I've posted a new working paper. Title above, links, abstract, contents, and introduction below.
Academia.edu: https://www.academia.edu/128029987/Computation_Text_and_Form_in_Literary_Criticism_A_Conversation_with_Claude_3_7
SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5166930
ResearchGate: https://www.researchgate.net/publication/389591524_Computation_Text_and_Form_in_Literary_Criticism_A_Conversation_with_Claude_37
Abstract: Literary criticism operates with contradictory definitions of “text,” rarely meaning simply the marks on a page. This makes it difficult to establish what “form” means. While critics do analyze features like rhyme and meter, or the distinction between story (fabula) and plot (syuzhet) criticism rarely seeks to understand how words are arranged in texts beyond these basics. Literary criticism selectively borrowed from Lévi-Strauss's structural analysis of myth (e.g. the concept of binary oppositions), it ignored a systematic methodology that was essentially computational in nature and about form. Now, Large Language Models present a watershed moment for literary studies - they're unavoidable and demonstrate sophisticated capabilities. A cohort of younger scholars using corpus linguistics and computational methods may represent a bridge between computational and literary approaches. Will these scholars extend computational thinking from method to theory? - using computation not just as an analytical tool but as a framework for understanding how literary texts function – that’s a key issue currently before the discipline.
Contents
How Do You Explain Computation to a Literary Critic? 1
What am I trying to explain to myself? 1
Why do literary critics need to know that? 2
The study of literary form 3
My Conversation about Computation with Claude 3.7 5
What is a text? 6
Literary criticism after World War II 7
What is Literary Form 8
Lévi-Strauss on Myth 10
The Current Situation: LLMs 12
Digital Humanities 13
How Do You Explain Computation to a Literary Critic?
I have been interested in conceptualizing literary processes as computational ones ever since I wrote an MA thesis on “Kubla Khan” in the early 1970s. Literary process? How texts are created and how they are read and assimilated, that’s what I mean. Ever since then I have been trying to explain that to literary critics. That has not gone well. I’ve also been trying to explain it to myself, which is a different kind of problem. The purpose of this paper is to, once more, attempt to explain computation-as-literary-process both to literary critics and to myself.
In the rest of this introduction I talk about explaining computation, first to myself, and then to literary critics. The rest of this paper consists of a discussion I had with Claude 3.7 about computation and literary criticism. I begin by discussing the text, then discuss the post WWII focus on interpretation as the central activity of literary criticism. Then Claude and I take up form, the influence of Lévi-Strauss, and the current emergence of large language models (LLMs). We conclude with some remarks about digital humanities.
What am I trying to explain to myself?
I know what kind of questions follow from thinking about literary processes as computational. I’ve done practical criticism on that foundation for decades, some quite rigorous, some rather informal, but all based on the belief that literary processes are computational processes. What is it that I’m trying to understand?
Let me explain: My basic conception derives from the use of parentheses in arithmetic or algebraic expressions. Without parentheses or a specific convention concerning operator precedence, expressions like this next one are ambiguous: 5 × 6 – 3. We can resolve the ambiguity by adding parentheses, giving us either of these two expressions:
(5 × 6) – 3 = 27
5 × (6 – 3) = 15
Adding parentheses places boundaries in the expression that affect how we evaluate it.
The same applies to texts, any texts, not just literary texts. Texts are strings of alphanumeric characters. Small sets of characters are grouped together into words and words are organized into longer strings by using spaces, punctuation marks, and capital letters. Boundaries divide long strings into substrings to whatever depth is necessary. Linguistics, psycholinguistics, and computational linguistics all proceed on that basis.
As far as I’m concerned “computation” is the best name for that process, whether we’re talking about language or arithmetic or mathematics in general. In this context arithmetic is, in effect, a very constrained and specialized kind of language.
Since I’ve known that for years, what is it that I’ve only just now realized? It’s not a matter of new information or ideas, it’s a matter of emphasis, of framing. What I’ve realized is simple: That’s all there is. Let me repeat that: Computation is all there is. That is to say, if we are going to treat literary texts as physical objects, as strings of symbols realized in some physical medium, such as vibrations in air or marks on a surface, if that’s what we’re going to do, then computation is the only physical process we know of that gives us ways of thinking about how strings can support the communication of meaning between individuals. That’s it.
Commensurability is the issue. Computation is a process that is commensurate with the physical text. Why? Because computation, real computation, is itself a physical process. That’s obvious when you consider computation with an abacus, for example. The same is true for a mechanical calculator or a slide rule. The physical nature of computation with a digital computer is less visible, much less, and hardly palpable. But it is no less real. Bits and bytes may appear to be abstract things, but they take up physical space in computer memories; that’s why the capacity of computer memories is measured in bytes, kilobytes, megabytes, gigabytes, etc. Computation takes place in time, and time is physical as well, though one can’t see it, touch it, or taste it, much less lift it. Computers and their processes are physical things, as are human brains. As such they are commensurate with physical texts.
That’s what literary critics need to know. Alas, they are incapable of hearing it, of taking it seriously. Why? Computers are machines, and, as such are at suspect at best and thus to be ignored, and at worst they are evil, and actively to be opposed. This reflects the historical distinction between the sciences and humanities, a distinction that has been written into the institutional structure of the academic world and so has come to seem inscribed in the foundations of the universe.
Like it or not, that universe is all but finished. We are now living in a world where anyone with an internet connection can engage in sophisticated linguistic activity with a machine, a chatbot. That changes everything. The physical text can no longer be effaced in favor some vague abstract object conjured into (virtual) existence by the act of interpretation.
Why do literary critics need to know that?
They need to know it because it allows for a much more robust and substantial analysis of literary form than is currently practiced. Literary texts are linguistic strings. The form of strings is a matter of how elements are arranged on the string. That’s obvious in the case of, say, colored beads on a string. And that’s how musicians and musicologists treat musical form, as the arrangement of sonic objects in a line, where the line tracks the passage of time.
Literary critics do this as well. When critics work out the rhyme scheme of a poem, they’re treating it as a physical object. When narratologists distinguish between story (fabula) and plot (syuzhet), they are treating the text as a physical object, one where story episodes are ordered into a narrative string. One can mention other examples as well. While those operations are well-established, they are a minor aspect practical criticism. Form is treated as a framework in which one can approach the problem of meaning. Its description and analysis are at best a secondary activity.
Interpretation is primary. And when critics start interpreting a text, they tacitly, surreptitiously, efface the physical text in favor of an abstract and poorly defined object to which they then direct their interpretive activities. To complete the mystification, that abstract and poorly defined object is called – you guessed it – the “text.”
But these chatbots – I’ve been working with OpenAI’s ChatGPT and Anthropic’s Claude – they handle language in a way that is astonishingly humanlike, and they are undoubtedly machines, albeit in the strange and special way that computers are machines. They’re not made of gears, levers, axles, pulleys and motors, but their operations are mechanistic in kind, as the word “machine” has come to be understood. Yes, there are serious philosophical issues about whether or they really understand language, whether or not they’re really dealing in meaning. I’ve thought about these issues, and written quite a bit about them, but this is not the place to entertain those discussions. Whatever these machines are doing, to a first approximation it looks like they’re going language.
While I believe that this technology is a source of useful tools, both for literary critics and others, and will become more so in the future, that’s not why I mention them here. I mention them because they constitute an unmistakable “existence proof” that thinking about language process as computational is fruitful, both intellectually and practically. Once we can do that, we can take literary form as a serious object of critical activity.
The study of literary form
Yes, I know that form is an important concept in literary criticism and that the discipline has seen various formalist critical approaches. But as Claude and I discuss, formalism doesn’t not imply close attention to the formal properties of literary texts. Rather, formalism is a philosophical position about the nature meaning in which meaning in literary texts is taken to be somehow different from meaning in other texts by virtue literary form. But form itself is given only cursory attention.
My primary object here is to mention what is perhaps the best-known essay by Susan Sonntag, “Against Interpretation.” It originally appeared in 1964 in the Evergreen Review and became the title essay in her first collection, which was published in 1966. In that essay she asserted:
In some cultural contexts, interpretation is a liberating act. It is a means of revising, of transvaluing, of escaping the dead past. In other cultural contexts, it is reactionary, impertinent, cowardly, stifling.
Today is such a time, when the project of interpretation is largely reactionary, stifling. Like the fumes of the automobile and of heavy industry which befoul the urban atmosphere, the effusion of interpretations of art today poisons our sensibilities. In a culture whose already classical dilemma is the hypertrophy of the intellect at the expense of energy and sensual capability, interpretation is the revenge of the intellect upon art.
Several pages later she asserts:
What is needed, first, is more attention to form in art. If excessive stress on content provokes the arrogance of interpretation, more extended and more thorough descriptions of form would silence. What is needed is a vocabulary—a descriptive, rather than prescriptive, vocabulary—for forms.
It seems like everybody read that essay at that time, but as far as I can tell, no one took Sonntag seriously. Now’s the time; to quote Charlie Parker, now’s the time.
The proof, as they say, is in the pudding. But this is not the place to serve pudding. If you want to read some examples, here are some examples from my own work that are unlike work readily available in the standard literature:
Articulate Vision: A Structuralist Reading of "Kubla Khan", Language and Style, Vol. 8: 3-29, 1985, https://www.academia.edu/8155602/Articulate_Vision_A_Structuralist_Reading_of_Kubla_Khan_. This is a revised and somewhat shorted version of my 1972 master’s thesis.
Heart of Darkness: Qualitative and Quantitative Analysis on Several Scales, Version 5, Working Paper, October 27, 2019, https://www.academia.edu/8132174/Heart_of_Darkness_Qualitative_and_Quantitative_Analysis_on_Several_Scales_Version_5. This is an eclectic collection of remarks about Conrad’s novel, but there is a formal analysis that is crucial, one that takes paragraph length as its point of departure. The longest paragraph is both structurally and thematically central.
The Gojira Papers, April 15, 2014, https://www.academia.edu/7905287/The_Gojira_Papers. This is about the 1954 Japanese film that started the Godzilla franchise.
Here is a theoretical and methodological account of some length (23,000 words, 11 diagrams):
Literary Morphology: Nine Propositions in a Naturalist Theory of Form, PsyArt: An Online Journal for the Psychological Study of the Arts, August 2006, Article 060608, https://www.academia.edu/235110/Literary_Morphology_Nine_Propositions_in_a_Naturalist_Theory_of_Form.
"Now, Large Language Models present a watershed moment for literary studies-they're unavoidable and demonstrate sophisticated capabilities. A cohort of younger scholars using corpus linguistics and computational methods may represent a bridge between computational and literary approaches."
ReplyDeleteNo textbooks are required. We will mainly use course notes (.pdfs below)
Computing
In this course, we will develop many powerful analytic tools. Equally important is the ability to implement these tools on a computer. The instructor and TAs use Python or Matlab, and all examples in class will be in these two languages.
You are good with languages
...
(Lecture 24) Cauchy Integral Formula (notes)
(Lecture 25) ML Bounds and examples of complex integration (notes)
(Lecture 26) Inverse Laplace Transform and the Bromwich integral (notes)
https://faculty.washington.edu/sbrunton/me564/