David Kurnick, Associate Professor of English at Rutgers University, will present a talk at UC Berkeley entitled “The Erotics of Large Numbers: Realism’s Demographic Passions.“ The event will take place on Friday, January 22nd at 3:00pm in 300 Wheeler Hall.
Kurnick is the author of “Numberiness,” a short essay responding to Eric Bulson’s article “Ulysses by Numbers” (Representations 127 [Summer 2014]), published on the Representations blog last year.
The Resistance to Counting, Recounted
Eric Bulson takes it as given that “quantitative readings of literature . . . get a bad rap.” Indeed, the presumed hostility of literary scholars toward quantitative analysis provides the necessary friction for his essay, lending argumentative force and methodological point to what might otherwise seem a rather narrowly focused piece. And it is to highlight the wider stakes involved in Bulson’s contrarian decision to count rather than simply read the words of Ulysses that the editors have invited this accompanying cluster of responses and reflections.
I’m in no position to challenge the view of literary studies as a bastion of numerophobia. I wrote a few years ago that a “negative relation to numbers” is “foundational” to literary studies, which occupies a structural position in the university as the quintessential non-counting discipline. But what strikes me now is that neither Bulson nor I, nor anyone else hoping to expand the space for quantitative analysis in literary research, has presented any quantitative evidence to support this picture of literary scholars as the determined enemies of counting. Wouldn’t “quantitative data… actually help us” in this respect, too, enabling us to take the measure of our presumed hyper-commitment to the qualitative, to calculate its degree and scale relative to other disciplines and to other moments in our own history?
How might such data be gathered? Andrew Goldstone and Ted Underwood have laid down one possible pathway in their “Quiet Transformations of Literary Studies: What Thirteen Thousand Scholars Could Tell Us.”[1] Analyzing a corpus of some 21,000 articles published over the last 120 years in seven major generalist journals of literary study, they counted not just the number of words but the number of “number words” (“from two to hundred and first to tenth”), calculating these as a percentage of all words in the corpus for each year. What they found, surprisingly, was that there used to be a lot more counting going on in our research than there is today: about 50% to 80% more if we compare the first half of the 20th century to the last couple of decades. In fact, based on this particular analysis of this particular dataset, it appears that we have never been so disinclined to quantify as we are at present. It is as if the society’s dominant tendency toward ever more ubiquitous computation and digitization has prompted literary scholars to mount a desperate rear guard action.
Proportion of “number words” by year in a corpus of articles from seven leading journals of literary study. (Source: Andrew Goldstone and Ted Underwood, “The Quiet Transformations of Literary Studies: What Thirteen Thousand Scholars Could Tell Us,” NLH 45.3 (2014): Fig. 1, p. 362.)
So it would seem, based on this particular analysis of this particular dataset. That of course is the rub. To Goldstone and Underwood, the declining frequency of number words suggests that our antagonism to quantitative modes of cultural study may be less foundational than we imagine: a “relatively recent” development (360). That could be right. But it might rather be the case that even our initially “high” level of about 100 number words per article a century ago placed us emphatically at the low end of the disciplinary spectrum as it then stood. And depending on what’s happened since then in other disciplines, our relative position may actually be more centrist now than it used to be, more proximate to some of the social and cultural sciences. Or if that seems too farfetched, what about our position relative to other humanities disciplines? Perhaps in that softer company our current average of about 60 number words per published article represents an unusually open disposition toward counting, a degree of interest in numbers that, however modest, exceeds that of neighboring fields. Because the dataset provides no basis for comparison with published scholarship besides our own, it cannot be used to support or refute any of these hypotheses about our stance in the larger system of social, cultural, and historical research.
To pursue these matters at all rigorously, then, we would need an expanded dataset that includes long runs of more or less comparable generalist journals from a representative range of disciplines. This in itself would present non-trivial challenges, and would require at the very least our consultation with experts from other academic domains to help us pin down the discipline-specific meanings of “generalist” and the relevant hierarchies of journals. But we would then also need to extract from the expanded dataset some different kinds of numbers about the use of numbers. It’s likely that we would have to look beyond “number words” alone for a reliable proxy of quantitative orientations across the disciplines, including for example mathematical terms such as mean, median, percentile, quartile, ratio, variable—maybe the word number itself. And then we would want to know not just the yearly frequencies of these words but how evenly or unevenly dispersed they are across the articles of a given discipline. Does the vocabulary of counting tend to cluster heavily in a subset of articles, which we might then interpret as representing a discipline’s “quantitative wing”? This information would help us to see not just whether literary studies is (as we presume) more dominated than other disciplines by qualitative scholarship, but whether it nonetheless harbors a statistically discernible zone or subfield in which counting is normal practice. The evolving historical position of that subfield within our discipline might then be compared to corresponding trends across the disciplines, be they toward more militant methodological divisions (with increasingly tight concentration of number words in more distinct subsets of articles) or rather toward increasing accommodation of mixed methods (with number words scattering more evenly across each disciplinary corpus). Other historical trends might be brought into the analysis, as well, probing correlations between, for example, declining frequencies of number words and declining percentages of male faculty.
By this stage in our expanding program of research, we would, I hope, begin to question our procedure of treating all the articles in the corpus as though they are equally representative of a discipline’s general practices and preoccupations. We know there are journal articles that practically everyone in the profession has read, and others, even in “major,” “leading,” or “generalist” journals, that practically no one has. Why, if we are serious about gathering and analyzing numbers, would we be satisfied to count every article (or every word of every article) as 1 instead of assigning them weights based on the extent (however approximated) of their circulation or subsequent citation? This indeed has been a failing of most of the work in literary studies that deals with large digitized corpora. Franco Moretti was right to insist that the data appropriate to the study of literary history exceeds the small canon of classics and includes a vast “slaughterhouse” of forgotten works. But that does not mean we can gain much purchase on literary history by treating every book in the slaughterhouse as equivalent, whether it sold in the millions initially but didn’t last, endured as a kind of coterie favorite, or failed ever to find any readers at all. Working with data on a larger scale does not relieve us of the burden of valuation any more than it does that of interpretation. On the contrary, scale and value present themselves as the entwined problems of all quantitative research. As soon as you settle on a scale for your data and metadata, determining to include some things in your counting but not others (authors but not author genders; journals but not books; journal articles but not journal circulation figures; articles in Modern Language Review but not in Modern Language Notes; number words but not mathematical concept words, etc.), you have assumed a whole set of value propositions, most of them highly contestable and capable of exerting a significant effect on your conclusions.
Goldstone and Underwood are well aware of these challenges, careful to specify the assumptions built into their dataset and the uncertainties involved in its interpretation. Their aim in constructing topic models of literary research is not to produce by this means a new history of our discipline but merely to sketch a few provisional contours of that history as a provocation to further inquiry. Likewise Bulson, whose data are strictly limited to the “264,448 graphic units (or language tokens)” of Ulysses and whose claims do not at this stage venture beyond that one supremely canonical novel. He leaves open the question whether Joyce’s relationship to the serial format is generalizable to other authors, whether this particular story, wherein the rigorous constraints of publishing lead unexpectedly to radical freedoms of language and style, might be scaled up to yield a larger historical narrative about modernism’s debts to seriality.
Is Bulson’s ongoing work on this question—with all the sifting and counting it will involve (his data “expanded to include more novels and little magazines,” more authors, more varieties of modernism)—part of a quantitative turn, what Goldstone and Underwood describe as “a recent tendency for literary studies to develop stronger connections to social science” (379)? This is another question that must be left open for further and better-designed forms of inquiry. We have a way to go yet before we can even discern such relational tendencies with confidence, let alone gauge their significance in the history of disciplinary practices.
James F. English
University of Pennsylvania
[1] Andrew Goldstone and Ted Underwood, “The Quiet Transformation of Literary Studies: What Thirteen Thousand Scholars Could Tell Us,” New Literary History 45.3 (Summer 2014): 359. Just as engrossing as this article is the Quiet Transformations website (http://www.rci.rutgers.edu/ag978/quiet), which offers user-friendly tools for further exploration of the dataset.
Eric Bulson’s “Ulysses by Numbers” (Representations 127) asks the literal question, “Why is James Joyce’s Ulysses as long as it is?” Here we have three responses to his question, his methods, and his conclusions:
JAMES F. ENGLISH | The Resistance to Counting, Recounting
Eric Bulson takes it as given that “quantitative readings of literature . . . get a bad rap.” Indeed, the presumed hostility of literary scholars toward quantitative analysis provides the necessary friction for his essay, lending argumentative force and methodological point to what might otherwise seem a rather narrowly focused piece. And it is to highlight the wider stakes involved in Bulson’s contrarian decision to count rather than simply read the words of Ulysses that the editors have invited this accompanying cluster of responses and reflections.
I’m in no position to challenge the view of literary studies as a bastion of numerophobia. I wrote a few years ago that a “negative relation to numbers” is “foundational” to literary studies, which occupies a structural position in the university as the quintessential non-counting discipline. But what strikes me now is that neither Bulson nor I, nor anyone else hoping to expand the space for quantitative analysis in literary research, has presented any quantitative evidence to support this picture of literary scholars as the determined enemies of counting. Wouldn’t “quantitative data… actually help us” in this respect, too, enabling us to take the measure of our presumed hyper-commitment to the qualitative, to calculate its degree and scale relative to other disciplines and to other moments in our own history? (Read full response… )
DAVID KURNICK | Numberiness
“We can indeed count” words, Eric Bulson observes, and concludes that therefore “the counting must go on” (4). The reasons to move from the first remark to the second will not be self-evident to everyone. But “Ulysses by Numbers” gives an unprecedentedly intimate sense of Joyce’s compositional practice, offering not just a fascinating picture of how Ulysses grew but also an account of why it grew in the increments it did. Perhaps the most surprising discovery here for Joyce scholars is the fact that, as Bulson puts it, “even after serialization stopped, Joyce was still writing by the numbers” (26): even released from the 6,000-word increments suggested by Pound for the novel’s serial installments, Joyce kept creating at scales of 6,000. It turns out that “Circe,” which seems to obey no rules save the volcanic logics of the unconscious and Joyce’s own ambition, is dutifully designed to fit into eight installments of The Little Review. Figure 9, where you can see this finding visualized, offers a startling picture of genius in compromise with the materiality of publication.
Bulson thus indisputably helps us get a sharper sense of how “the serial logic of length” (6) conditioned this particular masterwork. Accordingly, my questions about his essay are less about the findings themselves than his account of them, and they concern the charisma that the rhetoric of number itself exerts in the essay. Surely Bulson’s most provocative claim is that his method will help us get at Ulysses’ “numerical unconscious” (4). The formulation suggests an opaque but determining structure whose revelation will be decisive for our sense of the meaning of the whole. And Bulson does tend to connect number with causality in just this way. “More words on the page but fewer seconds passing in the plot: that is a discovery Joyce made while writing Ulysses” (19). This can’t really be said to be a discovery, though, since Joyce could have learned that discursive time affects diegetic time from (to pick a name not quite at random) Homer, who interrupts a classic action-movie moment—an arrow whizzing by Menelaos—with a startling simile about Athena deflecting it “the way a mother / would keep a fly from settling on a child / when he is happily asleep”[1]: the words take longer to read (or to hear recited) than an arrow to miss its mark, and even longer if you pause to think about them. And “more words” is only one way texts slow down story-time: arcane or boring or made-up words can achieve a similar end with relative verbal economy, as can disorienting shifts in point of view, or a lot of jokes, or odd images. Every attempted reader of Finnegans Wake knows that the number of words on the page has relatively little to do with how long it takes to read that page and how much time it seems is passing in the “plot” as you do so (if I had to quantify, I’d say that word count in the Wake isn’t even the half of it). (Read full response … )
HOYT LONG and RICHARD JEAN SO | “A Hail of Information”: Ulysses, Topic Modeled
What can a quantitative analysis of style tell us about James Joyce’s Ulysses? Quite a lot, according to Eric Bulson. In his “Ulysses by Numbers,” Bulson uses some of the simplest forms of “stylometrics”—word counts and measures of lexical diversity—to provide new insights into some fundamental questions: why do the novel’s episodes get longer? What’s the relationship between an episode’s length and its plot? Bulson productively correlates the concrete evidence given by word counts with questions of composition and the material constraints of serialization. While the straightforward empiricism of his argument is a strength, it left us to wonder what it misses by treating words as homogenous numerical units abstracted from their semantic contexts. But not because we believe numbers and counting are unsuited to an interpretation of the novel. One of Bulson’s great insights is that counting is hardly alien to the project of reading Ulysses, an insight encapsulated in an epigraph from Hugh Kenner (“‘Words’ are blocks delimited by spaces. So we can count them.”). For us, the question is how to push this counting further. Can we count the words in ways that do not elide their contextual signifying power? Kenner too was interested not just in the number of words on the page, but the likelihood of certain words appearing with others, in what he called “space-time block[s] of words.”[1]
As quantitative approaches to text analysis have evolved, they have similarly shifted from counting words to counting collocations of words, and even collocations of collocations. One popular innovation along these lines is probabilistic topic modeling, which we propose here as a method for exposing what Kenner calls Ulysses’s larger “verbal systems.”[2] What we discover in the process is in part obvious—that topic modeling as a method of counting is also constrained by its assumptions about words as numerical units and their relation to each other. Ulysses troubles these assumptions, which amount to a highly particular theory of information. Precisely because it does so, however, topic modeling the novel also reveals something of how the novel functions as its own form of literary information. If word counts help us understand Joyce as a “mechanical counter,” topic models help us understand him as a careful “arranger” of latent verbal structures.[3] (Read full response … )
“We can indeed count” words, Eric Bulson observes, and concludes that therefore “the counting must go on” (4). The reasons to move from the first remark to the second will not be self-evident to everyone. But “Ulysses by Numbers” gives an unprecedentedly intimate sense of Joyce’s compositional practice, offering not just a fascinating picture of how Ulysses grew but also an account of why it grew in the increments it did. Perhaps the most surprising discovery here for Joyce scholars is the fact that, as Bulson puts it, “even after serialization stopped, Joyce was still writing by the numbers” (26): even released from the 6,000-word increments suggested by Pound for the novel’s serial installments, Joyce kept creating at scales of 6,000. It turns out that “Circe,” which seems to obey no rules save the volcanic logics of the unconscious and Joyce’s own ambition, is dutifully designed to fit into eight installments of The Little Review. Figure 9, where you can see this finding visualized, offers a startling picture of genius in compromise with the materiality of publication.
Bulson thus indisputably helps us get a sharper sense of how “the serial logic of length” (6) conditioned this particular masterwork. Accordingly, my questions about his essay are less about the findings themselves than his account of them, and they concern the charisma that the rhetoric of number itself exerts in the essay. Surely Bulson’s most provocative claim is that his method will help us get at Ulysses’ “numerical unconscious” (4). The formulation suggests an opaque but determining structure whose revelation will be decisive for our sense of the meaning of the whole. And Bulson does tend to connect number with causality in just this way. “More words on the page but fewer seconds passing in the plot: that is a discovery Joyce made while writing Ulysses” (19). This can’t really be said to be a discovery, though, since Joyce could have learned that discursive time affects diegetic time from (to pick a name not quite at random) Homer, who interrupts a classic action-movie moment—an arrow whizzing by Menelaos—with a startling simile about Athena deflecting it “the way a mother / would keep a fly from settling on a child / when he is happily asleep”[i]: the words take longer to read (or to hear recited) than an arrow to miss its mark, and even longer if you pause to think about them. And “more words” is only one way texts slow down story-time: arcane or boring or made-up words can achieve a similar end with relative verbal economy, as can disorienting shifts in point of view, or a lot of jokes, or odd images. Every attempted reader of Finnegans Wake knows that the number of words on the page has relatively little to do with how long it takes to read that page and how much time it seems is passing in the “plot” as you do so (if I had to quantify, I’d say that word count in the Wake isn’t even the half of it).
It’s not that word count is irrelevant to narrative pacing. But its status as the factor driving Joyce (or any other writer) in a particular novelistic project needs to be established. My sense is that Bulson is drawn to number because we have new and powerful tools to help us count with relative facility—and he has used those tools with precision and ingenuity. But we might be wary of installing the facts those tools let us assemble as the engine of textual construction; this is the methodological metalepsis Pierre Bourdieu identifies when he warns against “giving as the source of agents’ practice the theory that had to be constructed in order to explain it.”[ii] To put it simply: is number essentially operating in Bulson’s argument as a metaphor for length? We could formulate Bulson’s signal discovery about the pace of Ulysses’ growth in two ways: a) Joyce’s earliest episodes average 5,233 words, and later the average jumps to 11,179; b) with “Scylla and Charybdis” Joyce starts writing episodes at double the length he’d agreed upon with Pound, thereby facilitating the publication of the later episodes over two installments of The Little Review. The information referred to by each sentence is identical, but my sense is that the specifics of the first version would be news to Joyce, while he’d readily acknowledge the second, amused that the professoriat has finally caught up with him.
This doesn’t in itself argue for the priority of either version: literary scholarship is under no obligation to limit itself to insights that would have occurred to literary producers, especially scholarship aiming for the unconscious determinants of literary forms. But what makes 5,233 versus 11,179 a more compelling way to describe a relation between literary objects than twice as long? It seems that the appeal of the more precise version derives from the charisma of quantification more generally (as, I suspect, does the temptation to describe that precise version as exercising the shaping power of the unconscious). Ours is a numbery historical moment, and that numberiness has a marked technicist (and digital) bent.[iii] But to fathom the status of “the numerical” in Joyce, one would want to know, in addition to the numbers themselves, what conceptions of the numerical Joyce was working with, and what ideas of number may have been working through him. Does it matter to our sense of number in Ulysses that one of the pioneering efforts to map word frequency in the English Bible was published by the Reverend J. Knowles (who was developing a system to teach blind people to read) in 1904, the year in which Joyce set his novel? Or that one of the major advances in this same field—Edward Thorndike’s ranking of word frequencies in a corpus comprising 10,000 words—appeared in 1921, in the hiatus between Ulysses’ run in The Little Review and the publication of the completed book version? (Would Joyce have known or cared about either of these events?)[iv] What did “6000 words” mean to a word-processorless writer—a painstaking tally, a rough calculation based on the number of manuscript pages, a guess? (Did Joyce or Pound ever literally count anything)? And how did Joyce use these experiences of number to play with (or ignore) the rhetorics of number operative in his moment? With a writer as deliberately self-revolutionizing as Joyce, this will get complex fast: it’s not only that our numberiness is different from his, but that the degrees and meanings of numberiness varied over the course of his work.
These are not Bulson’s questions, but my suspicion is that his findings would become most resonant when seen in their context. To answer such questions we would need an ecology of number in Joyce—one that would account, in A Portrait of the Artist, for the “thousand times” Stephen Dedalus feels he has yielded to Ellen’s charms as well as the “ten thousand idolators” baptized (according to the catechizing rector at Clongowes) by Saint Francis Xavier. The rhetoric of number in Portrait reaches its peak in the hellfire sermon, where we are asked to imagine the walls—“four thousand miles thick”—that pen in the damned, as well as the stench emanating “from the millions upon millions of fetid carcasses massed together in the reeking darkness”; and the novel famously concludes with Stephen’s resolution to “encounter for the millionth time the reality of experience.”[v] Even this quick review makes clear that number in Portrait operates on a decimal system, the zeros intensifying what Joyce thus encourages us to understand as a unified vital energy: eroticism (“a thousand times”), religion (“four thousand”), and artistic vocation (“the millionth time”) are divisible by one another. We are still essentially here within a lyricism of number, number as expressivist amplification device.
Things are altogether different by the time we reach the numberiest of Ulysses’s sections: “Ithaca,” the penultimate episode, proceeds by alternating a series of questions about the ongoing action with the madly technical responses those questions elicit from the narrator (Joyce called the episode’s method “mathematical catechism”). The joke of “Ithaca” is the yawning discrepancy between the abstraction of the language and the experiential texture of the events it narrates. Nowhere is the effect more evident than in the answer to the question, “What relation existed between [Stephen and Bloom’s] ages?” Bloom is 38 in Ulysses, Stephen 22, but naturally the narration will not say so thus straightforwardly. Instead Joyce spins out a mathematical series: starting from 1883, the first year when the ratio comprised by their ages calculated in years was expressible in a non-infinite form (i.e., when Bloom was 17 and Stephen was 1), the narrator points out that had Stephen aged normally from that year while the ratio between their ages had somehow remained constant, by the 1904 in which the novel transpires, Bloom would be 374 (that is, Stephen’s age of 22 multiplied by the factor of 17 that separated them in 1883). By the time Stephen reaches 70 (in 1952), Bloom would be 1190 years old; and if Stephen were himself in turn to reach that fantastic age, Bloom would have to be 83,300 years old, “having been obliged to have been born in the year 81,396 B.C.”[vi]
Does it matter that Joyce’s numbers don’t work? (By my calculation, when Stephen reaches 1190, Bloom would be 20,230 …). Barry McCrea, commenting recently on these errors, has read them as indicating that “the world is neither perfectible nor fully describable … [T]he mistaken calculations serve as a reminder that this is a novel” and not a mathematical formula.[vii] His point is supported by the fact we can make “novelistic” sense even of the outlandishly large and faux-precise numbers: the increasing gap between the men might stand as a figure for Bloom’s sweetly anxious protectiveness toward Stephen, or of Stephen’s self-absorbed inability to imagine himself on the same time-scale as his companion. The very counterfactual ground of the thought experiment, whereby one man ages normally while the other outpaces him at ever more fantastic rates, captures the temporal warpings subtending any relation between reader and story (having first read Ulysses at 22, I will always feel that Leopold Bloom is older than me by a factor of about 1.7; now, unaccountably, I am 1.105 times older than him.) Most mysteriously, the coldness of these calculations doubles as a form of tenderness: the number-crunching of “Ithaca” might be parodying our humanistic orientation to literary character, but it does so by embarrassing us into a solicitude on behalf of literary character (by episode 17, Stephen and Bloom have been engineered to appear to exceed any numerical account of them.)
All of which is to say that Joyce’s episode offers both a micro-history of and a commentary on the coming-to-being of our number ecology, insisting on the headiness of its sublime technicism even as it conjures its most intensely felt reality effects. Joyce’s straddling of these two aesthetics might be one way to describe his continued interest for historians of the novel: Ulysses’ self-constitution as a professional-object-in-waiting makes it feel at once perfectly suited to and faintly mocking of the most technically precise accounts we might offer of its workings. But the suitability and the mockery both remain to be read, as do the relations between them. So, yes, “the counting must go on”—so long as we agree to remain unsure about what the counts mean.
[i] Homer, The Iliad, trans. Robert Fitzgerald (Oxford: Oxford UP, 1974), 60.
[ii] Pierre Bourdieu, In Other Words: Essays Towards a Reflexive Sociology, trans. Matthew Adamson (Stanford: Stanford UP, 1990), 60.
[iii] For incisive recent contributions to the discussion of method in the digital humanities, see Alan Liu, “The Meaning of the Digital Humanities,” PMLA 128.2 (2013): 409-423, and Andrew Goldston and Ted Underwood, “The Quiet Transformations of Literary Studies: What Thirteen Thousand Scholars Could Tell Us,” New Literary History 45.3 (2014): 359-384. Both articles address the interpretive question as it pertains to much larger corpuses than Ulysses, but their discussions of how meaning does or does not inhere in quantitative digital methods raise interesting questions even for what Bulson calls the “single data set” constituted by Joyce’s novel (6).
[iv] J. E. De Rocher, The Counting of Words: A Review of the History, Techniques, and Theory of Word Counts (New York, 1973), 5-9.
[v] James Joyce, A Portrait of the Artist as a Young Man (1914; New York: Penguin, 2003), 72, 115, 128, 130, 275-6.
[vi] James Joyce, Ulysses (1922; Oxford: Oxford UP, 1993), 632.
[vii] Barry McCrea, In the Company of Strangers: Family and Narrative in Dickens, Conan Doyle, Joyce and Proust (New York: Columbia UP, 2011), 143.
What can a quantitative analysis of style tell us about James Joyce’s Ulysses? Quite a lot, according to Eric Bulson. In his “Ulysses by Numbers,” Bulson uses some of the simplest forms of “stylometrics”—word counts and measures of lexical diversity—to provide new insights into some fundamental questions: why do the novel’s episodes get longer? What’s the relationship between an episode’s length and its plot? Bulson productively correlates the concrete evidence given by word counts with questions of composition and the material constraints of serialization. While the straightforward empiricism of his argument is a strength, it left us to wonder what it misses by treating words as homogenous numerical units abstracted from their semantic contexts. But not because we believe numbers and counting are unsuited to an interpretation of the novel. One of Bulson’s great insights is that counting is hardly alien to the project of reading Ulysses, an insight encapsulated in an epigraph from Hugh Kenner (“‘Words’ are blocks delimited by spaces. So we can count them.”). For us, the question is how to push this counting further. Can we count the words in ways that do not elide their contextual signifying power? Kenner too was interested not just in the number of words on the page, but the likelihood of certain words appearing with others, in what he called “space-time block[s] of words.”[1]
As quantitative approaches to text analysis have evolved, they have similarly shifted from counting words to counting collocations of words, and even collocations of collocations. One popular innovation along these lines is probabilistic topic modeling, which we propose here as a method for exposing what Kenner calls Ulysses’s larger “verbal systems.”[2] What we discover in the process is in part obvious—that topic modeling as a method of counting is also constrained by its assumptions about words as numerical units and their relation to each other. Ulysses troubles these assumptions, which amount to a highly particular theory of information. Precisely because it does so, however, topic modeling the novel also reveals something of how the novel functions as its own form of literary information. If word counts help us understand Joyce as a “mechanical counter,” topic models help us understand him as a careful “arranger” of latent verbal structures.[3]
Topic modeling was developed over a decade ago by computer scientists hoping to aid in tasks like “information retrieval, document classification, and corpus exploration.” A major enhancement on prior methods of automated document comparison, it was intended to “discover the themes that run through” a large collection of texts without any advanced knowledge of the texts themselves.[4] While this unsupervised approach was initially applied to the exploration of scientific and news articles, literary scholars have been quick to adopt its most common implementation—which employs latent Dirichlet allocation (LDA)—to track the “migratory formulae” of literary history in thousands of novels; to explore patterns of discourse in critical literature or across multiple types of corpora; and even to find themes in highly figurative poetic language.[5]
What topic modeling is good at is identifying words that occur together in multiple places across multiple documents. It connects words that tend to appear in similar contexts while helpfully distinguishing between uses of words that have multiple meanings. These clusters of co-occurring words are the “topics” produced by the model. For instance, if a topic model were applied to a large corpus of scientific articles, it might find that “matter” and “energy” frequently appeared together in a subset of those articles. Considered alone, these words are ambiguous and would not help the human reader intuit what the articles are about. But the topic model also returns additional contextual clues—words like “particle” and “dark”—thus allowing us to say that these articles probably have something to do with physics, as opposed to energy policy.
This idea that coherent topics like “physics” are latent within clusters of co-occurring words naturally relies on a set of ontological assumptions about what a topic is, but also what a document (or “text”) is and how it is generated. Topic modeling operates on the assumption that it can use the words it observes in documents to infer the “hidden structure” of topics that likely generated those documents.[6] Thus it assumes that there are topics that already exist in the world, like “physics,” and that individual words, such as “neutron,” are probabilistically associated with these topics. It then treats every document as if it were composed from some proportion of these topics, with the words in each topic more or less likely to be chosen based on their statistical distribution within that topic. So, for example, the model will say that a document contains a 50 per cent share of a topic because it contains lots of words frequently associated with that topic, like matter, energy, particle, and neutron. Based on these words, we could then infer that this document most likely has something to do with “physics,” even if it also contains smaller shares of other topics. The topic model essentially reverse engineers the process of composition based on the set of documents it is given, returning three things: lists of words for every topic it infers; the probability of those words being associated with that topic; and the proportion (or share) of a topic present in every document.
If there is any quantitative method that can relate the “blocks of words” in Ulysses to some larger verbal system, topic modeling would seem to be it. Imagine segmenting the novel into hundreds of smaller blocks—about 500 words each—and building an algorithm that infers a hidden structure of 60 topics from these blocks, which here represent our set of “documents” or “articles.” The assumption here would be that each block is composed from a subset of these 60 topics, and that words appeared in the block based on an existing association with those topics (and the words likely to appear within them). On the one hand, this is a preposterous way to consider how Ulysses was written. Surely Joyce did not draw on a limited set of topics as he wrote. And surely he did not write with fixed associations about which words were more likely to belong to which topic. By any measure, Ulysses is a “hail of information”—a spigot of words in which information itself takes precedence over narrative. Yet it is difficult to imagine this information being arranged as coherently as topic modeling would assume. That said, and as critics have long argued, the novel clearly possesses some kind of “hidden plan.”[7]
Indeed, when we ran a topic model on Ulysses, we saw that there was a latent structure in the patterns of co-occurring words, but that this structure was meaningful only to the extent it warped the core assumptions of LDA. For instance, we found that of the 539 total word blocks in the novel, less than a fifth had more than a 20 per cent share of any one topic. Instead, most contained a small share of many different topics, suggesting a lack of topic consistency within each block. This means that Ulysses has difficulty sticking with any single topic. If we look at the top five topics in each block, things get even more interesting. What we find is that the model identifies topics that cohere at the level of each canonical episode, even after we have excluded grammatical function words and character names.[8] All the blocks from “Sirens” and “Ithaca,” for example, huddle around a limited number of topics (Figure 1), meaning that they more or less draw on similar clusters of words. This visualization confirms what we already know from literary scholarship: that Ulysses is generally organized by “episodes,” each possessing its own loosely coherent form of language, style, and theme. In a sense, the topic modeling algorithm replicates the work that scholars such as Stuart Gilbert and Frank Budgen have done in revealing the novel as constituted by discrete “episodes,” thus making Ulysses’s overall “schema” relatively comprehensible to the reader.[9]
Figure 1. One way to visualize the relation of our “blocks of words” to the 60 inferred topics is to create a network diagram. In this case, we began with a network that linked all 539 text blocks to their five most prevalent topics (numbered). We then filtered out all blocks except for those in the “Sirens” and “Ithaca” episodes to show how little overlap there is in their highest ranked topics. The complete network of blocks and topics, along with a list of all 60 topics and their highest ranking words, is available at our website: chicagotextlab.uchicago.edu.
While such confirmation is useful, topic modeling also exposes patterns of language and meaning that have remained hidden to scholars. Consider Topic 35, which is highly associated with the words street, passed, corner, past, bridge, and walked. We might describe this topic as a “walking” topic, or more broadly, an “urban-spatial” topic. A graph showing the fluctuation of this topic across the novel tells us that it peaks in the “Wandering Rocks” episode (Figure 2). This again confirms what we know. This episode, as its most literal level, narrates the physical meanderings of several characters, such as Father Conmee, across the urban landscape of Dublin. The words associated with this movement are what the topic model has picked up on. Yet if we look at the topic’s overall distribution, what is surprising is its sustained presence across multiple earlier episodes, such as “Hades” and “Aeolus.” We know that both are framed by Leopold Bloom walking to or from a specific location, but the strength and persistence of the topic across and through these episodes is nevertheless striking.
Figure 2. A visualization of the relative topic share of “Topic 35” in each 500 word block of Ulysses. The points on the graph indicate the exact share for each block. The line represents a running average of these values.
Few scholars would say that a main organizing theme of “Hades” is its urban landscape. More typically, it is concepts, such as death and memory (particularly Bloom’s recollection of his father’s suicide), that are said to animate the form of the episode.[10] Discrete words like “street” or “car” are seen as mere scaffolding for this larger thematic action. What the topic model results suggest, however, is that this scaffolding may have a larger symbolic function as part of a cluster of words related to movement. Consider a passage from “Hades” that the model identifies as thickly materializing Topic 35: “Mr Powers choked laugh burst quietly in the carriage. Nelsons pillar … We had better look a little serious, Martin Cunningham said. Mr Dedalus sighed. Ah then indeed, he said, poor little Paddy wouldnt grudge us a laugh. Many a good one he told himself. The Lord forgive me! Mr Power said, wiping his wet eyes with his fingers. Poor Paddy!” What seem to be inconsequential markers of space or movement at the level of the individual passage (here captured in the word “carriage”), are exposed as the verbal supports of more significant themes, like death. The topic model suggests that there is a latent, but necessary spatial-conceptual link reinforced within the episode. But also, and this is its most intriguing contribution to a re-evaluation of the novel’s serial origins, across episodes.
In connecting word counts to the history of the novel’s serialization, Bulson also finds something interesting about the “Hades” episode. It is here that the episodes start getting longer. The reason, he speculates, is because the physical world of the narrative also began to “expand,” with interactions between characters becoming more complex. As a result, Joyce needed more words. It is with “Hades” that Joyce began rethinking the overall design of his work, imagining each future installment as potentially longer than the last, and thus transforming the kinds of things he could write about. What topic modeling adds to this argument is a way to see not just how serialization was changing Joyce’s attitude about how many words he could write, but in how many different combinations he could write them. Given the chance to pack in more and more information, we have to wonder how he decided to organize that information differently. Looking at Topic 35 again, it seems significant that Joyce was laying the groundwork in “Hades” for one part of a verbal system that would become greatly extended in “Wandering Rocks.” What might other topics reveal about the evolution of Joyce’s “blocks of words” as serialization of the work progressed?
These are the kinds of questions that will most likely appeal to literary critics and validate the felt usefulness of numbers as an interpretive method. More work, naturally, is needed to take these questions further. But Bulson’s essay, as well as our brief extension of his project, point to a new interpretative model for the modernist novel. Some will continue to insist that word counts or empirical models do violence to readings of Ulysses: such texts do not signify meaning via the quantity or collocations of words, but through the attention of individual human readers to the words on the page. We would not quarrel with that. Ulysses does not signify through word counts and topic models, but it can still be known through them. Indeed, the history of scholarship has left us a view of the novel as a mass of words that needs to be classified and schematized. As Kenner puts it, the novel is a “hail of information” that can be “retrieved and systematized,” and only through this labor do we “know a fraction of what we may think we know.” Bulson has shown how as simple a process as counting words can make new sense of this “hail.” With topic modeling, this project can be extended even further, inviting Ulysses into a much longer history of information theory. Not, however, because Joyce’s text can be construed as “information” in ways commensurable with current quantitative or computational methods. But precisely the opposite. Its incommensurability with the ontology of these methods exposes the continuities and ruptures that the novel shares with the contemporary information age and the tools that seek to order it. It is within this zone of continuity and discontinuity that new forms of reading will emerge.
[1] Hugh Kenner, Ulysses (London: George Allen & Unwin, 1980), 24.
[2] Kenner, 106.
[3] Here we refer to Kenner’s famous description of Ulysses as pulled together by some great “arranger.” Ibid., 61-71.
[4] David M. Blei, “Probabilistic Topic Models,” Communications of the ACM, vol. 55, no. 4 (April 2012), 77-78.
[5] For more thorough discussions of topic modeling as applied to literary and historical texts, see Matthew L. Jockers, Macroanalysis: Digital Methods and Literary History (Urbana, IL: University of Illinois Press, 2013), ch. 8; Andrew Goldstone and Ted Underwood, “The Quiet Transformations of Literary Studies: What Thirteen Thousand Scholars Could Tell Us,” New Literary History, vol. 45, no. 3 (Summer, 2014): 359-384; Tim R. Tangherlini and Peter Leonard, “Trawling in the Sea of the Great Unread: Sub-Corpus Topic Modeling and Humanities Research,” Poetics, vol. 41, no. 6 (December, 2013): 725:749; and also Lisa M. Rhody, “Topic Modeling and Figurative Language,” Journal of Digital Humanities, vol. 2, no. 1 (Winter, 2012).
[6] Blei, 79.
[7] Kenner, 23. See also Derek Attridge, “Introduction,” in James Joyce’s Ulysses: A Casebook (Oxford: Oxford University Press, 2004), 3-11.
[8] Before running the topic model, we excluded all grammatical function words as well as all personal names and all character names. For the personal names, we used Jockers’s expanded stopword list. See Macroanalysis, ch. 8.
[9] See Attridge, 10-11, for an excellent review of this scholarship and what it accomplished.
[10] See Michael Seidel, Epic Geography: James Joyce’s Ulysses (Princeton: Princeton University Press, 2014), 161, for just one recent example.
At the symposium “The Literary and Its Outsides” held at UC Berkeley on April 1, Representations editorial board members Kent Puckett and Stephen Best offered responses to the journal’s recent special issue “Denotatively, Technically, Literally.”
STEPHEN BEST | “Well, that was obvious.”
I would argue that the current period of disciplinary navel-gazing began in 2004, with the Critical Inquiry volume on the future of criticism that was organized to mark the journal’s 30th year of publication. Bruno Latour’s “Why Has Critique Run Out of Steam?,” in which he makes a number of observations regarding the impasses of critique in science studies, appears in that volume, and has since served as a touchstone for a generation of special issues on the limits of historicist “critique.” At Representations, in the fall of 2009, we published “The Way We Read Now,” a volume in which a number of the scholars assembled here today also participated. That volume identified methods of interpretation that depart from a long tradition of “symptomatic reading,” which understands texts as symptoms of hidden structural and psychic forces – from material histories of the book and “new formalisms” to sociologies of reading and data mining. In the introduction to that volume, Sharon Marcus and I offered the notion of “surface reading,” arguing that literary critics might productively locate meaning in what texts say rather than in what they do not, cannot, or will not say. One of the points of our introduction was to extend Latour’s observations to the domain of literary criticism, suggesting that the work of demystification, so central to Marxist, psychoanalytic, feminist, and New Historicist criticism of the 1980s and 1990s, loses much of its effectiveness in a moment when its moves can be anticipated. Heather Love’s “Close But Not Deep: Literary Ethics and the Descriptive Turn,” published in a 2010 special issue of New Literary History called “New Sociologies of Literature,” suggested that practices of description offered a useful alternative to depth hermeneutics, drawing for her method the accent on interactions and operations in the sociology of Latour as well as Erving Goffman. At least two special issues appeared in the last several months that I would consider a part of this moment of critical self-reflection. In the commemorative volume of American Literary History, entitled “History, Historicism, and Historiography,” a journal that had relentlessly scrutinized the categories American and literary since its inception has turned its attention toward the third category – history – which over that same time, in the thinking of at least one of the contributors, had been taken more in the Jamesonian sense of an “imperative.”[i] And, finally, the Victorian Studies volume “The Ends of History” responds to these calls for alternatives to historicist “critique,” setting forth a range of practices that aim to reshape the historicisms of the past.
Cathy Gallagher, in the “Afterword” to the Victorian Studies volume, observes that all this controversy “sounds like an echo chamber for almost every debate regarding historical criticism that has emerged since the 1950s.”[ii] Thus when Love asserts that “deep” reading and textual “richness” serve as carriers for “an allegedly superannuated humanism,” she echoes Althusserian critics’ accusations that Raymond Williams merely propped up bourgeois humanism.[iii] Or when anti-critique critics assert their hostility to ideology, they are merely grafting their ideas onto Foucauldian roots that had been crucial to the New Historicists, where power “denied that texts (and discourses) were pathological,” and as such both devoid of symptoms and incapable of generating symptomatic readings (Gallagher 686). Each one of the contributions to the Victorian Studies volume, she writes, “reverberates with a unique arrangement of echoing judgments,” and in that respect, like the broader debate to which the volume contributes, rings “familiar rather than unique to the current critical moment” (683-4). We’re not dealing with our issues in these special issues — that seems to be the conclusion.
But ten years of special issues sure seems like a lot of anxious worrying about the future of criticism and the limits of historicism, certainly more than can be reduced to an “echo” of a previous generation’s terms of debate. I understand that, in this circumstance, I might be expected to issue a resounding “no” to this charge – to deny the claim that the terms of recent literary critical debate simply rehash those of an earlier one. Still, I’d prefer to inhabit the charge, and argue that we ought to find interesting the reasons behind the echoes and repetitions.
The current issue of Representations (“Denotatively, Technically, Literally”) hones in on some of the concerns regarding literality that were raised in “The Way We Read Now,” taking up, in particular, language that, as literary critics, we often leave unread, “as if it were denotative, literal, and technical,” as Freedgood and Schmitt write (1). When we read so as to concede that language “either means what it says or stands outside our purview,” they continue, “we create or recreate these categories of language within the literary, lightening our work but also thinning texts to predetermined sites of meaning or interpretive possibility . . . . To read denotatively, technically, or literally is not to explicate . . . . It is the reverse: to restore obscurity to the apparently clear, to stop language from working” (1). What if we pitched Gallagher suspicions at this description of denotative reading: does this way of configuring denotative reading echo judgments about literary language formed by a prior literary critical generation? By folding denotation into the act of literary reading, does it not bear a familiar ring with Paul de Man’s judgment of the literary as the kind of writing uniquely aware of the instability of the distinction between the literal and the figurative, between grammatical and rhetorical modes of meaning? And by offering “nonfigurative reading” as one way of approaching the text, are the editors simply coming at the literary from the opposite side to de Man’s “rhetorical reading” – denotative reading and rhetorical reading sharing a basic premise, i.e., the idea that “we organize – [that is] stabilize – language as we read it,” specifically the instability between the literal and the figurative?[iv] When de Man notes that the irony of the literary text is that it “says one thing and performs another,” couldn’t that apply to its denotative aspect as well if to read denotatively, technically, or literally, again, “is not to explicate . . . [but] to restore obscurity to the apparently clear” (4)? And when he claims that a text “simultaneously asserts and denies the authority of its own rhetorical mode,” isn’t that precisely what the editors have here claimed for its denotative mode?[v] All that said, and accepting the link back to de Man for the sake of argument, should it be a problem that “denotative reading” echoes “rhetorical reading” when the former proves so phenomenally productive: giving us more to read, as the editors put it, that is, directing our attention to the “understudied languages of the novel” (3), the stuff we’ve ruled out, ignored, deemed opaque, or passed by?
In Louis Menand’s recent review of the Evelyn Barish biography, he observes that the de Manian project of bracketing off the “real-life aspects of literature” – that it’s written by people, that it affects people – the putting aside of these aspects allowed for a more purposeful burrowing into the way literary language works. Menand celebrates this way of reading as de Man’s attempt to (as he puts it) “get inside the atom.” What if denotative reading doesn’t alter the fact of that burrowing, but simply shifts the terms a bit, from rhetoric and figurality to literality, denotation, reference and description? What we might lose in that case is precisely the configuration “the literary and its outsides” — some requirement that one must jettison denotative or literal language beyond the pale of critical scrutiny in order to preserve the distinctiveness of literature. What we might stand to gain is a sense of another generation of literary critics’ attempting to “get inside the atom.” To come at it another way, as Kent put it to me in another context, rather than ask whether something is a description or an interpretation, a text’s surface or its depths, what the text says or what it means, a denotation or a connotation, rather than make “dualist” distinctions, perhaps we ought to be ask what is currently drawing our attention to description?, what is currently drawing our attention to denotation? The answer might be something like the urgency of coming to terms with the nature of things. The interest in description, the interest in denotation: both reflect an anti-dialectical or post-dialectical turn in recent critical theory, and share the “monist” impulses of other critical tendencies:
There look to me to be monist impulses in Roland Barthes’s “return of literality,” which Rachel Buurma and Laura Heffernan attribute to his turn away from the “metalanguage” of critique and toward that sense of “notation” necessary to begin thinking about the writing of a novel: Barthes’s attempt to (in his words) “guid[e] the work toward the presence of the world, making the world co-present to the Work.”[vi] This turn Barthes achieved over the course of the ten years separating “The Reality Effect” (published 1968) from his lecture course “The Preparation of the Novel” (which he began to teach at the Collège de France in 1978). Barthes bemoaned the way the earlier essay unwittingly replicated the distinction between the meaningful and the living (the meaningful and the meaningless) in the very act of critiquing it. The critic distrusts the evidence of the senses and experience: the problem for Barthes (in “The Reality Effect”), as Buurma and Heffernan’s phrase it, is that “what at first seems to be referential meaning turns out to be ideologically laden meaning designed to produce the effect of a meaningless, found world whose ‘resistance to meaning’ lies apart from language and signification.”[vii] A meaningless, found world: our distrust of experience is a product of ideology critique, and the critic’s metalanguage is what fuels that distrust. In place of the critic, Barthes substitutes the note taker who knows “that there is no life that is not already meaningful,” and that he or she “is already living in a fully meaningful world already, and knows this in a way that does not necessarily hurt, does not necessarily entail an enormous loss” (84-5). But such hard-won knowledge isn’t to be arrived at immediately; it is only available in the form of a “return to literality.” Barthes’s literality, Freedgood and Schmitt remind us, “is not the beginning but ‘the end of a journey’” (Freedgood/Schmitt, 5). This is why Barthes designates it “literal II.” From The Preparation of the Novel: “In order to reach the third state, the literal II, or the return of the literal, you first have to pass through interpretation . . . . The first literality: an arrogance [is not equal to] the second: a ‘wisdom’” (Barthes, 81). In order to “reach the third state,” Barthes had to abandon the identity of critic, had to put “a stop to the division of the subject,” and take on the mantle of novelist.
Barthes’s “return of the literal” sounds an awful lot like some descriptions of the obvious that I’ve happened upon as I’ve worked on Benito Cereno; descriptions that seem less intent to classify the obvious, or define it, and more keen to temporize it as an irreducible aspect of reading. Here is one example that appears in A. S. Byatt’s Possession, she writes: “the knowledge that we shall know the writing differently or better or satisfactorily runs ahead of any capacity to say what we know, or how. In these readings, a sense that the text has appeared to be wholly new, never before seen, is followed, almost immediately, by the sense that it was always there, that we, the readers, knew it was always there, and have always known it was as it was, though we have now for the first time recognized, become fully cognizant of, our knowledge.”[viii] Now, any critical position worth its salt will almost always finally rest upon calling certain other claims obvious and thus to be ruled out, ignored, passed over.[ix] The trivialization of the obvious in literary criticism feels almost like an inaugurating gesture. The obvious is something, as Stanley Cavell has put it, “which for some reason is always underestimated, habitually . . . by critics, even when the art which hosts [it] is devoted to [its] seeing, and the artist set against that underestimation” (82). What criticism trivializes, novelists valorize (even one who, like Barthes, simply hoped to be a novelist). But what interests me are not these attempts to host the obvious, but this desire to “pass through interpretation” on one’s way back to the literal or the obvious; this desire to make the obvious or easily available not something superficial, but an achievement, something that requires critical work.
Let me close with a gesture toward another field of inquiry: the philosophy of perception. I think that it is here, in particular, in the work of philosophers who reject what is called the “myth of the given,” that we find a parallel means for working out the claim that it can be an achievement and require work to see what’s right in front of you. I have in mind the argument, say in John McDowell’s Mind and World, that even though perception allows us passive access to knowledge of the world, we wouldn’t be able to have such access unless we also had certain rational capacities in the background that we needed to be trained into. McDowell has no philosophical use for an idea that our experience contains representations that are not conceptually structured, that one has access to a genuinely passive perception. It is possible that the concepts of surface reading or denotative reading do not lend themselves to this type of analysis, but there does seem to be some consensus for the view that it can require work to see what’s right in front of you; or, to give Freedgood and Schmitt the final word (and go ahead, hear a bit of Neil Sedaka when I read them, a bit of melancholy for that old flame, rhetorical reading): “literal reading . . . is hard to do.” (9).
[i] Jennifer Fleissner, “Historicism Blues,” American Literary History, vol. 25, no. 4 (Winter 2013): 1-19.
[ii] Catherine Gallagher, “The Ends of History: Afterword,” Victorian Studies (Summer 2013): 683.
[iii] Heather Love, “Close But Not Deep: Literary Ethics and the Descriptive Turn,” New Literary History vol. 41, no. 2 (Spring 2010): 371-391.
[iv] Louis Menand, “The de Man Case,” The New Yorker, 24 March 2014
[v] Paul de Man, Allegories of Reading (1979), p.17
[vi] Roland Barthes, The Preparation of the Novel, trans. Kate Briggs (New York: Columbia University Press, 2011), 205.
[vii] Rachel Sagner Buurma and Laura Heffernan, “Notation After ‘The Reality Effect’: Remaking Reference with Roland Barthes and Sheila Heti,” Representations 125, Winter 2014: 80-102, 84.
[viii] Antonia S. Byatt, Possession; quoted in Morrison, Playing in the Dark: Whiteness and the Literary Imagination (New York: Vintage Books, 1992), xi-xii.
[ix] Stanley Cavell, “The Avoidance of Love: A Reading of King Lear,” in Disclaiming Knowledge (Cambridge: Cambridge University Press, 1987), 82-4.
KENT PUCKETT | Response
This special issue of Representations is an important contribution to a larger conversation happening in the humanities today. What do we want or expect from interpretation? Should we read for meaning, for significance, for depth, for connotation, or is it time to read for something else?
We need, however, to ask: are these methodological or historical questions? Or are they methodological questions that contain or conceal an idea about history? Or are they historical questions that must for some reason take the form of methodology?
The age of interpretation, of suspicion, of the symptom is the age of Freud and Marx, of the unconscious and ideology, and a consensus seems to be emerging that those models have somehow outlived their usefulness. What, though, do we think about Marx and Freud, about ideology and the unconscious when we imagine them as having passed their respective dates of expiration?
If we take it that these models demand interpretation because they rely on “vulgar” oppositions between consciousness and the unconscious, base and superstructure, surface and depth, connotation and denotation, then good riddance.
It seems to me, though, that both Marx and Freud are more subtle than that (surprise!). In both cases, interpretation is a response to relations between things, and, indeed, it is the fungibility of relations that demands and enlivens the act of interpretation. When we interpret the dream (something analogous could be said about the commodity), we’re not exposing the unconscious, we’re rather exploring a vital relation between things, between latent and manifest contents as well as between the past and the present, inside and outside, the pleasure principle and what lies beyond.
As everyone knows, the content of and explanations for dreams are boring. Relations, on the other hand, are always interesting. This is why the object of Freudian dream analysis is not the dream but the dreamwork, the processes of condensation and displacement that sit between two contents that were always sort of obvious to begin with.
Rather than think of interpretation as a final revelation of what’s behind the dream or the symptom, psychoanalysis—at its best and it’s, of course, not always at its best—is an invitation to remain in the space of that relation. This is why Freud imagines the dream as having a navel, why he requires threshold concepts like secondary revision and deferred action, and why, finally, he sees psychoanalysis as interminable.
Barthes makes a similar case in Writing Degree Zero and S/Z. In his effort to identify the literary, he draws on the terms of structural linguistics and imagines two poles of significance that bracket literary writing. On the one hand, there’s the pole of pure style: “it is the writer’s ‘thing,’ his glory and his prison, it is his solitude.”[1] This is parole without langue, a private writing that is so personal, so raw that it must be incomprehensible to others. On the other hand, there is what he calls language: “a language is a kind of natural ambience wholly pervading the writer’s expression, yet without endowing it with form or content.” This is langue without parole, the storehouse of conventions out of which the writer selects his or her already cooked materials.
We might think of both these poles as being only available to denotation; they can be named, which is to say repeated, but not explained or interpreted. When faced with either style or language, the magic word or the phone book, the imaginary or the symbolic, all one can really do is point because neither style nor language can ever on their own be like anything else. Because literary writing exists somewhere on the line that links these poles, and because denotative limits make that continuum possible, it represents a point where, like the space between the manifest and latent content of the dream, denotation passes into connotation and connotation into denotation.
The literary happens, perhaps, when we can’t tell the difference. This is, I think, what Genette calls “the paradox of every poetics”: “always torn between those two unavoidable commonplaces—that there are no objects except particular ones and no science except of the general—but always finding comfort and something like attraction in this other, slightly less widespread truth, that the general is at the heart of the particular, and therefore (contrary to the common preconception) the knowable is at the heart of the mysterious.”[2]
This is why instead of saying that we need to choose between denotation and connotation, Barthes says in S/Z that denotation is always only the place where connotation comes to a provisional halt; they are, he says, only “supposedly different systems.”[3] This is not to deny the reality of denotation; as Barthes understands, denotation is a real aspect of language. It is, as Jakobson said before him, one and only one of its several functions. Literary reading is, in that case, always suspended between denotation and connotation; it is the recognition that literature is meaningful not because it is surface or depth, denotation or connotation, but rather because it is a protean point of suspension between the two.
So, from a methodological point of view, the recovery of denotation as a term of art is crucial because it helps us to avoid arguments that emerge from reifying differences that are projected as metaphors onto an account of language that we should understand not in terms of seeing oppositions but, rather and following Wittgenstein, seeing aspects. Language has different functions, different aspects and literature happens when some or all of those functions are held in a certain kind of suspension. To assert these differences as ontologically as opposed to heuristically real, or in fact to read Barthes or Freud or Genette as asserting them as ontologically as opposed to heuristically real is to miss the point.
The essays in this volume help us to see that moment of suspension between connotation and denotation in several different expressions: in the space between belief and disbelief conjured by the Victorian ghost; in the space between invention and discovery, induction and deduction in Victorian science; in the space between the personal impersonality of technical maturity and the impersonal personality of the classic bildungsroman; and the space between known of literary convention and the unknown of early subaquatic discovery.
What these examples also share is a rough time period, one that coincides with the kind of interpretive suspension that I’ve been describing.
A historical question thus remains: do denotation and its interpretive demands work in the same way at other moments in time?
This is the question that seems to motivate some recent turns away from interpretation toward surfaces, toward new ontologies, toward the assemblage. Implied here is the idea that something has changed that demands a new way of seeing meaning. Have materialist discoveries about the brain mooted talk of the unconscious? Has a widening sense of ecological precarity undermined the urgency of the phenomenological? Has the financialization of the culture industry undone a historically specific alienation that underwrote the antinomies of bourgeois thought and the critical theories that went along with them?
There’s obviously too much to say here. I want, though, to suggest that these essays begin to imagine a politics of denotation and thus of interpretation that we need carefully to consider. It’s there in Cannon Schmitt’s invitation for us to look differently at the work, the labor of the novel, in Margaret Cohen’s suggestion that we consider what happens when there are no coral reefs left to denote, in Ian Duncan’s discussion of the passage between the figural and and the empirical that authorizes science, in Elaine Freedgood’s suggestion that ghosts not only allow us to diagnosis a moment’s relation to belief but also to follow the traces of its political disavowals: ghosts “always return: as farce, as tragedy, as religion, and as an exercise in choosing what to believe, and more radically perhaps, what to see.”
Indeed, Freedgood’s essay—and the whole of the issue—made me think of another example. In 1919 Abel Gance made a film called J’accuse. It’s a silent antiwar melodrama that follows the experience of two soldiers who love the same woman during the First World War. One dies and the other goes mad with shell shock. At the film’s end, the latter returns home, wild-eyed and apparently paranoid, to tell the people of his town that he’s seen the ghosts of the war dead rise from their graves. Before he finishes, the war dead appear for all to see and march through the town, accusing those who betrayed the war dead and saying goodbye to the faithful. Gance handles the moment with remarkable restraint, using montage and subtle, ghostly transparencies to create a believable encounter between the living and the dead. The gesture is also a moment of political denotation: Gance uses the sequence to point at what’s hard to point at, the whole human costs and consequences of war. To accuse, for Gance, is to try and denote cinematically, which is also to demand that his audience interpret.
In 1938, Gance remakes the film as a sound feature; it is once again called J’accuse. Where the earlier version was a response to a war that had just ended, the later version is made on the eve of another war, a war that seems to depend on our having forgot the suffering and stupidity of what came before. He also restages the earlier film’s ultimate appearance of the dead. Where in the first film he relies on montage, in 1938 he uses lurid anamorphic lens effects to suggest the breakdown between the worlds of the living and the dead. And as opposed to the quiet, if chilling confrontation between dead soldiers and survivors, he films the sequence as if it were a horror film. The townspeople run, scream, and riot as if the soldiers were Romero zombies as opposed to Shakespeare’s ghosts. And instead of actors, Gance uses real veterans from the First World War, soldiers whose faces were mutilated or, as they said, “broken” in an apparent attempt to bypass the merely believable in order to show us the real. This, too, is an act of denotation, a fantastic effort to name directly and not merely to imply or to figure the costs of war.
So, we have two moments of denotation that share the same subject (war in all its complexity), the same narrative context (the story of two soldiers), the same historical object (WWI), and the same director (Abel Gance). They are, though, entirely different. What, then, has changed? I want to say that denotation has changed; the basic coordinates of aboutness, of denotation, of connotation have changed. Gance demands something different from denotation after one war and before another and so allows us to track a passage in the related histories of war, belief, and reference. Because denotation has a history, turning to it is necessarily not to turn away from either interpretation or critique. So, while there’s much more to say here, I’ll end by suggesting that what Gance and these essays have to tell us is that there’s critical life in denotation, connotation, and interpretation still yet. We don’t have necessarily to give up the ghost.
[1] Roland Barthes, Writing Degree Zero (Macmillan, 1977) 11.
[2] Gerard Genette, Narrative Discourse: An Essay in Method (Ithaca: Cornell University Press, 1983) 23.
[3] Barthes, S/Z: An Essay (Macmillan, 1975) 9.