I discuss this finding of the third to a large extent in footnotes. As if to say ›it’s there, but it’s not important‹. Today I would place it squarely in the text. It is the important thing. And this is also the so-what question. The so-what question has to be answered on two levels: one is the material level, the maps, but then there is the second so-what, which is conceptual. And in this case one has to establish an intelligible relationship with the previous theory. In the case of the ›third‹, the relationship is a polemical relationship. In other cases it may not be as polemical.
FG: Let’s stay with the fundamental questions for a bit. Braudel’s idea of a map is a specific idea of a map. It’s a geographical map as we know it, only with unconventional units, as you may say. It made you think about maps tracing the movement of literature in space. It’s also an economic map, showing where novels and perhaps other literary products were sold and where they were not. Is this being tied to specific geographic regions what you call the »ortgebundene«, place-bound nature of literary forms?
FG: But then you want to go beyond Braudel’s maps. You claim that maps can bring out the internal logic of literary forms. And this is, of course, a different matter. The internal logic of a novel appears to be completely independent of the book’s economically steered movement in the real world. And yet, it seems that for you the connection between a character’s behavior in a Dickens novel or a Balzac story and his coming from this or that part of the city, the East or the West End, the Rive Gauche or the Rive Droite, also suggests that the material story of a book in the real world and its inherent logic are not as separate as one may think at first glance.
Would you agree that Braudel’s want for somehow traditionally geographical maps in a new field, the aesthetic field, fascinated you as such, but also made you aware of the more general possibilities of maps? That they are able to visualize much more than just geographical relationships? Would you agree that through Braudel you got interested in maps, but that your use of maps slowly turned them into a tool capable of visualizing any connection?
FM: True. The project of the Atlas of the Novel started as a graduate seminar. I wanted to do the whole work in the spirit of Braudel. But then it became clear that this work is extremely time-consuming, and that it would not work out with the students. So I started to do the space in literature thing just to keep the seminar afloat. Only later I realized the actual potential of this approach. Where did the potential lay? It lay in the fact that the map forced me to be explicit about the criteria of analysis. Choosing units means that you have to be explicit about what you want to isolate in the text at hand. I basically located the value of maps in their being a tool to figure out the internal logic of a genre.
Some maps show the internal logic of the form and some the external, socio-historical pressure. The simplest case, but it’s also banal, is Sherlock Holmes. In Sherlock Holmes we have crimes all over London, but not quite all over London. It stops where the Jack the Ripper Crimes begin. It would have been absurd to send the master detective into an area of unsolved crimes. So here you see how history creates a boundary to the logic of a story that cannot be transgressed. A less banal and potentially more interesting case was in Graphs, Maps, Trees in the map section where I chart village stories and then, you know, simply by showing where the various stories take place, you can see the village disintegrating under the pressure of urbanization. So there you really get a sense not just of literary form, but of literary form within a social structure. Ideally this is the kind of map that I think would tell us most.
FG: So let me rephrase my earlier question. I think that your elaboration on the beginnings of your own interest in maps showed an important modification or development of your work very nicely: starting out with Braudel’s rather classically geographical maps, you seemed to realize the power of maps as a general tool of visualization enabling a new way of dealing critically with literature. You started out with maps as just an interesting gimmick within a more or less traditional literary history and you ended up with maps as basically a figure of thought introducing and carrying a new way of thinking about literature. This is interesting. And this also makes it somehow proximate that the book after the Atlas was the already mentioned Graphs, Maps, Trees.
Would you agree that with the Atlas something like the epistemological (or heuristic) value of visual order called your attention and that this epistemological value of visual order sort of haunted your work ever after?
Since then, visualizations, visual arrangements of literary data allowing readers to see a principle of order underneath a full conceptual architecture, play an important role within your research. The English – and also the German – title of your book Graphs, Maps, Trees: Abstract Models for a Literary History somehow water down the double provocation that the Italian original still holds: La letteratura vista da lontano. Not only is the dominant methodical paradigm of literary criticism attacked – instead of a ›close reading‹ looking for the smallest nuances you demand a ›distant reading‹ based on a quantitative analysis of whole populations of literary texts – there is also another interesting shift in the researcher’s attitude toward his object: instead of immediate interpretation, he first has to »see«, he has to perceive it as it appears from a distance. What exactly does that mean?
FM: First of all: yes, La letteratura vista da lontano I liked better as a title, but it could not really be translated into English and German. It has been kept in some Romance languages, but in English and German we did not find a working translation. Ok, what does ›seeing‹ in this context mean? Let me begin with an oblique answer to that question: For quite a while now I’ve been doing these visualizations as part of my work at Stanford. And I do not only use them for research purposes; I also use them in class or, of course, in presentations, etc. I am always amazed how bad people are at describing what they see. I mean there is this slogan that we are living in a culture of images, etc., etc. – and yet, there is an illiteracy in reading images which I find stunning. Perhaps this happens precisely because we live in a culture of the image, they are around and we take them for granted. For example: I have a chart about what happens to the length of sentences in the nineteenth century; we had an enormous discussion about what this chart shows. We could not even agree if there was a trend or there wasn’t. This is the situation. It’s so difficult to say what one sees and people are just not able to do it. Even books or articles that use graphs and other visualizations are very reluctant to really describe what they see in these graphs. Powerpoint presentations are the worst. For me, visualization means to put up an object of study, which is important not primarily as a result, but as the beginning of a new work of analysis. You create an object that, as Pomian says, no one has ever seen and no one could ever see. So it’s important to acknowledge these new objects by describing what they look like. What are their dimensions, their shapes, etc.
There is a subjective moment in which you choose your units, and then there is the process of gathering data and presenting it – that’s not subjective. It’s rule-bound; every researcher would reach more or less the same results. Once you have it there, seeing is again a subjective moment. You have to interpret or to explain the graph, the map, the tree. Interpretation means a transformation of the evidence, turning a story into another story. Explanation is different. It is digging out the force behind the chart that has produced it. Its inner mechanism, so to speak. Explanation is a causal or a functional explanation. I am not sure yet what a visualization is calling for. It seems that at times the one thing happens, at times the other.
FG: Probably you cannot even keep them absolutely separate. An interpretation at one point or another most likely will have to be explanatory and the other way around.
FM: Sure. The important thing is: The choice between explanation and interpretation is again subjective …
So if you ask me what you see when you see a chart, I have to answer: you see the result of a rather objective work. But what happens then is again very hypothetical, conjectural, subjective.
FG: This is an important clarification of how you conceive your own work. The polemics against traditional literary criticism pervading your writing made many people think you would like to erase and eliminate any subjective moment within it in favor of some kind of positivist quantitative analysis, of a statistical measurement of literature. Now you made it absolutely clear: the elimination of all subjective moments is not only impossible, it would end not just traditional but literary criticism altogether.
FM: Let me give you one short example. I think the first graph in Graphs, Maps, Trees is the rise of the novel in four or five different countries. Assuming that the data are right and that the comparison is plausible, what caused that development? Among historians of the book there are basically two explanations: One that says that novels increased because the number of readers increased. The other says that novels increased because readers wanted to read novels instead of other things. The number of readers did not increase but their interest changed. These are basically two opposite explanations of the same phenomenon. I cannot remember a hundred percent, but I think it was Popper, in reaction to Nietzsche’s claiming more or less that science reduces the unknown to the known, who said that it is in fact the other way around: Science starts with the known – the visualization in our case. What is unknown is what comes out of it … I see you feel uncomfortable with this …
FG: Well, yes and no. But let’s have a look at the Sherlock Holmes example from Graphs, Maps, Trees. In this case the interesting starting point is that there were hundreds of different authors who wrote detective stories around 1900 but no one was as successful as Arthur Conan Doyle. His stories outdid all the others. How come? This probably is the moment Popper is speaking about. You have a fact, the known, and now the scientist starts to search the unknown in it.
So now you had to come up with a hunch as some kind of filter that allows you to collect data. This is the subjective moment at the beginning of your work we were talking about. In this case your hunch was that Conan Doyle’s practice of putting clues at specific places in his stories would make the difference. Could you just quickly elaborate that a little bit?
FM: Sure. Incidentally this is a very good case. Of course there had been clues in other detective stories. And of course I was not the first who became aware of them. I just made the connection between the specific strategy of putting clues in Sherlock Holmes and Conan Doyle’s historical success. I started making these trees – if clues were present or not, if they were necessary or not, etc. – and then I looked at them, and what I saw was what I just gave you as the punch line of this study: that clues were the strongest single factor explaining why Holmes outdid the others. What is surprising, however, is the fact that no other detective story author was able to imitate Conan Doyle successfully and that even some Sherlock Holmes stories did not use clues at every moment where they could or perhaps should have used them. So in a way the deviation is as important as the positive result. This probably is why I agreed with Popper that science does not simply reduce the unknown to the known but starts with the known and, even though it produces knowledge, leads more and more toward the unknown …
FG: This is a rather surprising claim for someone whose mission is to push statistical and quantitative work in a field where people usually react aversively when it comes to such ›positivistic‹ methods.
In 2010 you, together with Mathew Jockers, founded the Stanford Literary Lab that prolongs and intensifies the research you started with the Atlas and continued with Graphs, Maps, Trees as a laboratory to discuss, design and pursue literary research of a digital and quantitative nature.
The projects pursued here comprise a study on the fictional geography of English and American nineteenth-century novels, a project on the stylistics of the novelistic sentence and a network-theory analysis of dramatic structure in a corpus of dramas from antiquity to our days.
Could you tell me more about the work at the lab? Since we have talked about the Atlas, would you elaborate a little bit about what exactly you are doing in the project on the stylistics of the novelistic sentence and the network-theory analysis of dramatic structure?
FM: Well, the first publication we did at the lab was an article we sent to New Literary History, not least because they had asked me for something. So we proposed them our text – and they rejected it! So we had to publish it ourselves, and we did this as the first pamphlet of the literary lab. You can find it on our website. In this first collaborative project, we used word frequencies in order to differentiate several literary genres. We usually said that we were working on style. But then, about two years ago, we asked ourselves: was it really style that we were dealing with? And most of us agreed that it was not. That we mistook something that is not style for style. The percentage of definite and indefinite articles, third person and first person pronouns allowed us to differentiate genres reasonably well, but it was language on such an elementary level that you could not really call it ›style‹. The units with which we worked were too small. So we kept moving up, expanding the units to sentences with a simple structure and then with a more complex structure. In an article I found the formula ›The x of y‹ – ›The castle of Otranto‹ – in Gothic titles. This was better, but it was not enough. Marissa Gemma, one of the members of the lab, in one chapter of her dissertation on Edgar Allan Poe, found a formula that Poe uses with insane frequency: ›The x of y of z‹, ›The Fall of the House of Usher‹. Was this style? It was at least interesting because it was not functional. Sarah Allison, another lab member, found a sentence type that is frequent in Eliot, e.g., the opening sentence of Middlemarch: »Miss Brooke had that kind of beauty which seems to be thrown into relief by poor dress.« It’s a narrative sentence (›Miss Brooke had that kind of beauty‹), then there is a relative pronoun (›which‹), then it shifts to the present and to an essayistic writing (›which seems to be thrown into relief by poor dress‹). This seemed to be so idiosyncratic that we thought: ok, here we have something that may be called ›style‹. So we decided to study sentences. We tried to do that on a quantitative basis and it became very complicated. I will not go into the details right now. But in the end we are pretty happy with what we have. We got some results that classical stylistics, moving freely from the word to the sentence and the paragraph, did not get. Our findings have not changed the panorama of classical stylistics, but they have established a new unit to work with.
FG: Thus you are not looking for the one structure that defines the style of the novel, but you are looking for different ideas of the specific style of this and that author and how you can compare these different styles. For this work you need a steady unit and this unit is the sentence.
FM: Yes. We are looking for the structural characteristics of a specific style at the scale of the sentence.
FG: So the focus on the novel does not come out of the assumption that there is something that all the different novels stylistically have in common, because your idea of style is linked to the author – it’s personal, right?
FM: Well, yes, it seems like that.
FG: But still there is a difference between the author of a novel and the author of poems or dramas …
FM: Well, style is linked to authors but not completely, e.g., the use of verb forms separates genres reasonably well. The progressive forms and modals for example are very good indicators of the Bildungsroman vis à vis other types of novelistic genres. Every Bildungsroman needs progressive tenses because a lot of it has to do with processes, with something that’s becoming but not crystallized yet. The Bildungsroman also needs to use a lot of modals because modals have to do with the interiorization of social norms, of social expectations, with the epistemological uncertainty that defines youth, etc. What is striking, however, is when these two things come together (as they do in Eliot, for instance). The genre needs the two things. It does not need them together in the same sentence. But that’s exactly what we found. We found elements that the genre needed and so they were functional, but their coexistence in the same sentence was not functionally necessary. And here it was, our initial definition of style: Something which is not dysfunctional but also not necessary. Something the genre wants but that also adds something more. So it’s personal but not only personal. It’s a personal component within a field of possibilities defined by the genre.
This is our style project. As for the network-analysis-of-drama project, there we are trying to think more in terms of plot. But a plot that is conveyed through words. Dramatic action is almost always echoed or doubled in words. There are exceptions, of course. The first instance of action happening on stage and not being echoed in words I can think of is Posa in Don Carlos: »Geben Sie Gedankenfreiheit!« and then he kneels in front of the king. If you did not have the stage direction, the words would not give you a clue. Whereas, when Hamlet kills the king, he does not only do the deed but also says the words. The action is not kept implicit, but is immediately made explicit.
FG: Let’s linger for a moment on the basic idea of this network-theory approach to drama. Again, you have to start your work looking for a basic unit structuring your visualization, your map or rather your diagram. In this case the basic unit is defined as explicit interaction between two people. That two people talk to each other is marked as a straight line between them. Then, of course, you can specify your network, taking out links between minor characters, differentiating a one-time conversation from recurrent interaction between two protagonists through the width of a link, e.g., looking for second-order connections between two characters mediated through a third character, searching symmetries and asymmetries in the diagram – all that gives you pretty different visualizations of one and the same drama. What do you think is the use, the advantage of this approach that basically ignores the process, the development of dramatic action in order to picture events as a static tableau, a spatial configuration with no time axis?
FM: Networks are initial maps of the forces at play in a drama. At least for my work they are mostly tools to develop a question, which I then try to answer with all kinds of methods, quantitative methods in particular, of course. In Hamlet, e.g., I was very interested in the importance of ›succession‹ to the different characters. The network with the different interactions suggested a certain ranking. So Hamlet would probably be most interested in the question of ›succession‹, minor characters less, etc. The statistical analysis, however, showed that Hamlet talks the most of succession but only because he talks so much more than any other character in the play. In terms of percentage he talks significantly less about succession than, e.g., the player king and the player queen. This deviance from the initial expectation is what makes the finding of the research. The network helps to clarify your interpretation. Of course there are also other cases. Racine and Shakespeare are both great playwrights. But they are also fundamentally different. How could you pinpoint the difference? The networks of their plays show a recurrent difference: whereas Shakespeare is writing dramas with a completely asymmetrical structure, Racine creates pieces of perfect symmetry! This is interesting. It does not give you a final explanation or hypothesis about the disparity between the two authors, but it is a visualization that longs for interpretation. As I said before: visualizations are objects of study. And the better the visualization, the more it forces you to think about it.
Besides that, however, we are also working on network visualizations that reintroduce time. The network then becomes some kind of movie, focusing on specified interactions within a defined time window. You can then see the rhythmical structure of a play, which may even be independent of the rhythm of the plot. Ryan Heuser from the lab is working on that. Combining his work with the network visualization of the play as a whole will even lead to the connection of the process of the play and its final shape. Different types of interaction in the time windows lead to different symmetrical or asymmetrical shapes of the play.
FG: However, there is one basic question that keeps bugging me about your network visualizations of drama: You assume, as you said, that everything that is done is also there in words, is made explicit. How do you encounter the criticism that with this assumption you are actually missing one important aspect, the aspect of silence?
FM: Well, there are several responses to that question. The first is that, as I said, my focus does not lay on the modern drama, where silence becomes important, but on that drama in which every deed is doubled in words. Then I could also reply that no scientific approach to literature will capture all aspects of it. So maybe I do miss some performance aspects of some plays. That’s just how it is. But then there is also a third response to your question: I tried to make clear that these visualizations are not primarily the result but rather the object of study. The moment you interpret them what they capture positively is as interesting as what they fail to depict. That Hamlet and Ophelia, two ardent lovers, don’t talk to each other until the first scene of the third act somehow weakens the link between them in my network diagram, because there is nothing to be directly depicted. Within the interpretation of this visualization, this can become a very interesting negative fact, as it were. This long silence imposes itself on the beholder precisely as a missing link within the network …
FG: Would you agree then that looking at literature from afar is no strategy to uncover the real, the actual, the true in literature; it is rather a way to change perspectives, to break through the incumbent habit of traditional literary criticism in order to detect new aspects of a seemingly already known object? Much more than the positivistic gathering of data, quantitative research in this context is some kind of methodically induced estrangement. What you do with the data, what you see in your charts, is not just evidence but is up for interpretation and thus argumentation.
FM: You may say so.
1 Franco Moretti: An Atlas of the European Novel. 1800–1900, London, New York 1999, S. 3.
ist Danily C. and Laura Louise Bell Professor of English and Comparative Literature der Stanford University in Kalifornien. Er ist Mitglied der American Academy of Arts and Sciences, war Fellow des Wissenschaftskollegs zu Berlin und Berater des französischen Bildungsministeriums.
Der Begriff des Zeigens ist in den Geistes- und Kulturwissenschaften verstärkt und dauerhaft relevant geworden. Gegen Sprachfixierung und Präsenzvergessenheit als doppeltes Erbe des linguistic turn fasst er die Eigenart visueller, körper- und raumbasierter Medien ebenso wie eine Dimension der Aisthesis, die nicht im Symbolischen aufgeht. Die Beiträge des Bandes bilden einen aktuellen Querschnitt einer sich weiterentwickelnden Debatte: von philosophisch-systematischen Grundlagen über die Analyse technischer und sozialer Dispositive, die Revision von Bild- und Evidenzkonzepten bis hin zu Gesprächen aus der Praxis.
Mit Beiträgen von Hanne Loreck, Dieter Mersch, Michaela Ott, Markus Rautzenberg, Ulrich Richtmeyer, Robert Schade, Philipp Stoellger, Sabine Wirth. Und Interviews mit Franco Moretti und Thomas Hettche.