In 1963, J.M. Coetzee published a curious poem in a University of Cape Town student magazine. Titled “Computer Poem,” the piece of writing was the first public acknowledgment of Coetzee’s ongoing experimentation with computer-generated literature. Such experimentation had begun a year earlier when Coetzee had moved from South Africa to England to work as a programmer for IBM. “A chess player, mathematics graduate, and aspiring poet” (Roach 312), Coetzee proved an attractive candidate to companies like IBM who needed employees with both creativity and a healthy respect for formalism to thrive in “the black art of programming” (Ensmenger 40). Working during the evening on the same IBM 1401 computer that occupied his days, Coetzee explored the potential of generative literary computation. “Computer Poem” was the result of a rudimentary algorithm that instructed a computer to generate poems from an 800-word lexicon. Of the 2,100 poems that the computer produced during its half-hour runtime, Coetzee saw potential in frustratingly few. But, with the right kind of editorial intervention, some of the outputs could be reworked into something passable.
(Coetzee, “Computer Poem” 12)
All of his subsequent experiments would show similar results. At best, the computer would produce a provocative series of words that gave the impression of a poem. But it would be a quasi-poem that always demanded a good degree of (human) revision for it to begin to exhibit qualities of “the literary.” This is because, as Coetzee would write years later, “if we are trying to generate lines that have the unexpectedness and originality of ‘real’ poetry, we have, by definition, to rely on some formalized system of chance” (Coetzee, “Surreal Metaphors” 22). Coetzee’s “formalized system of chance” was to be found in the (pseudo-)random way in which his algorithm formed poetic collocations. But time and again he found that the chance of a computer creating an interesting poetic collocation from his basic algorithm was rather low. Like everything else, poetry has its own rules that are more than a little difficult to navigate. Adhere too closely to such rules, and a poem runs the risk of being labeled unoriginal; stray too far, and it becomes unintelligible nonsense. While recognizing that the poem of highest aesthetic value is, by this economy of things, the one that walks with a certain grace between the potential pitfalls of the commonplace and the absurd, Coetzee was left with hundreds of pages of output through which he diligently searched for moments of interest.
It was, though, during this time that Coetzee began to realize that those pitfalls of the commonplace and the absurd also described a largely unwritten grammar upon which the poetic and, by extension, the literary sits. Indeed, the observation formed the core of Coetzee’s doctoral thesis on the English fiction of Samuel Beckett, in which he interrogated the “grammar” of the Irish writer’s work in the hope of accounting for his turn away from English as a literary language. In line with the claims of other stylisticians, Coetzee expected his stylostatistical analysis of Beckett’s texts, which mapped “syntactical units, syllabicity, sentence length, rarity of words, and even Watt’s chains of logical speculation” (Roach 328), to offer up an objective examination of the works in a way that would put “criticism on a scientific basis” (Fish 109), rather than leave it the intuitive, impressionistic, and therefore eminently subjective art it had always supposedly been. However, as Rebecca Roach notes, “Coetzee ultimately concludes that the promise of this kind of objective critical insight is a hollow one” (328). Quoting from Coetzee’s dissertation, Roach continues, “For Coetzee, stylostatistics oversimplifies because it is ‘dominated by a metaphor of linearity, a conception of language as a one-dimensional stream extending in time’” (328). Thus, it becomes increasingly clear to Coetzee as he conducts his analysis of Beckett’s writing that our essential experience of the literary is one of “incessant recursion”—that “as we read we are continually reformulating formal hypotheses to account for what we are reading and what we have read” (qtd. in Roach 328). And stylostatistical analysis simply had no means of accommodating such a sense of recursion.
It is true that aside from a very small circle of specialists working in the field, when Coetzee was completing his thesis in the late 1960s it was not at all clear how computation might respond to such a demand for recursion.Footnote1 Although recursion became a design option in the 1960s, a robust version that went beyond simple tail recursion was not something that could be implemented in abstract programming languages (such as ISP, ALGOL, PL/I and Logo) until the 1980s. And even then, the programming language with which Coetzee was familiar could not accommodate it.Footnote2 Perhaps this, then, is why in an interview with David Attwell in the early 1990s Coetzee would characterize his interest in computational literary studies—both creative and critical—as “a wrong turning,” “a false trail [that] didn’t lead anywhere interesting” (Coetzee, Doubling 22).
Yet, at the same time that Coetzee was choosing to back away from computational approaches to literature, great strides were being made in domesticating recursive functions. Following a paper that introduced the first algorithm for backpropagation (Rumelhart et al.), Jeff Elman described the implementation of the first Simple Recurrent Network (SRN) in his 1990 essay, “Finding Structure in Time.” As James McClelland explains:
The paper was ground-breaking … since it was the first to completely break away from a prior commitment to specific linguistic units (e.g. phonemes or words), and to explore the vision that these units might be emergent consequences of a learning process operating over the latent structure in the speech stream. (McClelland 155)
Put simply, Elman’s SRN effectively ended the suspicion that computational approaches to language were forever destined to remain a one-dimensional instrument for simply counting words. The backpropagation algorithm, regardless of its guise (the simple recurrent network, the recurrent neural network, the convolutional neural network, or the long short-term memory network, to name but a few), was pivotal because it gave computational models the “memory” that Coetzee had realized was fundamental to the act of reading. Extended, in turn, by the introduction of Transformer architectures in 2017,Footnote3 such backpropagation networks allowed those who undertook computational analyses of literary works to engage directly for the first time with what George Steiner thought of as “the fundamental ‘looseness’ of natural language” found in such works (Steiner 239). To this end, the mathematics of conditional probability that underscored such computational networks not only decoupled number from the discrete and the definite that it continues to enjoy in the popular imagination, but by necessary consequence also pacified Coetzee’s later concern that such approaches could only breed a form of binary thinking. In distinction to encouraging a life lived through “a sequence of YES-NO decisions” (“Literary Thinking” 1152), the quantitative analysis of conditional dependencies showed that decisions or inferences could be made even (especially) in the presence of uncertainty. Put otherwise, conditional probability provided a principled framework for both acknowledging uncertainty in an environment of unstructured complex data and reasoning about relationships within that environment even when the information was incomplete. Here, computational approaches to textual analysis demonstrate the capacity to shift YES-NO decisions into the realm of SOMETIMES.
All of this is to say that such advances in computational approaches now mean that the contemporary literary critic has a powerful ally in the act of critique—an ally that continues to insist that Coetzee’s sense of an unwritten grammar that directs the possibilities of literary interpretation is an essential condition of all texts. To be clear, it is not that there is a single universal grammar that sits at the base of all literature, but rather that each individual text is conditioned by a complex assemblage of relations that emerge from nothing more esoteric than the language of the text itself. Gilles Deleuze and Félix Guattari have written about books most explicitly in this way. At the beginning of A Thousand Plateaus, they write:
A book has neither object nor subject; it is made of variously formed matters, and very different dates and speeds. To attribute the book to a subject is to overlook this working of matters, and the exteriority of their relations. It is to fabricate a beneficent God to explain geological movements. In a book, as in all things, there are lines of articulation or segmentarity, strata and territories; but also lines of flight, movements of deterritorialization and destratification. Comparative rates of flow on these lines produce phenomena of relative slowness and viscosity, or, on the contrary, of acceleration and rupture. All this, lines and measurable speeds, constitutes an assemblage. A book is an assemblage of this kind, and as such is unattributable. (3–4)
The notoriously idiosyncratic language of Deleuze and Guattari perhaps risks distancing the reader too quickly from the assertion that a book cannot be satisfactorily understood as either a singular, coherent, unitary form or as that which pursues a singular, coherent, unitary idea. But those familiar with the work of Mikhail Bakhtin, Roland Barthes, Julia Kristeva, Michel Foucault, or Maurice Blanchot (among others) would surely recognize the tenor of such a statement. Regardless, when Deleuze and Guattari write of “dates and speeds” as the factors by which a text is composed, they do so to exemplify a new ontology that allows for such a determination. It is an ontology of “essential forms” that are distinguished from other forms “solely by movement and rest, slowness and speed” (Deleuze and Guattari 254).Footnote4 Adding a little more texture to the notion of such forms, Deleuze and Guattari explain that “they are not atoms, in other words, finite elements still endowed with form. Nor are they indefinitely divisible. They are infinitely small, ultimate parts of an actual infinity, laid out on the same plane of consistency or composition” (254). In other words, essential forms are states of movement that describe moments of meaning. The notion of the date for example, which Deleuze and Guattari highlight in their description of the book, recognizes that the trajectory of those moments of meaning sometimes coalesce and in so doing assume an observable “density.” Here, philosophies of identity are, properly speaking, philosophies of density.Footnote5 Nonetheless, it is the presence of such densities that makes possible “a semiotic transformation that [is] a pure act or incorporeal transformation—November 20, 1923” (Deleuze and Guattari 82).Footnote6 The assertion here is that dates serve to envelope convergences of diverse themes, events, and complex historical processes in such a way that they are transformed into singular points of grand meaning. Translated to the context of the literary text, one might say that observable densities of essential forms (dates) make visible the meaning (speed) of the text in hand.Footnote7 A Deleuzo-Guattarian book is an assemblage of this kind. It is of vectors and densities, neither of which are to be mistaken for emergent properties of the act of reading for they are, in fact, essential properties of the text itself.
Understood in this way, to elucidate such vectors and densities is not only to elucidate the grammar of a text but also to render visible the vectors of meaning that constitute its very fabric.Footnote8 Put another way, to expose the “dates and speeds” of a text is to reveal a text’s network of relations and, for that, cratonic moments of meaning. The claim is provocative and enticing—that there is an objective structured “meaning” of a text that can be identified through the concatenation of the densities that are produced by the convergences of vectors in the vector space of the book. While the mathematics tells us that such a description of meaning can never be exhaustive, such moments of meaning will be always and only of the text and will condition what the critic is able to say about the text. To test this hypothesis, this paper draws its methodology from the field of computational literary analysis and enters into conversation with J.M. Coetzee’s short novel, Foe.Footnote9 The analysis begins from the premise that a text can indeed be profitably coerced into a vector space—precisely that which is envisioned and demanded by Deleuze and Guattari’s conceptualization of the book. And once coerced in this way, a text can then be subjected to rigorous statistical and algorithmic analysis that has the potential to reveal latent semantic structures that are, at the very least, statistically significant.Footnote10 More importantly, it will become clear that such latent structures constitute the complex terrain of the principled framework of meaning of a text.
Let us begin, though, at the beginning. When Coetzee began his first computational analysis, he did so by looking at the frequencies of word use in a text. If we do the same with Foe, then we get an immediate intuitive sense of the way in which densities might contour a text. After removing high frequency function words,Footnote11 one can easily determine the ten most frequently employed words in Coetzee’s novel (see ).
Table 1. Most frequent words in Coetzee’s Foe (function words excluded).
The result is perhaps unsurprising to those who know Coetzee’s novel well.Footnote12 Friday is of course a major character in the novel, as is Cruso and the author Foe. That the novel is widely regarded as a meditation on the act of literature—specifically, the power dynamics of authorship (understood in the widest possible sense)Footnote13—also allows the informed reader to have anticipated the appearance of words such as “say,” “know,” and “story.” After all, these words are intimately linked to the act of storytelling. On the other hand, the high frequency of “island” is easily explained by the fact that Coetzee’s Foe is a rewriting of Daniel Defoe’s Robinson Crusoe. It would surely be something of a curiosity if characters did not persistently reflect (in one way or another) on the landscape of their internment. In any case, these are the terms that form the initial “density” of Coetzee’s novel—something that one might visualize by mapping the co-occurrence of words (see ).
Figure 1. A feature co-occurrence network of Coetzee’s Foe.Footnote14
Although an interesting network that gives a first glimpse of the way in which certain densities look in Coetzee’s novel, it is a network that most would agree lacks necessary definition. The problem at hand is that raw frequency counts such as this are not the best metric for determining the significance of a word to a text, otherwise the function words which were originally discounted from the initial frequency count would always prove to be the most significant.Footnote15 A rather more nuanced metric to determine statistically significant words in a text that takes into account the relative triviality of ultra-high frequency words in a corpus is the term frequency-inverse document frequency (TF-IDF) weighting.Footnote16 In brief, this weighting divides how frequently a term occurs in a document by how many times it appears across a corpus of documents. below gives those terms that emerge as (most to least) statistically significant as one changes the size by which Coetzee’s Foe is partitioned.Footnote17
Table 2. Top-weighted words of Coetzee’s Foe, by partition.
There are a number of points to note about these results. First is the appearance of a number of new words to those found in the original word-frequency list.Footnote18 In addition to “friday,” “cruso,” and “foe,” the TF-IDF weighting of the text reveals the persistent significance of “tongue” and “write” to the text. Undoubtedly, the image of Friday’s tongue, and the speculation that surrounds how he came to lose it, is one that dominates the text. But it is interesting that the TF-IDF weighting algorithm raises this term from twenty-seventh on the raw word-frequency list to such a position of prominence.Footnote19 Similarly, the TF-IDF algorithm raises the word “write” from thirty-third on the raw word-frequency list and in so doing stresses its statistical significance to the text. It is a more direct indication to the critic, perhaps, that Coetzee’s novel concerns the act of authorship than those terms identified by simple raw word-frequency analysis—“say,” “know,” and “story.”
It is also interesting that the change in partition size reflects a change in the top-weighted features of the novel. Intuitively, this makes sense. As the size of each partition increases, the number of sections that constitute the “corpus” of the novel decreases, and so each individual word assumes a different “weight” across the text. However, one might also interpret this as indicative of the different intensities that inflect Coetzee’s novel. Thus, a high frequency examination of the novel—for example, one described by the partitioning of a text into sections of 200 words—results in a cluster of terms that is dominated by the characters of Foe and Cruso, the image of Friday’s tongue, and, more broadly, the nature of writing. But as the frequency changes, which is to say, as the high frequency examination of the novel becomes a low frequency examination, significance moves away from the figure of Cruso and the image of Friday’s tongue and moves toward the enigmatic nature of communication and expression (“slate,” “play”) and the curious character of Amy (which necessarily reinvigorates the story of Susan Barton’s lost daughter). It is as if the more sibylline features of the novel belong to an analysis that is performed at a certain distance—a lower frequency—from the text itself.
The TF-IDF weighting of the text also highlights the fact that the words “foe” and “write” are particularly significant to the novel as a whole. After all, these are the only terms that are recorded as top-weighted features of Coetzee’s Foe regardless of the size by which the text was partitioned. In other words, regardless of whether one is conducting a particularly close analysis or not, one must seemingly always return to the significance of these words in a critique of the novel. To ignore these terms highlighted by the algorithm is to ignore the significance of the act of writing to the novel. To this end, another density emerges for inspection that centers on these terms.
In this way, the hunt for densities that describe Coetzee’s novel encourages one to transition away from the examination of single words and move toward the explicit examination of word clusters. As the noted linguist J.R. Firth memorably quipped, “You shall know a word by the company it keeps!” (11). One important way of determining the “company” that a word keeps in a text is to run the text itself through a word embedding algorithm.Footnote21 Such an algorithm is capable of mapping in high-dimensional vector space the relative position of each word to every other word that appears in a text. Once configured, such a map can then be interrogated in order to find the words that cluster around a particular term. shows what that map looks like for Coetzee’s Foe once the high-dimensional vectors that describe each word have been translated into two-dimensional vectors (which can then be represented on x, y axes). The table below () lists the five words that most often appear in a list of top-weighted features of the novel, and the ten terms that are closest in association with them—that is to say, the ten terms that begin to shape the meaning of each word as it presents in Foe.
Figure 2. Word clusters (40 words) around the terms “foe” (red) and “write” (blue) in a word embedding map of Coetzee’s Foe.Footnote20
Table 3. Nearest ten words to each of the top-weighted words of Coetzee’s Foe.
The tongue of course belongs to “friday,” but the other terms that are closest to it in the vector space of the novel begin to suggest what the image means in the text. As one might suspect, the image of the tongue is closely aligned with those words that form the core of Susan Barton’s speculation about how it was “lost”—that it was lost to a “knife” of a “slaver” or, more worryingly, “cut” from his mouth by his “master,” Cruso.Footnote22 Similarly, Friday is described by words that most would have perhaps anticipated. None is more obvious than those that put him in relation to “cruso.” Again, there is an insistence on the notion of speaking or expression (“say,” a word that is, of course, also of high significance specifically to the image of Friday’s tongue), but this cluster of words also shows rather less obvious ways in which the text brings Friday and Cruso together. For example, both characters sit in close relation to the word “now,” which enlivens the sense that the eternal present suffered by Cruso is also lived by Friday. Barton makes clear that Cruso lived in a debilitating state of psychic exhaustion as the consequence of the past and the future fatally collapsing into the present.Footnote23 Yet, this may also be a way of understanding the enigmatic actions of Friday. The space of “now” is the space of immediacy and, as the etymology of the word reminds the critic, that is the space of the nothing between.Footnote24 There is nothing between Friday and Cruso, nothing between Friday and the island, nothing between Friday and the weather, nothing between Friday and the music he plays. Put simply, Friday stands as a point at which there is no space between the subject and object or the subject and its expressive acts. With this in mind, one is perhaps invited to recall what is surely the most striking passage from the hallucinatory ending of the novel: “But this is not a place of words. Each syllable, as it comes out, is caught and filled with water and diffused. This is a place where bodies are their own signs. It is the home of Friday” (157). Whether this is a positive or problematic rendition of (radical) otherness has been the focus of critical debate since the novel’s publication,Footnote25 but nevertheless, it seems that this enigmatic understanding of Friday is something that is only made possible because Coetzee’s writing of Friday does not allow one to drive a wedge between subject and object.
Friday finds relation to the figure of Foe in a similar way, though the word that draws them together is not “now” but “take.” Importantly, this word is not used in the narrative to describe Friday taking things from others, but rather because others—most often Susan Barton—either “take” things while in his company or attempt to order Friday to take something from them (orders that Friday invariably ignores). Not only does this return one to the crucial image of Friday’s tongue, which of course was taken from him by one means or another, it also brings into sharp relief Friday’s inscrutability. Playing no role in the economy of give and take, there can be no communication, no communion. Therefore, not to take is to remain the radical other, which is to say that which seduces rather than that which desires.Footnote26
If things are taken from Friday, then they are “given” to Foe. Think here of the stories, permission to write, choices, and other things that the careful reader will see given to Foe by others. As such, one is made to think of Foe as very much part of the economy of communication that cannot envelope Friday. This is an important observation because, as Deleuze and Guattari remind us, communication is one that is premised solely on the exercise of power. And Coetzee writes the character of Foe in such a way that he cannot be thought otherwise to this exercise of power. “Language is made not to be believed but to be obeyed, and to compel obedience,” Deleuze and Guattari assert. They continue:
Spengler notes that the fundamental forms of speech are not the statement of a judgment or the expression of a feeling, but “the command, the expression of obedience, the assertion, the question, the affirmation or negation,” very short phrases that command life and are inseparable from enterprises and large-scale projects. (76)
It is Foe who is given the authority by others to “command life,” which is to say, to marshal the thoughts and emotions captured in the “letters,” “papers,” and “words” of others, in such a way that Susan Barton’s story will eventually become The Life and Strange Surprizing Adventures of Robinson Crusoe, of York, Mariner.Footnote27 However, if Barton fails to see her own story published, if the reader witnesses “the political failure (in gendered terms) of Susan’s authorship” (Worthington 28), the rather uncompromising suggestion at hand is that it is because she has ultimately given her story to Foe and, in this way, abandoned it to the multitude of forces that constitute the circuit of literature at the beginning of the eighteenth century. Friday does not play the game of give or take.
In this way, then, the densities derived from word embeddings have the potential to enliven previously unexplored relationships between characters in a novel. But importantly, they also have the capacity to reveal unacknowledged relationships between specific sections of a novel. If one persists with Coetzee’s novel partitioned into 400-word sections as seen above,Footnote28 then the semantic relationship between various sections of the novel begins to produce another kind of density (see ).
Figure 3. A section network of Coetzee’s Foe, partitioned at 400 words.Footnote29
A close investigation of this network shows that the most “connected” sections in Coetzee’s Foe—which is to say, the sections that are semantically related to the most other sections of the novel—are sections 11, 12, and 18. Clearly, these sections are important to the novel in one way or another. But one can begin to get a sense of why these sections of the novel are so pivotal to the narrative by once again mapping the ten most statistically significant words from each section (see ).
Table 4. Top ten words (weighted by TD-IDF) of the most semantically related sections of Coetzee’s Foe.
The TF-IDF weighting of these sections of the novel draws the reader’s attention to the idea of law, the notion of punishment, and … spoons! While law and punishment are easy for most readers familiar with the novel to reconcile, one might initially struggle to understand how these high moral and ethical forms sit in relation to spoons. Yet, there is a relationship to be exposed here. It is a relationship that is revealed through an analysis of the other sections to which these key pivotal sections are connected (see ).
Table 5. Most semantically connected sections of Coetzee’s Foe and the sections to which they are related (with highest TF-IDF weighted feature).
When read together, what emerges from this network of key connected sections is the way in which the novel itself encourages the reader to begin thinking about its central theme—namely, the nature of writing, of storytelling, of literature caught in the event of its production.
If one concentrates on the three key pivotal sections identified by the distance matrix (sections 11, 12, 18) and a number of those sections that are connected by more than one such section (see ), then a slightly different discussion about the nature of writing emerges to the one that is commonly found in the critical literature on Coetzee’s novel. Section 11 is dominated by the idea of law. On the nature of law, Cruso tells the enquiring Susan Barton that “Laws are made for one purpose only … to hold us in check when our desires grow immoderate. As long as our desires are moderate we have no need for laws” (36). There is then perhaps more significance to the fact that Barton later mistakes Foe for a lawyer than has previously been attributed to it.Footnote31 If the writer can be mistaken for a lawyer, is it because writers surreptitiously apply the immoderate desires of others to the page in the same way that a lawyer seeks to make practical the abstract domain of the law? If so, then Robinson Crusoe is to be read in precisely this way—the immoderate desire of (European) man to dominate, to conquer, to own, to possess not only the natural world but also the men who walk upon it.
Figure 4. Simplified network showing the connections of key sections in Foe, with the highest TF-IDF weighted term for each section.Footnote30
Given this, it is significant that section 11 is also connected to section 12. Section 12 contains the discussion of punishment held between Barton and Cruso. But when read in association with the previous section, it becomes clear that section 12 deepens the way in which one is encouraged to think the context in which the writer commits such immoderate desires to paper. That the discussion between Barton and Cruso on the idea of punishment recalls the notion of original sin is important. In a moment of frustration early in the novel, Barton states that “This land is our punishment, this island and one another’s company, to the death” (37). It is of course a minor rehearsal of the divine punishment meted out to humanity in the Fall. However, those familiar with Coetzee’s own critical writing will note that this sense of things captures rather well his own reading of Defoe’s novel. In an essay that introduces a rather underappreciated way to think of the figure of Friday in Robinson Crusoe, Coetzee writes, “What, Crusoe muses, if mankind had been twice and separately created—in the Old World and the New—and what if in the New World there were no history of rebellion against God? What if Friday and his countryfolk are unfallen creatures, in no need of redemption?” (Stranger Shores 25).
It is an intriguing light in which to read Friday in Foe—a creature who was in no need of redemption. Either the enigmatic, the incomprehensible, and seemingly uncomprehending, figure of Friday in Coetzee’s novel undoes this speculation because he does not (or cannot) articulate his “self-evident good heart” in the way that Defoe’s Friday does and in so doing enlivens the justification for European colonialism, or Friday is a purer breed of human, something angel-like, which is to say a superior being that the fallen humans of the “Old World” cannot comprehend. If this is the case, then Friday makes clear that colonial discourse has always had its account of the world very badly wrong. Understood like this, Europe’s others are the chosen people, and it is therefore the Europeans who need to be brought a little closer to God. The civilizing mission, its theological basis, its philosophical basis, and the literature that served as its vehicle was part of a web of lies—a literature of the tongue with which, as Susan Barton notes, “we jest and lie and seduce” but noticeably do not tell the truth (85). The civilizing mission seduced, but it did not tell the truth. This is the true condition of European literature that is articulated in Foe; this is the mire in which the European writer writes.
It is, then, this observation that colors the other sections linked to section 12—those sections that, by and large, are concerned with interrogating the relationship between word and world. The keyword in section 6 is “firewood,” a noun that is used in the novel to indicate the fact that Friday appears to have a different categorical understanding of the world to that of Cruso and Barton. Barton remarks, “I found it strange that Friday should not understand that firewood was a kind of wood, as pinewood is a kind of wood, or poplarwood; but I let it pass. Not till after we had eaten, when we were sitting watching the stars, as had grown to be our habit, did I speak again” (21).
The point is repeated later in the novel in section 48 with the rather more abstract notion of “freedom.” Raised not in terms of pursuing freedom but rather in questioning what the word means, it becomes clear that freedom only assumes something like meaning when one is cognizant of bondage. Putting this equation together, Barton finally realizes that if Friday does not understand the notion of freedom, it is precisely because he does not know the idea of bondage. And if Friday does not know bondage, it is because, as discussed earlier, he stands as a point of immediacy at which there is no space between the subject and object, the subject and its expressive acts. As a man who enjoys freedom as an absolute condition, Friday does not play (and neither does Cruso) Hegel’s game of recognition that Barton tries to introduce to the island and which Defoe writes into Robinson Crusoe—a novel that one should properly think of at this point as an iteration of the literature of the tongue.Footnote32
Perhaps it is this lack of differentiation in the figure of Friday that explains why Barton’s attempts to teach Friday the English language in section 18 of Coetzee’s novel end in failure. She writes, “I hold up a spoon and say ‘Spoon, Friday!’ and give the spoon into his hand. Then I say ‘Spoon!’ and hold out my hand to receive the spoon; hoping thus that in time the word Spoon will echo in his mind willy-nilly whenever his eye falls on a spoon” (57).
The Saussurian triumvirate of sign, signifier, and signified relies on difference, and such difference relies on a space that can inaugurate a distance between that which is and that which is not. Friday’s inability or lack of desire to appreciate such distance means that he cannot make (or has no intention of making) the connection between the object and word. Here, the arbitrary relationship of the signifier and signified in the sign is fully exposed. That Barton later sells “a case of spoons … to buy necessaries” (87–88) surely symbolizes the end of her attempts to induct Friday into the English language and, perhaps, the harsh realization that the story she is struggling to see told is a story that language itself is not capable of telling.
Here, then, the structured meaning of Foe begins to emerge. These important pivotal sections of the novel certainly condition the way in which the central interest of the text is to be initially thought by the reader. At its most simple, these sections make clear that the act of writing is not an epistemologically transparent activity—an observation that most critics of the novel have made. But read under the direction of computational analysis, these sections form an intricate network of associations that encourages the critic to investigate a very particular sense of European literary history, the nature of storytelling, and the relationship of language to truth. For this reason, and regardless of whether the critic fully recognizes the significance of the interplay between these sections, the fact is that they constitute a density within a principled framework that outlines the interpretative possibility of the novel.
Ever more definition is given to the principled framework of Coetzee’s Foe as the latent semantic relationships that exists between words in the novel are disclosed. Indeed, a number of significant word clusters (or “topics”) that transit the novel are illuminated when one submits the text to a topic modeling algorithm (see ).
Table 6. Five most coherent topics in Coetzee’s Foe.Footnote33
Call these “tight densities,” highly coherent topics are clusters of words that are bound together in a text but do not necessarily recur over the text at large. Typically, these are groups of words that belong to a few specific passages in the text, as can be seen from the following graph (), which shows when and to what extent (known as the theta score) topic 24 (“play tune”) occurs throughout the novel.
Figure 5. Theta for topic 24 in Coetzee’s Foe.
Although prefigured in an early passage of the novel where Barton hears Friday “play over and over again on his little reed flute a tune of six notes” (27–28), it is clear that the cluster of words that describe topic 24 predominantly belong to a later scene in which Barton (again unsuccessfully) attempts to join Friday in playing his ostinato in the hope of enacting some kind of a communion with him (95–97). The signal of this topic is of such strength that it marks a density in the text that readers would be unwise to ignore. Clearly, there is something of significance to the interpretation of the novel to be noted in this cluster of words, if only for the fact that this density produces the most defined and therefore, perhaps, most abiding image of the text.
In distinction to such tight densities are topics that are significant for the fact that they are ubiquitous to the novel. Indeed, it is perhaps sensible for one to think of such prevalent topics as capturing the persistent interests of a text or, put another way, those aspects of a text that are constantly put in front of the reader (see ).
Table 7. Five most prevalent topics in Coetzee’s Foe.
Such “recursive densities” as these might lack the coherence of other topics, but they nonetheless represent persistent aspects of a novel. Again, mapping the theta score of the most prevalent topic in Coetzee’s novel makes clear just how persistent the topic of Friday’s lost tongue is in the novel at large (see ).
Figure 6. Theta for topic 25 in Coetzee’s Foe.
In the same way that the attentive critic cannot in good faith ignore the image of Friday “spinning and dancing and singing” that is illuminated by an examination of the tight densities of Coetzee’s novel (118), the critic cannot ignore the significance of Friday’s lost tongue to a critique of the work as a whole. Indeed, an analysis of the recursive densities in the text shows that the text itself is insistent on this image. Read in concert with word densities and sectional densities, such topic densities ultimately bring into focus Deleuze and Guattari’s vision of a book understood in terms of the dynamic movement of essential forms. Like any other book, Coetzee’s Foe is an assemblage of a multitude of vectors that at times coalesce to form densities. And it is these densities that not only constitute the interest and objective structured meaning of the text but also necessarily condition all interpretative acts of the literary critic. In short, such densities determine what can be said of a novel—they allow a text to mean. Some densities encourage the critic to read a text in a particular way while others compel a very specific reading. Those densities that cannot be ignored, and those that otherwise direct the reader to specific sections of the novel, to particular words, phrases, or themes, or to certain characters, are precisely that which constitute the complex principled framework of meaning that conditions a novel. As such, those few densities that this essay has highlighted in Coetzee’s Foe should be understood for what they are—the first indication of a profound network of meaning that sometimes encourages, sometimes directs, but always structures our interpretation of a text.
Disclosure Statement
No potential conflict of interest was reported by the author(s).
Additional information
Notes on contributors
Grant Hamilton
Grant Hamilton is Associate Professor of English Literature at the Chinese University of Hong Kong. He teaches and writes in the areas of twentieth-century world literatures, African literatures in English, literary theory, and computational literary studies.
Notes
1. See Daylight for an interesting history of “recursion” in programming.
2. Rinderknecht writes, “With even more abstract programming languages, fully-fledged recursion became a design option, first advocated in print by Dijkstra (1960) and McCarthy (1960), and implemented in LISP, ALGOL, PL/I and Logo (Martin, 1985, Lavallade, 1985), with the notable exceptions of Fortran and COBOL” (89). Coetzee worked with Fortran while at IBM in the 1960s.
3. See Viswani et al.
4. Actually, the ontology is not “new” but is rather one borrowed from Spinoza. See Deleuze and Guattari (255). Nonetheless, it is this ontology that Deleuze and Guattari develop throughout their philosophical work.
5. In distinction, Deleuze’s philosophy is widely regarded as the philosophy of difference. See, for example, Bell, Grosz, and Patton.
6. The date to which Deleuze and Guattari refer here recalls the day when the Rentenmark replaced the Papiermark as the official currency of Germany.
7. This is not at all the same as claiming that such densities constitute the full and final meaning of all interpretative acts performed upon the book. Rather, such densities constitute the condition of possibility for all such interpretative acts.
8. Meaning, here, is defined in terms of a semantic coherence that emerges from patterns of word co-occurrence. See Ouyang et al.
9. J.M. Coetzee’s Foe was chosen for this study because it is both well-known and reassuringly complex at both a structural and linguistic level. I say “reassuringly complex” here in recognition of one of Nan Da’s major complaints against computational literary studies—that it reduces the various “complexities” of literature. See Da.
10. The software environment for statistical computing and graphics used in this paper is R. See R Core Team.
11. On the significance of removing high frequency function words (also known as “stopwords”) such as articles, common adverbs, conjunctions, pronouns, and prepositions see Gerlach et al. and Jockers (131).
12. The raw word-frequency list has been produced after the text has been lemmatized. It is perhaps noteworthy that part of this process included setting the text to lower case. This is why character names, such as Cruso and Friday, are given in lower case in this table.
13. See, for example, Attridge, Macaskill and Colleran, Poyner, and Spivak.
14. This feature co-occurrence network shows the relationship between the 200 most frequent words in Coetzee’s Foe. The frequency count threshold is 0.85, and those features which have not met this threshold have been omitted.
15. For example, the word “the” is written 2,322 times in Coetzee’s novel.
16. On TF-IDF, see Manning et al.
17. To be clear, it is necessary to partition a single text into sections in this way in order to fabricate a “corpus” of sections—a corpus that is necessary in order to find the inverse document frequency of a word. In this example, Coetzee’s Foe is partitioned into 200-word sections (producing 101 sections in total), 300-word sections (producing 68 sections in total), and so on. Obviously, the novel partitioned into 200-word sections is of a higher “resolution” than the novel partitioned into 1,000-word sections. However, that is not to imply that results from a 200-word partition are in some way “better” than results from a 1,000-word partition. Rather, what this indicates is that different words come to the fore as the resolution of examination changes.
18. As with the raw word-frequency list, this table has been produced after the text has been lemmatized.
19. The word “tongue” is used 75 times in Coetzee’s novel.
20. This visualization of a word embedding map of Coetzee’s Foe is a 2-d expression of a 50-dimensional object. As such, one should be careful when drawing conclusions premised on the proximity of terms on this particular graph. See Melville.
21. The algorithm used is from the text2vec package. See Selivanov, Bickel, and Wang.
22. After all, Susan Barton tells the reader that “A knife, let us remember, was the sole tool Cruso saved from the wreck” (84).
23. For example, Cruso clearly rejects the past when he responds to Barton’s suggestion that he should try to keep a journal of his time on the island. He tells Barton, “Nothing is forgotten … Nothing I have forgotten is worth the remembering” (17). The rejection of the future is implicit in his refusal to facilitate rescue. It is Barton’s keen observation on Cruso’s demeanor following her question about why he hasn’t attempted to escape the island that is important: “‘And where should I escape to?’ he replied, smiling to himself as though no answer were possible” (13). Without a sense of the possible, the future must collapse.
24. “Immediate” from the late Latin immediatus, meaning “nothing intervening.”
25. See, for example, Kim Worthington’s excellent commentary on, and intervention in, the debate.
26. Jean Baudrillard writes on the significance of this “reversal of desire” enveloped by seduction in a somewhat different context.
27. The full title of Daniel Defoe’s 1719 novel reads: The Life and Strange Surprizing Adventures of Robinson Crusoe, of York, Mariner: Who lived Eight and Twenty Years, all alone in an un-inhabited island on the Coast of America near the Mouth of the Great River of Oroonoque; Having been cast on Shore by Shipwreck, wherein all the Men perished but himself. With an Account how he was at last as strangely deliver’d by Pyrates. Written by Himself.
28. Admittedly, the choice to pursue a reading based on a 400-word (that is, 51 section) partitioning of Foe is arbitrary. One might just as well choose a partitioning resolution of 200 words or 1,000 words. However, a 400-word partition does seem to sit in that happy ground between high frequency and low frequency examinations of the novel, returning results typically seen in both domains.
29. This network visualizes a distance matrix that is generated from partitioning the novel into 400-word sections and then coercing the result into a document feature matrix. See Benoit et al.
30. This simplified distance network shows the relationship between the “nodes” (400-word section) with the most “edges” (connections) in Coetzee’s Foe. The minimum frequency threshold is 0.98.
31. Talking of Foe, Barton writes, “I admit, when I first laid eyes on you I thought you were a lawyer or a man from the Exchange” (47–48).
32. I refer of course to Hegel’s master-slave dialectic. Terry Pinkard makes the “game” extremely clear in the introduction to his translation of Hegel’s Phenomenology of Spirit.
33. The k value determines the number of topics with which the topic modeling algorithm works. While the k value is usually a supervised part of this algorithm (that is, set by the analyst), the k value used here is also selected by algorithm. For an unsupervised way of determining k, see Jones “3. Topic Modeling.” The topic modeling package used here was textmineR, and the k value was 40 (Jones textmineR). The labels given to each topic were produced natively by the topic modeling algorithm.
Works Cited
- Attridge, Derek. “Oppressive Silence: J.M. Coetzee’s Foe and the Politics of the Canon.” Decolonizing Tradition: New Views of Twentieth Century “British” Literary Canons, edited by Karen Lawrence, U of Illinois P, 1992, pp. 212–38.
- Baudrillard, Jean. Fatal Strategies. Translated by Phil Beitchman and W. G. J. Niesluchowski. Semiotext(e), 2008.
- Bell, Jeffrey A. Philosophy at the Edge of Chaos: Gilles Deleuze and the Philosophy of Difference. U of Toronto P, 2006.
- Benoit, Kenneth, and Akitaka Matsuo. Spacyr: Wrapper to the ‘spaCy’ ‘NLP’ Library. 2020, ver. 1.2.1, https://CRAN.R-project.org/package=spacyr. Accessed 3 Feb. 2023.
- Benoit, Kenneth, and Adam Obeng. Readtext: Import and Handling for Plain and Formatted Text Files. 2020, ver. 0.80, https://CRAN.R-project.org/package=readtext. Accessed 20 Oct. 2022.
- Benoit, Kenneth, et al. “Quanteda: An R Package for the Quantitative Analysis of Textual Data.” Journal of Open Source Software, vol. 3, no. 30, 2018, pp. 774. doi:10.21105/joss.00774. Accessed 19 Apr. 2023.
- Coetzee, J. M. “Computer Poem.” The Lion and the Impala, vol. 2, no. 1, 1963, pp. 12–13.
- Coetzee, J.M. Doubling the Point: Essays and Interviews. Edited by David Attwell. Harvard UP, 1992.
- Coetzee, J.M. Foe. Penguin, 1987.
- Coetzee, J.M. “On Literary Thinking.” Textual Practice, vol. 30, no. 7, 2016, pp. 1151–52.
- Coetzee, J.M. Stranger Shores: Essays 1986–1999. Secker & Warburg, 2001.
- Coetzee, J.M. “Surreal Metaphors and Random Processes.” Journal of Literary Semantics, vol. 8, no. 1, 1979, pp. 22–30.
- Da, Nan Z. “The Computational Case Against Computational Literary Studies.” Critical Inquiry, vol. 45, no. 3, 2019, pp. 601–39. doi:10.1086/702594.
- Daylight, Edgar G. “The Advent of Recursion in Programming, 1950s–1960s.” U of Amsterdam EPrints, 2010, https://eprints.illc.uva.nl/id/eprint/383/1/PP-2010-04.text.pdf. Accessed 10 Nov. 2022.
- Deleuze, Gilles, and Félix Guattari. A Thousand Plateaus: Capitalism and Schizophrenia II. Translated by Brian Massumi. Athlone P, 1999.
- Elman, Jeff. “Finding Structure in Time.” Cognitive Science, vol. 14, no. 2, 1990, pp. 179–211. doi:10.1207/s15516709cog1402_1.
- Ensmenger, Nathan L. The Computer Boys Take Over: Computers, Programmers, and the Politics of Technical Expertise. MIT P, 2010.
- Firth, J. R. “A Synopsis of Linguistic Theory, 1930–1955.” Studies in Linguistic Analysis, edited by J. R. Firth, Blackwell, 1962, pp. 1–32.
- Fish, Stanley. “What is Stylistics and Why are They Saying Such Terrible Things About It?” Approaches to Poetics, edited by S. Chatman, Columbia UP, 1973, pp. 109–52.
- Gerlach, Martin, et al. “A Universal Information Theoretic Approach to the Identification of Stopwords.” Nature Machine Intelligence, vol. 1, no. 12, 2019, pp. 606–12. doi:10.1038/s42256-019-0112-6.
- Grosz, Elizabeth. “Bergson, Deleuze and the Becoming of Unbecoming.” Parallax, vol. 11, no. 2, 2005, pp. 4–13. doi:10.1080/13534640500058434.
- Jockers, Matthew. Macroanalysis: Digital Methods and Literary History. U of Illinois P, 2013.
- Jones, Thomas. “3. Topic Modeling.” 2021, https://cran.r-project.org/web/packages/textmineR/vignettes/c_topic_modeling.html. Accessed 20 Oct. 2022.
- Jones, Thomas. TextmineR: Functions for Text Mining and Topic Modeling. 2019, ver. 3.0.4, https://CRAN.R-project.org/package=textmineR. Accessed 20 Oct. 2022.
- Macaskill, Brian, and Jeanne Colleran. “Reading History, Writing Heresy: The Resistance of Representation and the Representation of Resistance in J. M. Coetzee’s Foe.” Contemporary Literature, vol. 33, no. 3, 1992, pp. 432–57.
- Manning, Christopher, et al. Introduction to Information Retrieval. Cambridge UP, 2008.
- McClelland, James L. Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises. 2nd ed., DRAFT, 2015, https://web.stanford.edu/group/pdplab/pdphandbook/handbook.pdf. Accessed 11 Mar. 2023.
- Melville, James. “Uwot: The Uniform Manifold Approximation and Projection (UMAP) Method for Dimensionality Reduction.” 2020, ver. 0.1.10, https://CRAN.R-project.org/package=uwot. Accessed 11 Mar. 2023.
- Ouyang, Long, et al. “Semantic Coherence Facilitates Distributional Learning.” Cognitive Science, vol. 41, no. S4, 2016, pp. 855–84. doi:10.1111/cogs.12360.
- Patton, Paul. Deleuze and the Political. Routledge, 2000.
- Pinkard, Terry. Introduction. The Phenomenology of Spirit. Edited by G. W. F. Hegel, 1807. Translated by Terry Pinkard. U of Cambridge P, 2018, pp. ix–xxxvi.
- Poyner, Jane. J.M. Coetzee and the Paradox of Postcolonial Authorship. Routledge, 2009.
- R Core Team. R: A Language and Environment for Statistical computing. R Foundation for Statistical Computing. 2020, https://www.R-project.org/. Accessed 20 Oct. 2022.
- Rinderknecht, Christian. “A Survey on Teaching and Learning Recursive Programming.” Informatics in Education, vol. 13, no. 1, 2014, pp. 87–119. doi:10.15388/infedu.2014.06.
- Roach, Rebecca. “J.M. Coetzee’s Aesthetic Automatism.” Modern Fiction Studies, vol. 65, no. 2, 2019, pp. 308–37. doi:10.1353/mfs.2019.0015.
- Rumelhart, David, et al. “Learning Representations by Back-Propagating Errors.” Nature, vol. 323, no. 6088, 1986, pp. 533–36. doi:10.1038/323533a0.
- Selivanov, Dmitriy, et al. Text2vec: Modern Text Mining Framework for R. 2020, ver. 0.6, https://CRAN.R-project.org/package=text2vec. Accessed 20 Oct. 2022.
- Silge, Julia, and David Robinson. “Tidytext: Text Mining and Analysis Using Tidy Data Principles in R.” JOSS, vol. 1, no. 3, 2016, pp. 37. doi:10.21105/joss.00037. Accessed 20 Oct. 2022.
- Spivak, Gayatri Chakravorty. “Theory in the Margin: Coetzee’s Foe Reading Defoe’s ‘Crusoe/Roxana’.” English in Africa, vol. 17, no. 2, 1990, pp. 1–23.
- Steiner, George. After Babel: Aspects of Language and Translation. Oxford UP, 1998.
- Vaswani, Ashish, et al. “Attention is All You Need.” Advances in Neural Information Processing Systems, vol. 31, 2017, https://arxiv.org/pdf/1706.03762.pdf. Accessed 5 May 2023.
- Wickham, Hadley. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag, 2016.
- Wickham, Hadley. tidyr: Tidy Messy Data. 2021, ver. 1.1.3, https://CRAN.R-project.org/package=tidyr. Accessed 20 Oct. 2022.
- Wickham, Hadley, et al. dplyr: A Grammar of Data Manipulation. 2021, ver. 1.0.6, https://CRAN.R-project.org/package=dplyr. Accessed 20 Oct. 2022.
- Worthington, Kim. “‘A Place Where Bodies are Their Own signs’: Re-Reading J.M. Coetzee’s Foe via Gubar, Spivak, Parry and Levinas.” Philia & Filia, vol. 2, no. 1, 2011, pp. 5–40.