The Science of Free-Indirect Discourse:An Alternate Cognitive Effect
Recent cognitive research has indicated that free-indirect discourse (FID) can promote empathy in readers. We broaden and refine this research by distinguishing between two formal features of FID: (1) its pivot from the third person into the first, and (2) its pivot back out of the first person into the third. We then suggest that a historical survey of literature provides grounds for hypothesizing that the second formal feature of FID might have a very different cognitive effect from empathy: an acceptance of alterity. We provide some supporting evidence for our hypothesis through an original psychology study and conclude by proposing that our identification of a second cognitive effect of FID reveals how scientific reduction might be used to develop multiple, even divergent, models of rhetorical function.
cognitive narratology, empathy, free indirect discourse, perspectivism, reader response, Theory of Mind, rhetorical narrative theory, scientific experiments on literature [End Page 82]
THE recent surge of interest, both popular and professional, in cognitive research on literature has placed literary scholars in a difficult position. On the one hand, literary studies has long been committed to welcoming new methodologies into its fold. But on the other, cognitive science employs an epistemology of “reduction” that seems not only foreign, but hostile, to much of literary studies (Easterlin 20). Where literary studies values the qualitative, subjective, autonomous, indeterminate, and endlessly variable, scientific reduction privileges the quantitative, objective, peer-reviewed, controlled, and single variable (Starr; Gottschall). Moreover, rather than displaying curiosity (or even tolerance) for the alternative perspectives of literary studies, cognitive scientists have frequently minimized them. At times this minimization has been explicit, as when Steven Pinker, Joseph Carroll, and other proponents of “consilience” have disparaged the methods of literary studies as anachronisms ripe for a scientific “revolution” (Carroll 271). But even when it has been tacit, a reflection of the fact that cognitive scientists have such faith in their own epistemology that they are content to research literature without consulting literary scholars (Bloom), it has been lent a cutting edge by the public reputation of science. Buoyed by large federal grants and promulgated by media enthusiasm for pop-Darwinist explanations for literature (Konnikova), cognitive experiments have already attracted such disproportionate attention that they are threatening to overshadow the rest of literary studies (Kramnick), making them seem less like a source of diversity than a danger to it (Crews).
Because of these concerns, large numbers of literary scholars have chosen to resist cognitive studies of literature, either by openly critiquing them, or more commonly, by simply not acknowledging them (Menand). Neither response, however, has managed to slow their growth (Bruhn and Wehrs), and so in this article, we would like to explore an alternative way for literary scholars to protect the autonomy and open-endedness of literary practice from scientific reduction: designing cognitive experiments of their own. That is, rather than rejecting scientific reduction outright, we will inform it with a literary perspective, one that by no means encompasses the vast diversity of literary studies, but that nevertheless draws (and depends) upon it. To model the potential benefits of this more inclusive method of reduction, we will use it to revise a currently popular cognitive interpretation of literature: the theory that novelistic techniques (and in particular, free-indirect discourse, or FID) can improve empathy or “Theory of Mind.” Versions of this theory have been advanced by several different psychology labs, published in Science, reported through The New York Times and National Public Radio, and presented at the annual conference of The Modern Language Association. Many literary scholars, however, have been skeptical, arguing that scientists have not so much innovated the study of novel-reading as reverted back to an uncritical humanism that threatens cultural difference and personal autonomy. This disagreement over FID thus exposes some of the broader methodological differences between scientists and literary scholars, and yet as we will show, it can also serve as a case-study on how to put the two fields into deeper conversation. By designing an original cognitive experiment (performed at the University of Southern California’s Brain and Creativity Institute) to test a literary theory of novel-reading that suggests that FID can have an entirely different cultural function than promoting empathy, we will demonstrate that this literary theory can be supported in the same [End Page 83] empirical manner as the dominant cognitive one. On the logic of scientific reduction itself, there is therefore more than one possible function for FID, so that rather than imposing a single biological purpose on the novel, our literary-cognitive approach preserves the variability of individual choice, allowing authors and readers to make their own decisions about what to use literature to do.
Over the past decade, the claim that novel-reading improves Theory of Mind has been popularized by a team of University of Toronto psychologists loosely headed by Keith Oatley. Oatley and his team begin with a widely-accepted observation: when we are very young, we assume that everyone thinks and feels as we do, but somewhere around our fourth birthday, most of us come to recognize that other people’s heads have their own distinct mental contents (Goldman). Psychologists call this recognition “Theory of Mind” (or more descriptively, “Theory of Minds,” since it involves an awareness that the world is filled with many different minds), and like most of his scientific colleagues, Oatley believes that Theory of Mind (or ToM) serves as prompt to try to determine what is going on in all those unfamiliar heads. One way for us to do so, he suggests, is by using our imagination to “simulate” other people’s thoughts—a process that psychologists refer to as “mind reading”—and it is here, Oatley speculates, that novels might be useful. Because novels can present the interior thoughts of characters, offering an imaginative window into other minds, they seem a potential tool for practicing (and improving) our real-life ToM skills. To test this hypothesis, Oatley and his team recruited ninety-four participants into a cognitive study (Mar et al. 2006), discovering that fiction readers (as defined by a familiarity with authors such as Danielle Steel and Robert Ludlum) were better than non-fiction readers at predicting people’s thoughts from photos of their faces (although conversely, both kinds of readers showed a diminished capacity to predict people’s thoughts from their actions).
This finding led Oatley to conclude that fiction could in certain contexts improve “empathy or theory-of-mind” (703), and he and his team have since run other studies that suggest that fiction readers are more social (Mar et al. 2009), that romance fiction (such as Danielle Steel) can make us more empathetic (Fong), and that fiction can increase the empathy of low-empathy individuals (although it unexpectedly can also decrease the empathy of high-empathy individuals [Djikic]). Further support for Oatley’s basic hypothesis, moreover, has come from a pair of outside labs. In 2012, Dan Johnson showed that individuals who had just read a short piece of fiction were more likely to help others (though in another twist, they were also more prone to false empathy). Then, in a 2013 study published in Science and reported in The New York Times, David Kidd and Emanuele Castano demonstrated that the readers of National Book Award finalists were better able than readers of Danielle Steel novels to gauge people’s internal moods from their outward appearance, suggesting that “reading literary fiction may hone adults’ ToM” (380).
To dig deeper into this connection between ToM and literary fiction, Oatley has recently turned to Jane Austen, and in particular, to her free-indirect style:
Good writing … needs to offer readers a set of cues to start up and run the simulation-dream of the story world with its characters and incidents … Consider how Jane Austen, towards the end of the first paragraph of [End Page 84] Chapter 3 of Pride and Prejudice, handles her characters’ thoughts, when the prospect is announced of Bingley bringing a large party to the forthcoming dance. “Nothing could be more delightful!” she writes. “To be fond of dancing was a certain step towards falling in love.” These two sentences are in free indirect style, newly invented in Jane Austen’s time. This style is a cue to the reader to imagine himself or herself into the minds of the Bennet girls in their anticipation of the dance … They can [thus] be thoughts in the mind of the reader, who thereby becomes an intimate part of the scene.
As Oatley describes it, FID thus involves a quasi first-person pivot in which we look “into the minds” of literary characters, allowing us to run a “simulation-dream” that improves our “imaginative empathy.”1 Empirical support for this model of FID has come from further psychology studies, which have shown that free-indirect discourse can encourage readers to identify more strongly with characters (Bortolussi and Dixon) and to perceive them as “clear and transparently understandable” (Kotovych et al. 262). FID thus seems able to stimulate the vicarious identification that Oatley’s other research has associated with improved mind reading, suggesting that a free-indirect style may be one of the features of literary fiction that explains the heightened ToM abilities measured by the experiments of Kidd and Castano.
There is an attractive (and intuitive) conciseness to this scientific account of freeindirect discourse: by offering a peek into the first-person experience of others, FID makes us more attuned to minds outside our own. As National Public Radio has recently summed it up: “Want to Read Others’ Thoughts? Try Reading Literary Fiction” (Greenfieldboyce). And yet for all its reductive clarity, this account of novel-reading has seemed too simple to literary scholars. Although Oatley treats fictional identification as a source of a pro-social mind reading (Such Stuff as Dreams 148), a variety of literary critics from Suzanne Keen to Dorothy Hale have noted that such identification can instead desensitize us to differences of perspective, encouraging (as in the experiments of Dan Johnson) a false empathy that marginalizes the views of others. Moreover, in the specific case of Austen, Oatley seems to have it exactly backward, for when the passage that he cites from Pride and Prejudice is read in context, it works not so much to promote mind reading as to warn against it:
Not all that Mrs. Bennet, however, with the assistance of her five daughters, could ask on the subject [of Charles Bingley] was sufficient to draw from her husband any satisfactory description of Mr. Bingley … and they were at last obliged to accept the second-hand intelligence of their neighbour Lady Lucas. Her report was highly favourable. Sir William had been delighted with him. He was quite young, wonderfully handsome, extremely agreeable, and to crown the whole, he meant to be at the next assembly with a large party. Nothing could be more delightful! To be fond of dancing was a certain step towards falling in love; and very lively hopes of Mr. Bingley’s heart were entertained.(1.10) [End Page 85]
In keeping with the novel’s generally wry view of Mrs. Bennet’s flights of romantic fancy, this passage is satiric. So eager are Mrs. Bennet and her daughters to hear wedding bells that they clutch at the “second-hand” information provided by a gossipy neighbor, misconstruing Mr. Bingley’s dance plans as a sign of matrimonial seriousness, and setting themselves up (as events shortly reveal) for heartache. Far from emphasizing the transparency of other minds, this passage thus illustrates how easily we misread the thoughts of others, mapping our own feelings over theirs. And so rather than urging us to identify with the Bennets’ free-indirect swoon—“Nothing could be more delightful!”—the narrative encourages a lightly ironic distance.
Austen’s free-indirect method in this passage is by no means exceptional. Many of her earlier uses of FID—and many of the uses of English authors back to Chaucer and classical authors back to Horace—are firmly satiric. Vocalizing the inner vanities of egocentrics such as Mrs. John Dashwood and the Monk of The General Prologue, they encourage a wry separation from the minds of the self-absorbed (Fletcher and Benveniste 2013). It is not, moreover, just the details of literary history that raise questions about the completeness of Oatley’s model of FID; two scientific bodies of evidence do as well. First, there are the unexpected findings of Oatley and his fellow researchers. Why does literature make high-empathy individuals less so? Why do novels sometimes encourage false sympathy and mis-identification? Why does the FID in popular romances seem to affect readers differently from the FID in National Book Award finalists? And second, there is an entirely contradictory cognitive study never cited directly by Oatley or Kidd and Castano (although they seem aware of it).2 In this study, the psychologists Joan Peskin and Janet Wilde Astington gave children storybooks that were filled with cues about what the protagonists were thinking. Like Oatley, the researchers hypothesized that such cues would teach children to reason better about other minds, but in fact, they discovered the reverse. Rather than learning to mind-read faster, the children were more likely to lapse into false identifications, so that their overall development of ToM skills was delayed.
This counterevidence suggests that Oatley’s account of FID is not completely right, but complicating things further is the fact that it in no way proves that Oatley’s account is wrong. After all, while FID has a long history of functioning as a satiric tool, it has also (as Oatley observes) frequently been treated by novelists as a source of identification (Cohn 112). Likewise, while existing scientific research into freeindirect discourse has raised questions that it cannot answer, much of this research nevertheless implies that FID does (in certain situations) promote empathetic feelings (even if these feelings can be misdirected). Although Oatley’s model may not be wholly satisfactory, it thus still has a good deal to recommend it, steering us away from a zero-sum verdict and into a third possibility: perhaps the model tells only part of the story. Perhaps, that is, it captures one—but only one—of the possible cognitive functions for FID. To see if this were the case, we would need to run a psychology survey that employs Oatley’s protocols to test for another such cognitive function. If the survey yielded a positive result, it would indicate that there are multiple scientificallylegitimate interpretations of FID, suggesting that reduction could be used to support a richer, more open-ended account of what novel-reading can do.
Running this sort of survey is relatively straightforward, but designing it is less so. [End Page 86] For a cognitive study to be considered valid, it must reduce literature to a single variable, a procedure that would automatically seem to destroy the very open-endedness that we are trying to preserve. In fact, however, scientific reduction only has a culling effect on literary variety when it is applied insensitively. If we begin by assuming that the novel is a gigantic transhistorical entity that has evolved to serve one master biological function, then we will only ever run experiments to look for this function, and so reduction will inevitably serve as a tool for squeezing out individual variation. But if we draw on the expertise of literature scholars and instead attend to the cultural and authorial differences between novels, then we can design scientific experiments that acknowledge the endless possibilities of literary form, allowing us to catalog a diverse and ever-increasing set of novelistic techniques that can be adapted to different qualitative effect.
As an example of how this more open-ended approach might work, we can return to the Pride and Prejudice passage cited by Oatley and use it as the starting-point for a scientific reduction that does not attempt to explain how FID functions across time, but confines itself to a literary hypothesis specific to Austen’s particular context. One such hypothesis was articulated a half-century prior to Pride and Prejudice by Adam Smith, who was deeply attracted by the ethical ideal of sympathy (Nussbaum 337–43), but who had also been prompted by the skeptical critiques of David Hume to admit that it was impossible to know with certainty what anyone else was feeling (Griswold). Rather than being confident about our ability to grasp the contents of outside minds, Smith was therefore extremely cautious, and indeed, in his view, a prime cause of anti-social behavior was our tendency to overestimate the extent to which other people feel the same way that we do (Marshall). As he dryly remarked: “we sympathize even with the dead … overlooking what is of real importance in their situation” (The Theory of Moral Sentiments 1.1.13).
Like many literature scholars today, moreover, Smith believed that this problem of false sympathy could be managed in part through literary practice (Hale). He argues in his Lectures on Rhetoric that although the most “agreeable” literature seems to immerse readers in the first-person perspective of the author, its invitational effect is in fact made possible by a third-person detachment that works to “regulate [the author’s] naturall temper, restrain[ing it] within just bounds” (55). Rather than directly presenting the author’s first-person consciousness, agreeable literature thus re-presents the first person from just outside, drawing the reader into the author’s perspective through a slight remove. Smith referred to this literary style as “indirect” writing, and it suggests a mechanism for promoting social behavior that is almost entirely the opposite of the one hypothesized by Oatley. In indirect writing, it is not a shift into the first person that nurtures society, but a shift away from the first person, for the author’s admission of the limits of his own perspective is what allows others to see the value in his personal point-of-view.
Austen shared Smith’s view of the ethical importance of self-restraint (Knox-Shaw 146–8), and in the passage from Pride and Prejudice cited by Oatley, she uses free-indirect discourse as a device for encouraging the same third-person detachment that Smith associated with indirect writing. Plunging us into the Bennets’ romantic enthusiasm only to then quickly lift us into the less sentimental perspective of [End Page 87] the world beyond, she takes advantage of FID’s pivot out of the first-person to physically interrupt a moment of self-absorption. The result is a double discouragement of mind reading, for at the same time as the passage satirizes the Bennets for their error, it uses FID to summon and then disrupt an immersion in their point-of-view.
This free-indirect method for restraining sympathy is a regular feature of Austen’s later novels. In the final chapter of Emma, the narrator remarks: “Emma … had no sooner an opportunity of being one hour alone with Harriet, than she became perfectly satisfied—unaccountable as it was!—that Robert Martin had thoroughly supplanted Mr. Knightley, and was now forming all her views of happiness” (3.355). Here, the free-indirect pivot—“unaccountable as it was!”—communicates Emma’s sudden realization that she cannot see into Harriet’s mind. Although Emma has spent the entire novel believing that she can intuit her friend’s romantic preferences, she at last comes to accept that they are “unaccountable.” Reinforcing this lesson about false identification, moreover, is the gently ironic tone of the free-indirect pivot. Instead of immersing readers in Emma’s first-person epiphany, it encourages a smile at the fact that she has only now discovered the limits of her mind-reading abilities,3 maintaining a light third-person separation from Emma’s perspective that works alongside the narrative content to nudge readers out of repeating her empathy mistake.
This historical use of free-indirect discourse (which for ease of reference, we will call the “self-restraint method”) points us toward a very different scientific reduction from the one made by previous cognitive studies of FID. Where these studies have emphasized the pivot into the first-person (by measuring the effect of first-person free-indirect pivots that stretch out at length without a third-person pivot back),4 the self-restraint method suggests the value of a study that instead stresses the pivot out of the first-person (by measuring the effect of first-person free-indirect pivots that are promptly followed with a third-person pivot back). If the participants in such a study displayed no increase in tolerance for social outliers, or if they displayed an increase in both tolerance and identification, this would suggest a potential problem with the self-restraint method. But if the participants displayed increased tolerance and no identification, this would support the viability of the self-restraint method (and by extension, the practice of novelists such as Austen). Moreover, because this empirical finding would be rooted in a context-specific use of free-indirect discourse, it would not reduce FID to a universal cognitive function. Instead, it would allow for an openended number of other applications, including the one suggested by Oatley.
To run this experiment, we implemented a large-scale behavioral survey involving one-hundred-and-eight participants (for ease of reading, we have located the full Methods and Results sections in the Appendix). Our first major task was to select an outlying social behavior that would provide the story content for our free-indirect pivots. In keeping with the example of Emma, that is, we needed to select a behavior which—like Emma’s snobbish judgments of her friends’ romantic choices—would be recognized as anti-social by our readers. Given the diversity of our subject pool, we opted to avoid topics (like gun control or immigration) that might invite too many conflicting attitudes. Instead, we decided to choose a behavior that would be perceived as anti-social by the overwhelming majority of our participants, regardless of their personal history or political commitments. There are a variety of such behaviors [End Page 88] (for example, incest or boasting) that might have worked for our purposes, but in the end, we settled on perhaps the most ubiquitous literary example of a highly antisocial behavior: the decision to take the law into one’s own hands by seeking revenge. Protagonists who take revenge can be found in everything from ancient Greek epics (The Iliad), to medieval African praise songs (Sundiata Keita), to Renaissance English plays (Hamlet), to avant garde Korean films (Oldboy), and in addition to being widely portrayed in world literature, revenge is also directly anti-social. That is, unlike romantic incest, revenge is a non-consensual act that inflicts physical harm on others, and unlike boasting, it openly threatens the existing civic order. Even among behaviors that are perceived as socially unacceptable, revenge thus has a conspicuously high public cost, and so people are inclined to frown upon it unless they identify with the revenger’s motives.5 By making this behavior the focus of our free-indirect pivots, we therefore attempted to guard against a false positive in two ways, for the anti-social qualities of revenge not only make it difficult to tolerate, but mean that tolerance is typically driven by identification, not restraint.
With this behavior selected, our next task was to generate two sets of literary samples, each as identical to the other as possible, except that one (our FID-enriched samples) contained a much higher concentration of free-indirect third-person pivots than the other (our controls). To determine what constituted a “high concentration” of FID, we once again turned to historical precedent. Since Austen, novelists have shown a willingness to use free-indirect discourse more and more densely: at its peak, Emma averages about eight free-indirect pivots per thousand words, while Mrs. Dalloway opens at two-and-a-half times this rate. We therefore decided to follow the example of modern authors and enrich our samples to roughly twenty free-indirect pivots per thousand words, hoping that this concentration would be high enough to prompt a measurable response, without being so high that it would prove inaccessible to modern readers. Because FID is a complex formal technique that involves many different features of language and grammar (Kotovych), we did not attempt to capture every possible aspect of FID in our samples, but focused on the one feature highlighted by the self-restraint method: the interruption of a first-person, free-indirect pivot by a pivot back into the third. For example, one section of our control sample read:
Amerigo Bonasera remembered the judge’s pledges, his vows, and all his great promises that he would be just. And Amerigo felt betrayed.
While the FID-enriched sample read:
Amerigo Bonasera remembered the judge’s pledges, his vows, and all his great promises. Just. He called himself just. And Amerigo felt betrayed.
While the control sample was strictly third-person, the FID-enriched sample thus began in the third-person (“Amerigo Bonasera remembered …”) and pivoted into the first (“Just …”), before then pivoting quickly back into the third (“And Amerigo felt betrayed”).
As the above samples reveal, in order to introduce free-indirect pivots from third-to-first-to-third person, we had to introduce other grammatical changes as well. [End Page 89] For example, the average sentence length was shortened, while the word “just” was repeated. This, in turn, raised a potential concern: how could we be certain that if we saw an effect in our readers’ behavior, the true cause of this effect was not the shifts in sentence length, or the repetition of words, or some other grammatical factor only tangentially related to FID? To mitigate this possibility, we varied the grammatical form of our free-indirect pivots, so that the total concentration of the third-to-first-to-third-person shifts was greater than the concentration of any tangentially-related formal feature. For example, another section of our control sample read:
Medea had married Jason because people said he was a good man. But he had deceived her. She knew good men didn’t twist the truth. She knew good men didn’t abandon their children. She knew good men didn’t leave their wives to marry someone richer. Medea wished she’d never met the brute.
While the FID-enriched sample read:
Medea had married Jason because people said he was a good man. But Medea now sneered at the idea. Did a good man twist the truth? Did a good man abandon his children? Did a good man leave his wife to marry someone richer? Medea wished she’d never met the brute.
In this case, the free-indirect pivot did not involve increased repetition. Instead, it relied on a different grammatical technique, the rhetorical question: “Did a good man twist the truth?” In other passages, we drew on a variety of further techniques—timeshifting adverbs, for example, or exclamatory address—so that while maintaining the consistency of the third-to-first-to-third person pivots, we tried to generate enough diversity in the other grammatical features associated with FID to disperse any unwanted effects.
Even with this effort to minimize grammatical variables, there was still the possibility that readers might be influenced by a difference in literary quality between our FID-enriched and control samples (a difference that, as Kidd and Castano have shown, can have a measurable effect). For example, had we developed our samples simply by culling FID-rich excerpts from a modern novel and then pruning the freeindirect pivots to make our controls, our FID samples would all have been untouched originals while our controls would all have been chopped-down adaptations. To avoid this unevenness of literary quality, we therefore based our samples not just on post-Austen revenge novels (such as The Count of Monte Cristo or The Godfather) that already contained FID, but on popular revenge dramas and epics (such as The Odyssey or Medea) that predated the novel and had to have FID added after-the-fact. The control samples based on these older works were thus closer to their literary origins, while the FID-enriched versions were further away, balancing out any tendency for our controls to skew toward a lower literary quality. Further consistency of literary quality was then provided by the fact that all of our samples, both FID-enriched and control, were prepared by a single professional author, who relied upon the details of the original stories, but translated them into his own words and standardized them [End Page 90] to the same fifth-grade reading level and word count. In short, instead of presenting one group of our participants with a modern literary masterpiece, and the other with a doctored revision, both groups were presented with something in-between: a range of historical narratives, all crafted by the same author, and all written in a style that the average middle-schooler could comprehend.
Because these samples ran to over ten thousand words, we cannot reproduce them fully, but to provide a more detailed sense of our method, here are representative portions of each of our samples, in both their control and FID-enriched forms:
Sample 1, control. And then Ulysses had hungered to go home. He wanted back in his own bed. He wanted to see his family. He had missed Telemachus grow up. He had missed Penelope smiling. But it was not to be. For ten more years, Ulysses’ journey home was interrupted.
Sample 1, FID-enriched. And then Ulysses had hungered to go home. To be back in his own bed. To see the family he had left behind. Telemachus, now a teenager. And Penelope, her smile golden, her eyes glad. But it was not to be. For ten more years, Ulysses’ journey home was interrupted.
Sample 2, control. Antigone looked at the king. He said, “I have given you justice.” But Antigone scowled. She knew he knew nothing of justice. And she decided to teach him the truth. She knew a truth higher than the ways of man. She knew the gods were waiting with their justice. And she felt their strength.
Sample 2, FID-enriched. Antigone looked at the king. He said, “I have given you justice.” But Antigone scowled. Justice? He knew nothing of justice. And she decided to teach him the truth. Yes, there was a truth higher than the ways of man. The gods. The gods were waiting with their justice. And she felt their strength.
Sample 3, control. Medea smiled, thinking of poison. She liked poison because it was so painful. The thought of the princess in pain made Medea very happy indeed.
Sample 3, FID-enriched. Medea smiled. Poison. Yes, that was the way to do it. Poison was so painful. The thought of the princess in pain made Medea very happy indeed.
Sample 4, control. Dantes became pale at the sight; he felt he had passed beyond the bounds of vengeance, so he could no longer say that God was for and with him. Remorse gripped him at the thought of his revenge. But then Dantes thought of the fourteen years in his cell in Alcatraz. And he saw the inscription he had carved in white letters on the green wall, begging God to preserve his memory. And he realized that God had not forgotten him. And Dantes was happy. [End Page 91]
Sample 4, FID-enriched. Dantes became pale at the sight; he felt he had passed beyond the bounds of vengeance, so he could no longer say, “God is for and with me.” Oh, God. Oh, God. What did God want with revenge? But then Dantes thought of the fourteen years in his cell in Alcatraz. With the inscription he had carved in white letters on the green wall—“O God, preserve my memory! God do not forget me!” And now God had not forgotten. And Dantes was happy.
Sample 5, control. Only his forty years as a professional undertaker kept the overwhelming frustration and hatred from showing on Amerigo Bonasera’s face. His young daughter was still in the hospital with her broken jaw wired together. While the two animals went free. He saw now that it had been a farce. He watched the happy parents cluster around the sons they loved. They were happy. They smiled freely. The black bile, sourly bitter, rose in Bonasera’s throat. It overflowed through tightly clenched teeth.
Sample 5, FID-enriched. Only his forty years as a professional undertaker kept the overwhelming frustration and hatred from showing on Amerigo Bonasera’s face. His young daughter was still in the hospital with her broken jaw wired together. And now these two animals went free? It had all been a farce. He watched the happy parents cluster around their darling sons. Oh, they were all happy now. They were smiling now. The black bile, sourly bitter, rose in Bonasera’s throat. It overflowed through tightly clenched teeth.
As these examples demonstrate, in order to keep the vocabulary and syntax of the controls as close as possible to the FID-enriched samples, we allowed the controls to retain short sections of text—e.g. “nothing of justice” in Sample 2—that fall under certain modern definitions of FID.6 In other words, in keeping with our view that there are many different formal kinds of FID, we did not attempt to isolate every kind of FID at once. Instead we chose to focus on the one specific kind of FID exemplified by Austen’s novels. This kind of FID, as we have seen, involves explicit punctuation breaks (e.g., dashes or exclamation marks or periods) that accompany the third-to-first-to-third person pivots. Such punctuation breaks emphasize the interruption of the first-person that Smith and Austen associate with the development of self-restraint, and they are concentrated at over fifty-times the rate in our FID-enriched samples.
To measure the effect of this concentrated form of FID upon our participants, we then prepared two sets of questions (see Tables 1 and 2 in the Appendix). The first set of questions, the “Sample Response Questionnaire,” was designed to gauge our participants’ identification with the protagonists. For example, we asked our participants whether, on a scale from 1 to 10, they felt they understood the protagonist’s perspective, whether they felt they shared in the protagonist’s emotions, and whether they could see themselves taking similar actions in similar circumstances. If the participants who read the FID samples answered these kinds of questions with a strong affirmative, that would suggest that they had identified with the protagonist, while if they [End Page 92] answered in the negative, that would imply that they had instead kept their distance. Meanwhile, the second set of questions, the “Revenge Attitude Questionnaire,” was designed to measure broader shifts in our participants’ ethical attitudes that went beyond their assessments of the specific actions of the protagonists and into their more general views about revenge-taking. Some of these questions were meant to doublecheck the results from the first set of questions. For example, if the participants who had read the FID samples did not feel that it was acceptable to imitate the protagonists and take the law into their own hands (question 2), that would support the view that FID need not involve identification. Other questions, meanwhile, were intended to test whether, as predicted by the self-restraint method, a free-indirect detachment could be associated with greater tolerance for a character’s behavior. For example, if our FID-readers felt that it was unacceptable to take the law into their own hands, but nevertheless answered in the affirmative to questions such as “Do you think that revenge is ever justified?,” that would establish their ability to tolerate a behavior with which they did not identify themselves.
With these two sets of questions prepared, we ran the experiment, blindly dividing our 108 participants into two groups. One group was asked to read two randomly assigned and ordered FID-enriched samples, while the other was asked to read two randomly assigned and ordered control samples. The mean reading-time for all sets of samples was identical: roughly ten minutes per sample. After reading each sample, the participants were asked to fill out a “Sample Response Questionnaire” about that particular sample, and after finishing both samples and their associated questionnaires, the participants were then asked to answer the “Revenge Attitude Questionnaire.” To prevent communication between the participants, or any other unwanted variables of this kind, a staff member remained in the room while the study was being administered. To ensure that the participants did not change their answers out of fear of being watched, however, the staff member did not actively monitor them, and the participants were all assured that their responses were to be kept strictly anonymous. The participants were not told what the study was about, nor were they instructed on the nature of free-indirect discourse or asked to read in any particular way.
After the experiment was completed, the participants’ responses were statistically analyzed (as discussed in the appendix). To cross-check our preparation of the FIDenriched samples, we looked to see if there was any significant variation among them, and there was not. Whether our participants read the FID-enriched version of Medea, The Count of Monte Cristo, or any of the other three stories, their responses were the same. We then compared the FID-enriched samples as a group to the controls, discovering several statistically-significant differences (p-values < 0.05). To begin with, when we analyzed the results of the “Revenge Attitude Questionnaire,” we saw that the participants who had read the FID-enriched samples were more likely to answer in the affirmative to the questions: “Do you think good people are capable of violent actions?” and “Do you think that revenge is ever justified?” These participants, in other words, believed that the socially-outlying behavior of the protagonists was more acceptable,7 so that FID was associated with higher tolerance.
This result suggested that FID could indeed influence people’s social behavior, [End Page 93] but by itself, it did not shed light on whether this influence was due to self-restraint, identification, or some other cognitive process. To explore the mechanism behind FID’s function, we therefore examined the remaining results from the “Revenge Attitude Questionnaire,” finding no evidence that the readers of the FID-enriched samples had identified with the protagonists. When they were asked whether they were more willing to commit an act of revenge themselves—“Do you think that it is acceptable to take the law into your own hands?”—they answered in the negative, suggesting that the behavior they tolerated in the protagonist was not one they would tolerate in themselves.8 In other words, far from taking on the attitudes of the protagonist, they perceived a division between themselves and the literary consciousness portrayed through FID. This finding was then confirmed by turning to the “Sample Response Questionnaire,” which revealed that FID was no more likely to encourage our participants to claim that they understood the perspective of the revenger, shared the revenger’s emotions, or would themselves act in the same way as the revenger. In fact, quite the reverse: while FID was linked with an increased emotional response, our participants linked this response to the suffering of the victim. In line with the self-restraint method of FID, our results thus associated tolerance not with the first person, but with the third. Like Emma when she accepts Harriet’s taste in men, our participants displayed more acceptance of another person’s behavior even as they failed to identify with it.
A lone cognitive study cannot be used to draw overarching conclusions about how novel-reading must work, but our experiment does more modestly indicate that the self-restraint method of FID is scientifically plausible. As modern literary scholars have argued, and as the history of literature implies, the pro-social effects of novels need not be connected to empathy. Instead, our findings suggest that literature can encourage greater diversity and inclusion by promoting an acceptance of alterity. In addition to associating novel-reading with this specific social behavior, moreover, our study also suggests a few broader ways in which the methods of science and literature might enrich one another. To begin with the possible enrichment of science, our experiment design shows how literary studies can provide scientific researchers with a wider set of hypotheses about literature’s potential functions, a widening that in turn shifts cognitive approaches away from a deterministic, transhistorical model of literary practice toward more autonomous models in which writers and readers can choose what they want to use literature to do. In some cases, we might decide to produce and consume a free-indirect style that emphasizes the first-person pivot, encouraging empathy and identification. In others, we might opt for a free-indirect style that emphasizes the third-person pivot, promoting tolerance through an alternative cognitive route.
From the perspective of science, moreover, this literary sensitivity to a diversity of function has the potential to enrich not only cognitive accounts of literature, but accounts of cognition itself. Although most psychologists now root Theory of Mind in what Oatley calls “mind reading,”9 there are a number of less publicized studies that instead link ToM to what Smith and Austen portray as the source of social behavior: the “inhibition” of our assumption that other people think like we do.10 Susan Birch and Paul Bloom, for example, have traced young children’s failure at ToM tasks to the fact that they are “biased by [their] own knowledge.” Similarly, other researchers have [End Page 94] demonstrated that young children can learn to display ToM behaviors by adopting rules that check their egocentrism (Clément), while adults fail at ToM tasks when they do not recognize their own personal “motives and goals” (Ickes). By suggesting that ToM is improved not by imaginatively closing the gap between our minds and others, but consciously acknowledging it, these studies seem to contradict the mainstream view, and they have often been tacitly marginalized by it.11 Yet as our study of FID reveals, the differences between the mind-reading and the inhibition accounts of ToM need not be mutually exclusive. Instead, they can be taken as complementary, such that ToM is understood to be improved in some cases by closing the gap (as the identification model of FID suggests) and in other cases by widening it (as the self-restraint method suggests). Seen this way, the apparent contradiction between the two accounts of ToM could in fact be evidence of deeper cognitive nuance, explaining why (for example) the young children in Peskin and Astington’s study had delayed ToM skills after their storybooks were re-written to make identification easier. Since these three-year-old children innately assumed that other people thought like them, they could only (in keeping with the self-restraint method) develop ToM by being clued into the gap. But once the social maturation of their brains helped clue them in, the very storybooks that had at first hindered their ToM skills could then (in keeping with the identification model) potentially improve these skills through another cognitive pathway.
Meanwhile, from the perspective of literary studies, our experiment suggests a way to embrace scientific reduction without foreclosing the open-endedness of literary practice. Instead of imposing a single master-logic on the novel, our study uses the methods of science to test an alternative theory of FID, one that does not displace Oatley’s hypothesis, but partners with it to provide an expanded account of novels’ possible biological effects. Although individual experiments may boil literature down to a single function, our study thus shows that these experiments can be supplemented by others that isolate different formal qualities, introduce different environmental factors, and measure for different psychological responses, so that what any given scientific reduction establishes is not a final account of what literature must do, but simply a functional potential that allows for other uses. Nothing in our study, moreover, precludes non-scientific approaches to interpreting the function of FID, nor does our research in any other way present itself as absolute. Since scientific experiments are exercises in attempted “falsification,” the most that ours can be said to do is to strengthen the probability that an alternative theory of FID is not untrue (Ayala). While scientific reduction has in general been seen as antithetical to many of the core commitments of literary studies, our experiment thus suggests that it can work—in its own particular way—to support individual variation and choice. Offering us a different way to explore the psychological function of literature, it serves as another reminder of Emma’s discovery: there’s more than one right way of thinking. [End Page 95]
Angus Fletcher is an Associate Professor of English and Core Faculty of Project Narrative at The Ohio State University. He is the author of articles in Critical Inquiry, New Literary History, and The Journal of Narrative Theory, and of the book Comic Democracies from Johns Hopkins University Press. He can be reached at firstname.lastname@example.org.
John Monterosso is an Associate Professor of Psychology and member of the Brain and Creativity Institute at the University of Southern California. He is the author of articles in The Journal of Neuroscience, Biological Psychiatry, Proceedings of National Academy of Science, and numerous other behavioral science journals. He can be reached at email@example.com.
Participants were recruited through a research subject pool used for psychology courses (primarily for introductory psychology). All participants were read, and then signed, an informed consent that had been approved by the University of Southern California Office for the Protection of Research Subjects. Participants completed the study either individually or with as many as three other participants in the same room. Participants were not permitted to discuss testing material during the study session.
Each participant was randomly assigned to either an FID-enriched or control condition. Participants were not notified of this, nor were they given any orientation to this dimension of literature. Each participant filled out a brief demographic questionnaire, and then was given the first of two literary samples to read. Samples were selected randomly from a group of five literary texts (Homer’s Odyssey, Sophocles’s Antigone, Euripedes’s Medea, Dumas’s Count of Monte Cristo, Puzo’s Godfather) all of which dealt with revenge. Each sample had a FID-enriched and a control version, and the participant received the version that corresponded to his or her randomly selected condition. A researcher recorded the time the participant began reading the literary sample, and the time the participant finished. Participants were not asked to read for any particular details or in any particular way. After reading the first sample, the participant was given a 20-item questionnaire designed to assess their level of engagement in the sample (“Sample Response Questionnaire”). Upon completion of the Sample Response Questionnaire, the participant was given a second literary sample to read, again with time recorded, and again followed by another Sample Response Questionnaire. The Sample Response Questionnaires were identical, except for questions in which character names appeared (see Table 1).
After completion of both literary samples and both corresponding Sample Response Questionnaires, participants completed a 5-item questionnaire designed to assess their attitudes and beliefs about revenge (“Revenge Attitude Questionnaire”). These questions probed the participant’s general attitudes towards revenge, and made no reference to the specific literary samples (see Table 2).
One hundred and eight individuals participated in the experiments. Participants included 29 males and 79 females, ranging in age between 18 and 40 (Mean = 20.3 ± 3.08). Among the participants 36 self-identified as “Caucasian” (33%), 10 as “Black or African American” (9.3%), 8 as “Hispanic/Latino/a” (7.4%), 43 as “Asian” (39.8%), 1 as “Native Hawaiian or Other Pacific Islander” (0.9%), and 10 as “Other” (9.3%). Fifty-five (50.9%) participants were randomized into the FID-enriched condition, and 53 (49.1%) into the control condition. The groups did not differ in age, or in the balance of gender and racial identity (all p-values > .6). [End Page 96]
Mean reading time for literary samples did not differ by condition (10.08 ± 2.58 min for FID-enriched samples vs. 9.85 ± 2.90 min for control samples; t (106) = 0.44, p = 0.66). Prior to analysis of the effects of condition on Sample Response Questionnaires, all response variance related to Sample (but independent of Condition) was removed in separate Univariate Analyses of Variance. The standardized residuals from these analyses were then probed for all subsequent analyses. This is a standard scientific method for reducing noise, because it prevents random blips from being amplified in subsequent levels of analysis.
There was a positive association between time spent reading the literary samples, and higher responses on the Sample Response Questionnaires (r(108) = .36, p < .001). However, there was not a significant difference between male and female participants’ responses (t (106) = 1.04, p = .30). Of greatest interest presently, overall mean response on the Sample Response Questionnaire were significantly higher for the group randomly assigned to FID-enriched condition (t (106) = 2.1, p < .05). For exploratory purposes, we looked at individual items of Sample Response Questionnaire. Based on individual paired t-test comparisons, the individual items that most distinguished the FID-enriched and control conditions were item 3 (“To what degree did you enjoy reading the story?”; t(106) = 2.40, p = .02 uncorrected for multiple comparisons), item 5 (“To what extent do you think you felt the emotions that [target of revenge] felt?”; t(106) = 2.08, p = .04 uncorrected for multiple comparisons), item 8 (“How strong were the emotions that you felt during the story?”; t(106) = 2.08, p = .04 uncorrected for multiple comparisons), and item 20 (“Did the story seem mostly neutral or did it have a personality?”; t(106) = 1.99, p = .049 uncorrected for multiple comparisons). In each case, agreement with the item was higher in the FID-rich condition.
The Revenge Attitude Questionnaire was the primary dependent measure for the study. The summary data for the individual items, separated by condition, are presented in Figure 1. These raw scores allow easy qualitative assessment of the effect FID had on responses in the Revenge Attitude Questionnaire. However, in order to minimize the number of statistical comparisons made, we conducted a principal components factor analysis. In other words, rather than analyzing each question independently, we reduced the data into two core categories, each associated with a different set of questions. The initial reduction was derived using a threshold Eigenvalue cutoff score of 1, and without any rotation (although rotation is commonly used to force more divergence in item loading and easier interpretation of factors, we did not rotate our factor solution since the factor analysis was carried out for purposes of reduction only). We obtained a two-factor reduction, with the first component explaining 40.44% of the total variance, and the second component explaining 23.64% of the total variance. The robustness of the reduction was first assessed quantitatively, both by analysis of the scree plot, and by rerunning the analysis with removal of individual items. Then it was assessed qualitatively, confirming that the questions fell into two [End Page 97] broad categories that lined up with the reduction analysis: third-person presentation (“Do you think that good people …” “Do you think that it …” “Do you think that revenge …” “Do you think that force …”) and first-person presentation (“Do you think that we …”). All questions with the exception of question three (“Do you think that we can always trust the legal system to protect our interests?”) loaded more strongly on the first component than on the second component. The factor loadings were saved and were the basis for subsequent analyses. For exposition purposes, we refer to the factors as “primary” and “secondary” components.
We then carried out a standard statistical method for reducing noise and sharpening results: Separate Analyses of Covariance, or ANCOVA. In particular, ANCOVA models were carried out examining the association between the primary and secondary component and demographic variables, reading time, and most importantly, FID-enriched condition. For the primary component, male gender was associated with higher loadings (F(1, 99) = 8.39, p = .005) as was longer reading duration (F(1,99) = 4.53, p = .036). Most importantly, the FID-rich condition was associated with significantly higher primary component scores (F(1,99) = 4.60, p = .034). Based on Figure 1, it appears that this association was primarily driven by an association between the FID-rich condition and a tendency to agreement with the items, “Do you think that good people are capable of violent actions?” and, “Do you think that revenge is ever justified?” The same ANCOVA model was next used to predict loadings on the secondary component. None of the predictor variables were associated with secondary component scores (all p-values > .28).
[End Page 98]
[End Page 99]
2. Kidd and Castano, in the quotation above, associate fiction with the improvement of ToM in “adult,” as opposed to very young, readers, suggesting that they are perhaps tacitly distinguishing themselves from Peskin and Astington’s research into children. Oatley cites Astington’s research in “Bookworms Versus Nerds” (Mar et al. 2006, 710), but not the joint Peskin-Astington study that followed Astington’s original paper.
3. Unlike Austen’s early uses of FID in Sense and Sensibility (which brusquely satirize characters such as Mrs. John Dashwood and her husband for failing to recognize their own egoism) or in earlier portions of Pride and Prejudice (which, as we have seen, satirize the similar obliviousness of characters like Mrs. Bennet) this passage allows that the character herself may be ruefully sharing in the joke. By doing so, it avoids what Austen seems to have recognized was a potentially counterproductive effect of FID: allowing readers (like the narrator in Sense and Sensibility) to feel that they have achieved a god’s eye view in which they can look down scornfully on characters like Mrs. John Dashwood, falling back into the very egoism that they think they are escaping. To ward off this unwanted effect, many of Austen’s later uses of FID (including here and in multiple of the free-indirect pivots involving Anne in Persuasion) employ a more inclusive focalization that allows for the character to be ironizing herself, encouraging the readers to engage in a similar self-reflection that checks their moral judgment.
4. For example, by using literary samples rich with stream-of-consciousness narration, where “Perhaps we are not in love,” the man reflected, “perhaps I am not in love. Perhaps …” is converted to Perhaps they were not in love. Perhaps he was not in love. Perhaps … (Table 4 of Bortolussi and Dixon).
7. The p-value for this measurement was 0.034, indicating that there was a 3.4% chance that our results were purely random. The usual threshold for a publishable result in Psychology is a less than 5% chance.
8. This result is consistent with recent work on “Imaginative Resistance” (Gendler), which shows that people are uncomfortable imagining moral judgments that differ from their own, further suggesting that tolerance must ultimately come not from empathizing with alternative perspectives, but from restraining knee-jerk dislike of them.
9. E.g., Such Stuff as Dreams, 115. Alvin Goldman has recently summarized the general consensus among psychologists: “how is mindreading accomplished? In broad strokes, there are three competing answers: by theorizing, by rationalizing, or by simulating” (4, emphasis in original).
10. Leslie et al. The authors note that Theory of Mind “kick-starts belief-desire attribution but effective reasoning about belief contents depends on a process of selection by inhibition” (528). For the need for such inhibition, on into adulthood, see Kreuger.
11. The findings of Joan Peskin and Janet Wilde Astington, for example, have accrued citations at less than half the rate (6.7 citations/year) of Oatley’s sixth most-cited article on literature and cognition (Mar et al. 2006; 14.4 citations/year). So it is that in Kidd and Castano’s recent Science article on literature and ToM, they cite Oatley multiple times while not referencing Peskin and Astington. [End Page 100]