NOTE: I presented the following essay, which continues the argument begun in an essay I have previously posted to HASTAC, at the Futures of American Studies Institute in June 2014. I have made a few changes following comments received during the Institute.
Can an Algorithm be Disturbed?
Symptomology, Intrinsic Criticism, and the Digital Humanities.
James E. Dobson
Structuralism and the Systemic Criticism
The desire to produce scientific account of literature, such as we are finding in both the turn to social sciences and within the digital humanities, is not new. Prior to this moment of re-theorized attention to the surfaces of texts, prior even to the advent of symptomatic reading, those working within the humanities dreamed of the possibilities of a scientific criticism. Northrop Frye’s Anatomy of Criticism (1957) made the polemical case for a systematic and scientific criticism derived from an inductive reading of literature. Frye outlines the expansive scope of his approach by creating, explicitly modeled after Aristotle, “a theory of criticism whose principles apply to the whole of literature and account for every valid type of critical procedure.”  This approach would work, he argues, because like the sciences he assumes the existence of an order of nature, an order of words, lies behind the enterprise known as literature as a coherent whole. Discovering the laws governing this order becomes the task of the critic. This understanding enables Frye to read widely across numerous literatures extract major modes and archetypes and to produce a categorization of all these into a single organizing schema. Individual texts are then brought, either by Frye or another future critic, into the law of the schema and used to establish minor variations on a theme. This is what he believes to make his system scientific: each revision made by critics and scholars builds progressively on the entire body of prior humanistic research. Above all Frye’s schema works in pursuit of what he sees as a set of unalterable structural principles that can guide future criticism and reading. It is a “genuine” mode of criticism—to be differentiated from the accretion of judgments made by literary taste makers, or what he calls meaningless criticism—that follows the research models provided by science and “progresses toward making the world of literature intelligible” (9).
What Frye’s Anatomy creates is a “conceptual universe” in which all of literature can be plotted, located, and mapped. His schema are ultimately less rigid than we might expect and one particularly important feature of Frye’s system is its own open-endedness: he intended that categories beyond those major labels that have made his book famous—the mythic, generic, and archetypal—would come to improve and even obsolete his theory. Yet as Geoffrey Hartman notes in an important critique, the mythic holds a central place in Frye’s system. Hartman selects this category because he believes that myth will demonstrate a blindspot in Frye’s system. Like our present moment, the possibility of a scientific criticism was deployed as a “surface” against the concept of depth. This aspect leads Hartman to call Frye’s method a “flattening out” of literature in opposition to the “depth criticism and depth psychology,” of their shared historical moment. Frye’s Anatomy is ultimately spatial. The Anatomy charts and maps the literary terrain and in so doing it drops what Hartman believes to be an important dimension: time. Claiming that “literature unfolds in time rather than quasi-simultaneously in space” (33), Hartman criticizes Frye’s understanding of temporality and literary history. The system evades the question of historical development by treating all literature as essentially co-occurring and finding little use for concepts like tradition, influence, and inheritance. This leads to Hartman’s greatest concern. He worries that such “[a]rchetypal analysis can degenerate into an abstract thematics where the living pressure of mediations is lost and all connections are skeletonized” (30-31). Without the literary-historical network, the system that takes its place finds a series of dead-end nodal points.
But this network is precisely what myth requires and what it reworks. There are no “pure” forms of myth in Frye’s system only multiple appearances of historically situated myths. Hartman writes, “a writer does not confront a pure pattern, archetype, or convention, but a corpus of tales or principles that are far from harmonized” (37). Hartman was right about the disappearance of history from Frye’ssystem as a cause for concern. History is, after all, key to humanistic inquiry. Without the nuanced understanding of the ways in which ideas and representation unfold throughout time, literary critics would be the social scientists that Frye seeks to distance himself from. Frye’ go so far as to reject the sociological reading advocated by critics like Heather Love and by some contemporary digital humanists that his work anticipates:
I understand that there is a Ph.D. thesis somewhere which displays a list of Hardy’s novels in the order of the percentages of gloom they contain, but one does not feel that that sort of procedure should be encouraged. The critic may want to know something of the social sciences, but there can be no such thing as, for instance, a sociological “approach” to literature. (19)
What seems most interesting about the many contemporary digital humanities projects when compared to prior forms of scientific criticism is the deep focus on history. Indeed, these projects seem to have anticipated such critiques and made the temporal dimension central to their inquiry. Very large-scale archives such as Google’s Google Books enable heretofore impossible readings across the longue durée of literary history. Tools like the ngram viewer make the historical tracking of word or phrase genealogies through almost all of print history a trivial task. Thus these projects could be understood as answering Hartman’s main complaint of the systemic and scientific approach to literature, and yet the hermeneutical critique of what Hartman refers to as the “sweet science” remains helpful advice to the would-be scientific literary critic.
Frye’s archetypal system shares much with the structuralist critique of the 1960s and 70s. Both approaches seek to organize all of literature into well-defined categories and take as a founding assumption the existence of an ordered world that could be illuminated through progressive critique. Like Frye, the structuralists explicitly referred to their practices as a science. This was in part because structuralism came to the humanities from the social sciences, but also due to its status as a classificatory methodology. In his well-known essay “From Science to Literature,” Roland Barthes describes the structuralist commitment to taxonomy:
structuralism, by virtue of its method, pays special attention to classifications, orders, arrangements; its essential object is taxonomy, or the distributive model inevitably established by any human work, institution, or book, for there is no culture without classification; now discourse, or ensemble of words superior to the sentence, has its forms of organization; it too is a classification, and a signifying one. 
For many structuralism was essentially a formalism. Like Frye’s system it erased history and like the new historicism that would eventually follow structuralism, it operated synchronically rather than diachronically. It formed schema based on the presupposition of a closed-world of meaning that enabled the taxonomization texts and the components of a text. The forms or categories, however, were not necessarily considered objective and arguments over selection and categorization prevented the production of any real definitive readings.  It understood itself as an improvement upon what has become known as the New Criticism primarily through Ferdinand de Saussure’s introduction of the division between langue and parole; understanding each individual textual object, the closed-world of the poem as theorized by the New Criticism, as an individual instance of enunciation, what Saussure called parole, and its object of criticism the system, the langue, which produces the grounds of possibility for the individual poem, structuralists desired a larger object of critique and common language to be used by a community of scholars.
Structuralism was yet to be fully embraced when, in 1968, Sigurd Burckhardt produced a critique of the mechanical tendencies of both archetypal and structural criticism. Burckhardt’s essay appeared in an unexpected location; he publishes his “Notes on the Theory of Intrinsic Interpretation” as an appendix to his Shakespearean Meanings. This essay seeks to revitalize literary criticism primarily through his division of intellectual labor into two categories: explanation and interpretation. His own categorizations enable him to make an unusual defense of hermeneutics by arguing against the understanding of interpretation as the description of the way in which a work of literature “works.” Interpretation is not the accounting for why a text follows certain mythic laws or archetypes but a mode of discovery that takes as its primary object the text itself. At the same time, if the surface reading and digital readings of the present reject the conception of “depth hermeneutics” so too as does Burckhardt’s theory. For what calls his reading practice into action is not the deeply buried symptom, the sense of deep meaning to be revealed by the critic, but something on the surface that troubles our ability to give a structuralist account of the text.
Burckhardt argues that insofar as it has a methodology capable of supporting a theory, science is intrinsic. By this Burckhardt means, like Northrop Frye, that science understands the universe as ordered and organized by a set of discoverable laws. Like the religious belief that science has made obsolete, the entire world postulated by empiricism is subject to intrinsic analysis. Everything must have a place and meaning. Interpretation, according to his account, “would mean the attempt to know the law of a poem solely from the poem itself, on the necessary assumptions of the infallibility of the poem. Explanation, on the other hand, would mean the attempt to demonstrate how parts of a poem obey an already known, established principle” (298).
Explanation and interpretation map onto, respectively, extrinsic and intrinsic analysis. Frye’s conception of the “order of words” necessitates an intrinsic approach and this shares some assumptions within Burckhardt. Burckhardt, however, places his emphasis on the hermeneutical act called into being by the intrinsic method. While the residual New Critical focus on poetry and the single poem draws Burckhardt toward the poem as his unit of interpretation, there is no reason why this procedure should be limited to poetry or even a single novel. Indeed it seems entirely likely that Burckhardt’s hermeneutical approach is exactly what we need for the large archives studied with digital approaches. One does not have to necessarily follow Sigurd Burckhardt in his belief that each textual object is a “unit” or world with knowable rules in order to understand the force of his critique of certain strains of structuralist thinking. What I mean to say is that Burckhardt’s conception can revitalize the digital humanities-cum-structuralist reading practices that we find at the present moment.
Yet there are ways to turn the major presupposition of digital reading techniques such as machine learning back against itself. Classification, whether machine or human derived, fits observed data or objects into distinct categories. The important difference between human and machine classification, however, is what draws us to categorize data into categories and our doubt about this categorization. The algorithm assumes that all data will “fit”; within machine learning there are concepts to label the degree to which data fits into categories: we call any potential uncertainty within classification confusion or simply “error.” The outlier, that peculiar object not belonging to the domain of one law or another, might present some only difficulty to categorize for the algorithm, but it is of high interest to the human interpreter because it represents a problem. Burckhardt draws our attention to the way in which when we are reading we encounter something that he calls a “stumblingblock” that becomes the occasion for analysis:
What occurs, then, when I really do interpret? Something which in principle is very simple. I read a poem and the poem “speaks to me.” At the same time, however, or perhaps only after several readings, I get the impression that I have not yet grasped its true significance. Something ‘disturbs’ me. What it is that will ‘disturb’ me is never predictable. It may be a ‘discrepancy’ (a contradiction, sometimes purely factual, which seems to reside in the poem itself); it may be an apparent whim of the poet or a seemingly inappropriate word; it may be configuration whose meaning is obscure; or it may be (as with Hölderlin’s late hymns) that the coherence of the whole completely escapes me. Finally any conception of the poem which contradicts my own may also disturb me in this sense. (301)
Burckhardt here faults the dominant contemporary theory during this time, structuralism, for not paying enough attention to those objects that do not fit within pre-existing strategies. This critical science has pushed these difficult to categories elements aside in favor of generalizations and categorizations.
So the question remains, can an algorithm be disturbed? My guess is that when they work digital reading algorithms resist and reduce disturbance, but when they fail, they might be considered disturbed. It is in another essay found in Beyond Formalism thatGeoffrey Hartmann describes interpretation as requiring either the location of a space in between the text or the opening of that space by the critic: “Interpretation is like a football game. You spot a “hole” and you go through. But first you may have to induce that opening. The Rabbis used the technical word patach, “he opened,” for interpretation” (255). For Hartmann, literature is special because it has the capacity to sustain the hole. Interpretation exists within a space that might be thought of as in between the bits of language. When Hans-Georg Gadamer produced a definition of the hermeneutical circle he makes the point that the concept “is based on a polarity of familiarity and strangeness.” The poles in Gadamer’s circle are less rigidly defined (as should be obvious from the juxtaposition of poles and circles) than allowed by much digital humanities work. When we allow our algorithms to familiarize that which is fundamentally ambiguous and we risk turning our work, the project of literary criticism, into what Burckhardt would call explanation—not that there is anything wrong with that, but it might be the case that what we think we are doing is not what we are in fact doing. We can use machine learning and other computer-aided reading techniques to open holes by deploying the algorithm against itself, but ultimately interpretation is an interesting and compelling narrative of how one deals with being poked by a text, by being disturbed.
 Northrop Frye, Anatomy of Criticism: Four Essays (Princeton: Princeton University Press, 1971), 14.
 Roland Barthes, “From Science to Literature” The Rustle of Language. Translated by Richard Howard (Berkeley: University of California Press, 1989), 6.
 For a general theoretical and historical background on structuralism, its main currents of thought, and adoption within the American academy, see Robert Scholes, Structuralism in Literature: An Introduction (New Haven: Yale University Press, 1974). Robert Scholes describes the assumption of an a priori order of the world: “The perception of order or structure where only undifferentiated phenomena had seemed to exist before is the distinguishing characteristic of structuralist thought” (41). See also, Jonathan Culler, Stucturalist Poetics (Ithaca: Cornell University Press, 1975).
 Hans-Georg Gadamer, “On the Circle of Understanding” Hermeneutics Versus Science?: Three German Views. trans and ed. by John M. Connolly and Thomas Keutner(Notre Dame, IN: University of Notre Dame Press, 1988), 68-78. See also, Truth and Method. trans. Joel Weinsheimer and Donald Marshall (New York: Crossroad, 1989), 295.