Margaret Atwood has enjoyed a genre and medium spanning career very few other authors have achieved: her published fictional novels include The Handmaid’s Tale (1985,) a post-apocalyptic work of social criticism which has recently inspired a popular television show; Oryx and Crake (2003,) a well-known work of science fiction; and Alias Grace (1996,) a work of historical fiction which has been made into a six-part television series. This list, which is far from complete, does not include Atwood’s many plays, collections of poetry, works of critical non-fiction, or the transcriptions of various interviews she has given over her career. Atwood is a prolific writer who has achieved success in many, wildly divergent, genres and forms of writing: she is the authorial “Jill-of-all-trades.” However, while Atwood’s large body of work provides no shortage of material for critical analysis, the size of her corpus makes it difficult—if not impossible—to analyze the minutia of her career as a unified whole. Moreover, while it is certainly not impossible for an individual critic to closely examine all of Atwood’s works, it is difficult for one person to detect lexical, thematic, and topical similarities which span the many genres, forms, and mediums Atwood has contributed to over her career—a single critic simply cannot quickly survey the amount of text required to make such claims with authority. In other words, authoritative claims spanning Atwood’s entire career require a level of rigor the individual critic is incapable of achieving; no one can keep—within the biological constraints of a single human mind—the volume of divergent themes and the vast quantity of text required to meaningfully examine an author as productive as Margaret Atwood in their head simultaneously without some form of augmentation. To illustrate this issue, a paper written on the thesis that “Margaret Atwood’s entire career as a writer has been focused on the traumatic effects of war as they pertain to the women on the home front” would be hard-pressed to collect and keep track of the number of examples required to substantiate this central claim. Furthermore, a critical paper comprised of thousands of quotations in support of such a thesis would leave little room for a critical analysis and, at best, amount to little more than a catalog of examples with an analysis tacked on at the end. As a result, these types of universalizing theses cannot depend on the tools of literary criticism alone. However, by combining the mathematics of a statistician, the reductivist analysis of cultural sociology, and the subjective theory of the modern literary critic, literary criticism can gain the ability to make the broad, universalizing claims mentioned above without the need to support such claims with hundreds or thousands of unique examples. The tools of mathematics and machine learning can provide literary critics with the augmentations they need to move beyond the exemplary form of criticism they are accustomed to, and into an empericised form of criticism that enables the objectivity of scientific discourse within the subjective discourse of literary criticism.
The desire to make authoritative claims about an impossibly large corpora of divergent texts is the central purpose behind what is often called augmented (or algorithmic[i]) criticism—a form of criticism associated with the nascent, ill-defined field of “digital humanities.” Stephen Ramsay, a well known digital humanist, proposes an algorithmic criticism which “seeks, in the narrowing forces of constraint embodied and instantiated in the strictures of [computer] programming, an analogue to the liberating potentialities of art” (Ramsay x). According to Ramsay, an algorithmic criticism should “unite the reductive calculus of computation to the broader act of critical narrative” (xi). Put simply, augmented criticism should provide the critic with tools that enable a level of mathematical rigor without also supplanting the need for subjective analysis within critical discourse. The Augmented Criticism Lab[ii] (ACL,) a project I have spent the last year developing, is such a tool. The ACL toolkit is specifically designed for large-scale analysis of arbitrary volumes of text across any number of mediums; it enables a single critic to query a corpus of trillions of words in mere seconds and acts as a guide to areas within the corpus to focus a critical study. However, it is important to note that augmented analysis does not perform the critic’s job: these tools enable new forms of criticism but do nothing to remove the critic from the critical process as such.
The purpose of this paper is twofold: in part one, I will describe, in non-technical language, the augmented methods I employed to explore Margaret Atwood’s corpus of fiction, poetry, short stories, and critical non-fiction as a unified whole. Part one will focus primarily on correlating the results of a traditional, subjective analysis with the results of an augmented critical analysis. In part two, I will use the results of my algorithmic criticism to explore the central question of this paper: how does Margaret Atwood work the war?
Part 1 – Methods
The process of examining a large volume of text must always begin with the tedious task of data entry. The profound importance of accurate data entry cannot be understated: errors in the input data will render a resulting analysis invalid—any claims structured around incorrect data are themselves incorrect. In my analysis of Margaret Atwood, I began with parsing of few of her better-known works of fiction into the ACL database; in this initial phase I entered The Robber Bride, The Handmaid’s Tale, Alias Grace, and The Blind Assassin. To handle the data entry automatically, I wrote a simple Python script to parse the Kindle (mobi) editions of each text into sections the ACL database can handle. To enable a granular analysis of the texts, I split each work into small “chunks”—in this case chapters—so each chunk can be handled as a separate entity. These divisions allow a work to be examined with a high level of granularity and avoids treating a longer work of fiction as a single, unified entity. This initial phase was primarily focused on ensuring the database could handle longer works of fiction; up to this point the database had only been used to examine poetry. After carefully confirming the dataset against the source texts, I continued to the next phase of the process, topic modelling.
Topic modelling is a highly complicated mathematical process; the central theory governing how it operates is called Latent Dirichlet allocation (LDA).[iii] Any attempt to explain how LDA works in less than a few thousand words—the requisite graduate degree in mathematics aside—is destined to fail. Nevertheless, in very simple terms LDA assumes that each semantic unit (usually a sentence) in a piece of text has a definite topic. For example, the phrase “my love is like a rose” contains two topics, love and rose. In an LDA model, these two topics are added to a list of topics containing the topics of all sentences in the entire piece of text, and each topic is then ranked via a complex set of factors. These factors include frequency of occurrence, part of speech, and centrality. Nouns and proper nouns are always ranked higher in this matrix of factors than other parts of speech. This is because, in a purely grammatical sense, nouns are usually the subject of a given sentence, and proper nouns are most often the subject to which a sentence refers. Conversely, verbs, adverbs, and adjectives are actions and modifier terms rather than topics, and, as a result, often—but not always—rank much lower than nouns. Put simply, the LDA topic modeller determines which terms in a body of text are the most likely topics one can use to describe the input text as a whole. It is a process of inference whereby terms are ranked against the likelihood they describe the entire body of the input text. Below is an example of a topic model, specifically generated for Chapter One of Margaret Atwood’s The Robber Bride. Note that rank 0 of any topic model contains the most central topic, and its probability is shown both as a single factor—without any of the other topics in consideration—and comparatively. The percentage inside the brackets is the topic’s comparative probability.
|Rank||Probability||topics (word frequency)|
|0||85.1% (3.6%)||zenia (8) tell (2) point (2) hand (2) develop (1)|
|1||3.2%||enjoy (2) roz (1) crumble (1) despite (1) sense (1)|
|2||3.7%||would (8) know (2) charis (1) market (1) real (1)|
|3||53.3%||war (2) definitive (2) event (2) one (2) trouble (1)|
|4||3.1%||Friend (1) hole (1) across (1) october (1) interest (1)|
|5||2.7%||return (1) gulf (1) shift (1) dissolve (1) simultaneous (1)|
|6||3.5%||much (3) moment (2) choice (2) preface (2) toxique (1)|
|7||4.3%||tony (7) mix (2) someplace (2) begin (2) ontario (1)|
|8||3.4%||lie (3) nothing (2) rubble (2) disaster (2) slight (1)|
|9||2.8%||kill (2) hurt (2) blow (1) marriage (1) power (1)|
These charts can be difficult to understand for several reasons: the probabilities do not appear to be ranked logically, the word frequency is often quite low, and, in the chart above, the item in rank 0 comparatively scores roughly 15 times lower than the item in rank 3. Regardless, anyone who has read the first chapter of The Robber Bride would know the topic of “Zenia” is most certainly the primary topic, but how does a computer determine this? Similarly, why is the item in rank 3, with a probability of 53.3%, not in rank 0? In exceedingly simple terms, the computer determined that Zenia is the most likely topic because the word “Zenia” appears in relation to other high-ranking potential topics more often than any other term: the word “Zenia” is central (or related) to many of the other topics and therefor must be the central topic of Chapter One.
There is little doubt that Zenia is the central topic of Chapter One of The Robber Bride, and it would be difficult to argue otherwise. To take this one step further, can the topic modeller justify the claim that Zenia is the primary topic of the entire novel? There are, of course, non-computational methods for justifying this claim: as readers, we have no difficulty in determining Zenia’s central role in The Robber Bride, but can a computer reach the same conclusion without human input? If the topic models for the entire novel are examined, the word ‘zenia’ appears in 648 out of a total 2916 model predictions,[iv] an occurrence of 22.2%. This is direct empirical evidence that “Zeina” is one of the central topics in The Robber Bride. Here, the topic modeller has substantiated something subjectively known about a text in mathematical terms, providing objective evidence for a subjective observation. To make this interesting, what happens when the topic modeller looks at the central question of this paper: how does Margaret Atwood work the war? To keep things simple, if the same algorithm as the one used above is used to search for the word “war” rather than “Zenia,” 162 hits are returned, representing roughly 5.6% of the total predictions. If the same search is run on The Blind Assassin, 198 hits are returned out of a total 6318 predictions, representing about 3.1% of the total. These results are interesting, but they fail to provide anything of value to a critical analysis of either text. If topic modelling is going to augment the critic, these numbers alone are simply not enough.
Part 1.a – Robbers, Assassins, and the Adventures of Penelope: Women Against War
This section will begin to demonstrate the synthesis of traditional English criticism and the augmented tools described above. Specifically, this section will examine a common theme in Atwood’s The Robber Bride, The Blind Assassin, and The Penelopiad: all three texts examine the traumatic effects of war, not on the men who fight in them, but on the women who are left to pick up the pieces when their brothers, husbands, and fathers come home as strange shadows of the men they used to know or, in many cases, never come home at all.
In The Robber Bride, the three female protagonists and Tony’s mother have suffered profound traumas as a direct result of the traumatic effects war had on the men in their lives. Tony’s conflicted relationship with her parents—and their similarly conflicted relationship with each other—is the most profound example of this effect. In his suicide note, Tony’s father Griff remarks “that Tony’s high school graduation was all he’d been waiting for. … [And after her graduation ceremony,] he went home and shot himself in the head” (Atwood, Robber, chap.23).[v] While Griff’s suicide is certainly a consequence of his experiences with war, his final act is not the most troubling effect his war trauma has on Tony’s life. While the exact cause of an individual’s trauma is often difficult to determine, much of Tony’s trauma can be traced to how Anthea, Tony’s mother, often minimizes Griff’s experience of the war: Anthea claims “[Griff] didn’t go through [the war] the way she did. Her parents’ house in London was destroyed by a bomb during the Blitz and her parents were both killed” (chap. 22). Anthea’s antagonism of Tony’s father is deplorable and an understandable expression of her own trauma; effects of her own traumatic experiences aside, when Anthea asks Griff if the gun he brought home from the war “feels liberated” she reveals a level of sociopathic narcissism Mr. Trump would struggle to achieve (chap. 22). Tony’s lifelong obsession with war, and her decision to become a war historian, is likely a product of her father’s unwillingness to speak of his experiences during WW II in the face of Anthea’s lack of compassion—which is likely the result of her own traumatic experience of war. Tony acknowledges this when she remarks that she often “sits on the floor, looking at her father and wondering about the war, which is such a mystery to her but which appears to have been decisive in her life” (chap. 22). In this passage, Tony sees her father’s experiences as a mystery she needs to understand, not because of what happened to her father per se, but because of the effects of the war on her own life. Moreover, Tony’s desire to understand her relationships through the lenses of combat and war does not end with her parents: Tony seeks to explain her relationship with West in the same war-centric frame. When Zenia appears to return from the dead, Tony struggles with how to disclose her apparent resurrection to West, stating “the enemy is already within the walls. … [And,] the personal is not political… the personal is military. War is what happens when language fails” (chap. 6; emphasis added). Here, Tony sees her relationship with West as a product—and direct consequence—of a conflict with Zenia, and she rationalizes this conflict through the same process she uses to rationalize the conflict between her parents, war; Griff’s inability to discuss his trauma, coupled with Anthea’s narcissism and lack of empathy, have caused Tony to see all relationships with men as a product of conflict; for Tony, relationships, be they romantic, platonic, or otherwise, are always-already entangled with her own traumatic experiences of war.
Roz and Charis have less obvious war-related traumas, but their respective experiences are no less profound than Tony’s. Roz’s father’s life had always been a near-total mystery to her, and the truth of his experiences with war is something Roz often obsesses about in the novel. After his death, Roz interrogates one of her father’s long-time friends, whom she calls Uncle George: he describes her father as “a fixer before the war, he was a fixer in the war, and after the war he was also a fixer” (chap. 45). However, this explanation does little to assuage Roz’s curiosity; even the revelation that her father was a crook is no surprise—she had always suspected the source of her family’s sudden wealth was illegitimate. Zenia uses Roz’s curiosity about her father’s life—claiming “he saved [her] life… during the war”—to gain more access to Roz’s personal life (chap. 41). This is illustrated in how “it grates on Roz that Zenia has this news [about her father] and Roz does not. It’s as if her father has left something in his will, some treasure, to a perfect stranger, some drifter he’d met in a bar, and nothing for his own daughter. Didn’t he know how much she wanted to know?” (chap. 41). Another important point is how the troubled relationship between Roz and her parents, especially her relationship with her mother, forms the basis for much of her war-related traumas. During the scene of her childhood flashback, Roz describes her father as “the great unknown” (chap. 42). She recalls that her father was “doing important, secret things that could not be spoken about. They were war things, even though the war was over” (chap. 42). However, Roz doubts her mother’s claims and, consequently, locates the source of her trauma as her mother’s distortion of her father’s role in the war rather than the war itself.
Roz and Tony both share an insatiable desire to understand the relationship between the traumatic experience of war and their respective fathers, but what about Charis? To explore this, before Charis, there was Karen, and “Karen’s father was killed in the war when Karen wasn’t even born yet, leaving Karen’s mother to bring up Karen all by herself” (chap. 33). Here, Charis perceives the loss of her father as a direct product of war, but as with Tony, Charis’s mother is the true source of her war-related trauma; Charis’s mother tells her raising a child alone is “very hard, practically impossible,” and forces Charis to see herself as a burden at a very young age (chap. 33). Moreover, Charis’s mother abandons her at her grandparent’s farm, as “she was heading into a bad patch of nerves” (chap. 33). These so-called “nerves” are a condition which Charis describes as “the fault of the war” (chap. 33). Yet again, as with Tony and Roz, the traumatic experience of war—even though Charis’s mother may have lied about her father’s identity—plays a profoundly important role in Charis’s life. Moreover, the genesis of the Charis / Karen split is a direct result of her mother’s “nerves,” which Charis perceives as a product of war; had it not been for her father’s death, her mother would never have abandoned her, and Uncle Vern would never have created the divided personality of Charis / Karen during the traumatic rape scene where Karen’s “skin comes open like the dry skin of a cocoon, and Charis flies out” (chap. 35). Here, the very existence of Charis as an individual is portrayed as a direct result of the war-trauma her mother suffered when Charis’s father, allegedly, died during the war.
Everything noted above indicates a distinct constellation of topics are at work in Atwood’s The Robber Bride. Atwood structures the narrative of Tony, Roz, and Charis around the effect war had on their fathers, but she always directs this trauma through the locus of their mothers. Rather than a direct experience of war-related trauma via their respective fathers, it is their mothers who act as the catalyst of their traumas. For Tony, it is her mother’s constant antagonism of her father that causes her distress; for Roz it is her mother’s misrepresentation of her father’s role in the war the leads her to obsess over the true nature of his experiences; and for Karen it is her mother’s “nerves” that lead to her abandonment, rape, and eventual fracture into Charis. These topical connections provide a means for pursuing an augmented analysis over a much wider scope of works, but finding a pattern in a single text is not enough to substantiate a claim which covers all of Atwood’s corpus. To make such a claim, a similar pattern must be found in other texts: this will help substantiate any claims structured on topic modelling alone. Put simply, if one finds the topics of war, mother, father, and trauma are also related in other works via traditional critical methods, any claims made via augmented critical methods are much easier to substantiate.
In The Blind Assassin, the traumatic effects of war on the women who fight on the home front is central to the entire narrative. Norval Chase, Iris and Laura’s father, “had been wounded at the Somme” and, much to his wife Liliana’s chagrin, “is now an atheist” (Atwood, Blind, chap.“The gramophone”). Rather than accepting Norval’s new-found atheism as a direct result of his traumatic experience of war, Liliana “begged [Norval] to keep his atheism to himself. Then she was deeply ashamed for having asked this – as if what mattered most to her was the opinion of the neighbours, and not the relationship in which [Norval’s] living soul stood to god” (chap. “The gramophone”). Liliana’s ongoing attempts to sooth her husband and understand his trauma have no effect, and Norval would often “climb up into the stumpy turret of Avilion… [and] up there he would talk to himself and slam against the walls, and end by drinking himself numb” (chap. “The gramophone”). Unlike Anthea and Griff from The Robber Bride, Liliana does not attempt to antagonize her husband, but her inability to properly cope with his suffering is a major source of trauma for their children. After Liliana’s death, she leaves her children ill-prepared for the realities of their father’s war trauma: by allowing Norval to spend much of his time “drinking, and… tomcatting,” Liliana unwittingly places her children in the care of a broken alcoholic with no capacity to fulfil his duties as a now single parent (chap. “Bread day”). It is important to note that Liliana is not at fault for her inability to deal with her husband’s trauma, at the time no one could effectively deal with the traumatic effects of war. The purpose of this argument is to draw a connection between the traumas experienced by Tony, Roz, and Charise in The Robber Bride, with the traumas of Iris and Laura in The Blind Assassin. In both novels, Atwood uses the relationship dynamics between mother and father as a locus for exploring the inter-generational traumas experienced by their children. The pattern of topical associations between war, trauma, and parents form a continuity across both works. Nevertheless, finding a common theme in two of Atwood’s novels is not enough evidence to justify any objective conclusions found via topic modelling—more evidence of a pattern is still required.
In Atwood’s Penelopiad, the eponymous Penelope “is thrown into the sea” as an infant because an Oracle tells her father, King Icarius of Sparta, she will “weave… [his] shroud” (Atwood, Penelopiad, 1.1). If it were not for the intervention of some “purple-striped ducks” Penelope would certainly have drowned (1.1). However, Penelope’s struggles do not end with her father’s attempts to kill her: later in the play Icarius remarks that “if you have daughters instead of sons, you need to get them married off as soon as possible so you can have grandsons. The more sword-wielders and spear-throwers you can count on within your own family, the better” (1.6). Here, the connection between the traumas of war and one’s parents—common across many of Atwood’s works—begins to take form: Icarius sees his daughter as little more than a bargaining chip. Moreover, children are portrayed as “vehicles for passing things along” and “through children, alliances are formed” (1.6). In other words, Penelope’s entire life is directed by the exigencies of war; her parents see her as a vehicle for ensuring and perpetuating the martial capacity of her father’s kingdom. For example, Penelope’s husband is chosen after “holding an athletic contest to determine a suitable match” (1.6); her father is only concerned with the martial prowess of Penelope’s future husband—Penelope’s happiness and wellbeing never enter the equation.
Penelope’s war-related trauma may begin with her parents, but her life in Ithica is where the traumatic effects of war, specifically as they apply to women on the home front, become impossible to ignore. After a period where Penelope “had little authority in her new home” and spent most of her time “in anticipation of evenings—evenings in bed with Odysseus,” Helen, Penelope’s beautiful (and married) cousin, unexpectedly appears (1.11). Helen, “the septic bitch,” as Penelope calls her, had eloped with King Priam, and consequently Odysseus must “go to Troy” because of “an oath [he swore] concerning the husband of Helen” (1.12). Odysseus’ sudden departure leaves Penelope alone in her new island home as her husband—the only thing she enjoys on Ithica—goes off to fight a war against the Trojans. Soon after his departure, Odysseus’s mother “wrinkled up like drying mud and then died,” King Laertes “retreated to the countryside,” and Penelope is left to manage her husband’s estate in his absence (1.12). Here, the connection between The Robber Bride, The Blind Assassin, and The Penelopiad takes shape: Odysseus is gone for a long time, and his prolonged absence after the fall of Troy leaves Penelope wide-open to proposals from any eligible bachelor with a desire for the newly created wealth of Odysseus’s estates—wealth Penelope herself had created. These lecherous layabouts force Penelope and her maids into the subterfuge that, in the play’s conclusion, causes Odysseus to hang the maids without first consulting Penelope. In summary, Penelope’s war-trauma-by-proxy plays the same narrative role as the traumas of Roz, Charis, and Tony in The Robber Bride and the Chase children in The Blind Assassin. In all three of these novels the traumatic experiences of the main character(s) stem from the war experiences—or aspirations—of their brothers, fathers, husbands, and mothers. What does this signify? These texts all share a very specific topical similarity: namely, they all focus on the intergenerational effects of traumatic war experiences on the children—especially the female children—of those directly affected by war. Is it possible to extrapolate this topical connection into Atwood’s other works? In the following section, I will begin exploring the potentials of augmented critical analysis across the whole of Atwood’s corpus.
Section 2 – Topic Modelling a Career: How does Atwood “Work the War?”
“Simple word-to-word collocations… do not provide enough information to rise to the level of theme. What is needed in order to capture theme are collocations of collocations on a much larger scale. … latent Dirichlet allocation (LDA) takes us a very long way to fulfilling that need.”Jockers 122
How does LDA topic modelling manage to “rise to the level of theme,” as suggested by Mathew Jockers, above? While LDA models can provide insight into the thematic components of a piece of writing, it is not simply a matter of shunting a large volume of text into a formula and hoping for a positive result. For one thing, an LDA model provides none of the subjective insight required by a good critical analysis. Instead, these models provide a means of reducing and “deforming” a text into forms that enable new avenues for subjective critical discourse. In section one of this paper, I identified a topical similarity in three of Atwood’s works. Specifically, in these texts Atwood explores the intergenerational nature of war-related trauma with a specific focus on the female children of those traumatized, and this topical correlation will serve as the jumping-off-point for an interdisciplinary study of Atwood’s corpus. However, as noted by Jocker’s in the epigraph, a list of collocations provides little insight insofar as the critic is concerned, and subjective criticism cannot rely on the tools of a statistician alone. To be clear, this paper is unequivocally un-statistical: the data are 100% cherry-picked and the graphs are manipulated to favour my subjective interpretations. The purpose of this paper is not to demonstrate the capabilities of a statistical model or mathematical trick; the goal is to explore the possible fusion of natural language processing (NLP) and Franco Moretti’s notion of “distant reading:” a practice “where distance… is a condition of knowledge: it allows you to focus on units that are much smaller or much larger than the text: devices, themes, tropes—or genres and systems” (Moretti, “Conjectures on World Literature” 77). Topic models are used as a means of reducing a text to its thematic components, and subjective criticism is required to explore the relationship between these thematic components and Margaret Atwood’s entire body of works. In other words, the synthesis of a computer’s objective logic and a critic’s subjective analysis provides a means to answer this paper’s central question: how does Margaret Atwood “work the wars?”
In my analysis of Atwood’s The Robber Bride, The Blind Assassin, and The Penelopiad, I noticed a common theme: each work explores the inter-generational consequences of war-trauma on the children of those traumatized. In The Robber Bride, Atwood structures the narrative of Tony, Roz, and Charis around the effect war had on their fathers, but she always directs this trauma through the locus of their mothers. For Tony, it is her mother’s constant antagonism of her father that causes her distress; for Roz it is her mother’s misrepresentation of her father’s role in the war that leads her to obsess over the true nature of his experiences; and for Karen it is her mother’s “nerves” that leads to her abandonment, rape, and eventual fracture into Charis. In The Blind Assassin, Liliana Chase’s inability to properly cope with her husband Norval’s suffering is a major source of trauma for their children; by allowing Norval to spend much of his time drinking Liliana unwittingly places her children in the care of a broken alcoholic with no capacity to fulfil his duties as a now single parent. And finally, the eponymous Penelope of The Penelopiad lives a life perpetually structured around the exigencies of war—her father chooses Odysseus as her husband based solely on his martial prowess, and Odysseus’s long absence to make war with Troy is the cause of most of the play’s conflict. In all three works, the narrative is structured around the trauma-by-proxy experienced by the children of those directly traumatised by war. However, these qualitative similarities are impossible to quantify through the objective lens of a topic model. Per Franco Moretti, if these similarities form “parts of a pattern, then what we must explain is the pattern as a whole, not just one of its phases” (Moretti, Graphs 13). Put simply, before a topic model can quantify a topical connection within Atwood’s corpus each topic must first be reduced to its lexical components. To accomplish this, a list of terms and related terms (or collocations) must be generated for each topical pattern. Moretti describes these collocated list of terms, or “maps,” as a list that “disregards the specificity [of each element] … to focus almost entirely on their mutual relations” (Moretti, Graphs 54). In other words, to detect a topical pattern within any large corpus of works, the topic must first be reduced to a list of collocated terms that disregards the broad context of the work where the topic appears.
How does one go about generating these lists of de-contextualized terms? There is no clearly defined means of generating such a list automatically, but there is a tried and true method of examining a text any critic is already intimately familiar with. This is to say, such lists are created through very close reading and a bit of computational aid. Consider Tony’s story in The Robber Bride: how could Tony’s war trauma be reduced to a set of loosely related terms? A close reading focused on words related to war and trauma as they apply to Tony provides a good starting point, and, to aid in locating further examples, a custom search engine I devised can locate all instances where a series of terms are found in proximity to each other. In any case, this phase of the process depends heavily on traditional close reading; what value is a search engine if you do not know what to search for? Through close reading and proximity searching a short list of terms related to both war-trauma and Tony is devised:
Once a list of terms is identified, my custom search engine generates a key-word-in-context (KWIC) graph of all instances where the terms occur in proximity to each other. Some examples of these KWIC graphs are listed below:
- …teeter, car factories grind to a halt. The war in the Gulf is over. … (Atwood, Robber, chap.57)
- …or the war?” says Tony. “For the battle, it’ll definitely be technology… (chap. 5)
- …“You mean, is there going to be a war?” says Tony. … (chap. 5)
- …So Tony is a foreigner, to her own mother; and to her father… (chap. 22)
- …father and wondering about the war, which is such a mystery to her be on the farm. … (chap. 22)
This list is not complete: a total of 47 examples of these related terms are found in The Robber Bride alone, far too many to list here. After confirming the terms are relevant, this process is repeated for The Blind Assassin and The Penelopiad. Finally, after a list of common battle related words are added, the data is compiled into a list of 565 battle related words[vi] which acts as a guide to locations within Atwood’s corpus to direct further study. To proceed from this point, I must first answer a question I have left intentionally unanswered: what does a collection of war-related terms have to do with an analysis of Atwood’s body of work?
Before I answer this question, I must first explain a limitation of the LDA topic modelling process. Accurate topic models require a minimum amount of text; this limit is not well defined and depends heavily on the type of input material. For example, a short abstract of a paper—around 400 words—is more than enough to generate an accurate topic model: an abstract is specifically written to describe the topic of the paper. In contrast, a short piece of poetry is nearly impossible to topic model accurately—poems are not focused on a topic and are difficult to qualify objectively. As a result, topic modelling and KWIC lists are wholly ineffective at working with poems. Nevertheless, this deficiency is easy to rectify: poems are short. A single critic can manually topic model dozens of poems in a relatively short period of time. For this reason, I identified fourteen of Atwood’s poems from three of her poetry collections which provide further evidence of the topical similarities already discussed above. These poems include “The Loneliness of the Military Historian”where the speaker has, “in the interest of research… walked many battlefields,” as Tony does in The Robber Bride (Atwood, “Lonliness”); “Helen of Troy Does Counter Dancing” where the speaker states “I do give value. \ Like preachers, I sell vision, \ like perfume ads, desire \ or its facsimile. Like jokes \ or war, it’s all in the timing” (Atwood, “Helen”); and “Singing to Genghis Khan” where it is the speaker’s job to “make him [Khan] feel better. \ Then maybe she can get some sleep \ and will not be murdered” (Atwood, “Singing”). At the risk of waxing tautological, in the three collections of Atwood’s poetry I read while researching this paper each contained a considerable number of poems about war qua women’s bodies, roles, or experiences—the fourteen I chose were the most obvious examples, but do not amount to a complete list of all examples. These poems provide further evidence of a topical connection spanning Atwood’s entire authorial career: images and topics on the traumatic effects of war are central to many of her poetic and fictional works. Thus, lists of collocated terms allow a critic to explore how two texts are related to each other via purely lexical means.
Section 2.a: What do the Computers Say?
The observations above provide ample evidence of a topical connection across a small subset of Atwood’s corpus. However, all this evidence was gathered with traditional, subjective means and the aid of a souped-up search engine. How does one move from subjective criticism to augmented criticism, and what does a list of battle related terms have to do with said augmentation? The first step in answering these questions is to topic model as much of Atwood’s corpus as possible: to start The Handmaid’s Tale, Alias Grace, Cat’s Eye, Oryx and Crake, Year of the Flood, and Surfacing were divided into chapters and topic modelled—I feel these texts provide a good survey of Atwood’s career as a novelist. Of the 529 chapters modelled, 96 (or 18%) of the sections contained a topic from the list of war-related topics. From the 96 models flagged as potentially war related, I generated a KWIC list to locate how each word was used in the relevant section of the text. These lists provide the insight required to bridge the gap between traditional criticism, topic modelling, and augmented criticism. To begin exploring how this works, the topic models indicate that Cat’s Eye contains thirteen chapters with a war-related topic. For example, Chapter 22 is described with five topics: easier, perpetual, boy, enemy, and nose. A KWIC lists provides the following snippet of text in relation to these topics:
“…because she’s my enemy. Far from it. I know about enemies. There are enemies in the schoolyard, they yell things at one another and if they’re boys they fight. In the war there were enemies. … Our boys and the boys from Our Lady of Perpetual Help are enemies. You throw snowballs at enemies and rejoice if they get hit. With enemies you can feel hatred, and anger.”Atwood, Cat’s, chap.22
This passage seems to describe a conflict between school children, but the line “in the war there were enemies” indicates these school children are familiar with, or have been directly affected by, war in a literal sense. A search for the words war, house, and boy—inspired by the model of Chapter 12 (carol, rope, bury, boy, and house)—provides a bit of context:
“satisfying word, especially annihilating. Boys say it mostly, to one another to be intentional. I don’t have occasion to use my mean mouth on boys, since on the side of the page. … We are taking the Second World War. The teacher may have been in the war himself, or so rumor goes. On the board he’s drawn is disintegrating block by block, house by house, mantelpieces, chimneys, double beds.”chap.43
This indicates the topical connection described previously is at work in Cat’s Eye: the teacher is described as a potential soldier in the Second World War and has drawn a picture of a town being dismantled “block by block.” But what about the topic of intergenerational trauma? A search for collocations of “mother and war,” and “father and war” reveals that the speaker in Cat’s Eye is experiencing the war alongside her parents; thus intergenerational trauma does not come into play. To explore another of Atwood’s novels, the topic modeller indicates sixteen war-related topics in Alias Grace: Chapter 45 contains the topics chief, dead, father, help, and mr (Mr.); and Chapter 48 contains command, neck, behind, first, and except. A KWIC search of these 10 topics fails to provide any insight into the role war-trauma plays in Alias Grace. However, the civil war is identified as a potential topic, and a KWIC search on some of Chapter 8’s topics—head, blow, and Jordan—returns the passage: “of a sharp blow to the head, which is the only thing that would account for such idiocy. If Dr. Jordan keeps on with this disorderly course of thought in regards to Grace Marks” (Atwood, Alias, chap.50). This passage is ambiguous, but an online plot summary—like those used by students who have not read an assigned text—describes the significance of Dr. Jordan’s blow to the head: “upon the outbreak of the Civil War, Dr. Jordan enlists as a military surgeon. He sustains a head injury and, according to his mother, loses his memory of his time in Canada. However, despite his claim to not remember treating Grace, Mrs. Jordan admits that her son mistakenly refers to his betrothed as Grace” (“Alias Grace Summary”). While Alias Grace is, however tangentially, thematically related to war and war-related trauma, it is not a good example of the thematic connection described above—the war-trauma-by-proxy theme is not detectable in this text. However, the centrality of war related imagery in Alias Grace is further evidence of a topical focus on war and war related themes at work within Atwood’s corpus.
To avoid writing a paper that amounts to little more than a catalog of examples with an analysis tacked on at the end—a form of analysis I deemed useless in part one—I will attempt to summarize the 529 unique topic models generated for this project succinctly. Of the novels I modelled, both Cat’s Eye and Surfacing contain strong evidence of the war-trauma-by-proxy theme discussed at length in this paper. Cat’s Eye has been explored in detail above, and a few examples found in Surfacing are explored below.
- “Joe never mentions his mother and father, Anna says hers were nothing… childhood; it was in the middle of the war, flecked grey newsreels I never saw cars: like my father, Paul saves everything useful…” (Atwood, Surfacing, chap.2). This quote indicates the war played an important role in the lives of each character’s respective parents.
- “a talisman, my father had left me the guides, the man- animals and the maze would go elsewhere. To continue the war. I didn’t want to join…” (chap.8). Here, the speaker is tasked with continuing a war his father began.
In effect, topic modelling provides objective evidence of a very specific topical continuity across five of Margaret Atwood’s novels, and close reading of her poetry provides further evidence in support of this continuity. As further evidence, topic models were generated for 45 of Margaret Atwood’s short stories: these 45 stories are comprised of every story in The Tent and Moral Disorder and Other Stories—I did not read or curate the sample in advance. Of these 45 stories, two are unambiguous examples of the war-trauma topical continuity. The story “My Last Duchess” discusses “the war of zippers and buttons” and how the speaker “couldn’t expect help from her father… just after the war” (Atwood, “Last”); and “Life Stories” discusses a father and mother who go away, leaving “icebergs and war monuments” to mark their passing (Atwood, “Life”). The general subject of war is found in eight of the 45 stories I analyzed, suggesting the pattern of topical similarity, previously estimated at 18%, continues into Atwood’s shorter works. Put simply, when the results of the topic modeller are compared with the results of a subjective analysis the result is nearly identical: war-trauma-by-proxy is a common theme in nearly one fifth of Atwood’s entire corpus. Ultimately, the answer to my central question of how Atwood works the wars is simple: there is ample evidence, both subjective and objective, that much of Atwood’s work is centred on the effects of war, and a good proportion of her work deals specifically with the intergenerational nature of war-trauma. In conclusion, topic modelling has proven its use in the toolkit of augmented criticism: the synthesis of subjective criticism and objective data analysis enables a critic to explore a subjective observation across a huge range of texts without the need for the critic to examine each text individually.
“Alias Grace Summary.” LitCharts, https://www.litcharts.com/lit/alias-grace/summary. Accessed 16 Apr. 2019.
Atwood, Margaret. Alias Grace. McClelland & Stewart E-Omnibus 2016, 2016.
—. Cat’s Eye. Anchor Books, 1988.
—. “Helen of Troy Does Counter Dancing.” Morning in the Burned House, McClelland & Stewart, 1995, pp. 33–36.
—. “Life Stories.” The Tent, Anchor Books, 2006.
—. “My Last Duchess.” Moral Disorder and Other Stories, Emblem, 2009.
—. “Singing to Genghis Khan.” Interlunar, Toronto UP, 1983, pp. 68–69.
—. Surfacing. McClelland & Stewart, 1999.
—. The Blind Assassin. Kindle, McClelland & Stewart, 2000.
—. “The Loneliness of the Military Historian.” Morning in the Burned House, McClelland & Stewart, 1995, pp. 49–53.
—. The Penelopiad: The Play. Kindle, Faber & Faber Ltd, 2007.
—. The Robber Bride. Kindle, Emblem, 1993.
Jockers, Matthew. Macroanalysis. U of Illinois P, 2013.
Moretti, Franco. “Conjectures on World Literature.” New Left Review, vol. 1, 2000, pp. 54–68.
—. Graphs, Maps, Trees. Verso, 2007.
Ramsay, Stephen. Reading Machines. U of Illinois, 2011.
[i] These terms will be used interchangeably throughout this paper.
[iv] Predictions with a probability of less than 4% were removed.
[v] I am using chapter numbers because the kindle editions of The Robber Bride and The Blind Assassin do not include page numbers.