Telling Stories: Unreliable Discourse, Fight Club, and the Cinematic Narrator

"The first rule of Fight Club is, you do not talk about Fight Club."

—Fight Club (1999)

Forty-five minutes into David Fincher's film, the viewer enjoys a brief respite from its violence and gore: a surreal sex scene featuring the protagonist's would-be love interest, Marla Singer, and an unidentifiable man. At the scene's conclusion, our protagonist wakes up abruptly, presumably from a delightful dream, and enters the kitchen only to discover that his roommate, Tyler Durden, actually did have sex with Marla the night before. Tyler (Brad Pitt) explains that he rescued Marla (Helena Bohnam Carter) from a suicide attempt, that he brought her home, and that one thing led to another. The film then moves back in time to show the viewer how it happened. To a first-time viewer, this is just another example of Tyler's getting what our unnamed protagonist (Edward Norton) wants—good looks, enlightenment, and now the girl. Someone who has seen the film before, however, understands this flashback differently. Jack, as I shall call him, and his roommate are in fact the same person—Tyler is Jack's alternate personality.¹ And once we know that, we understand that Jack himself rescued and slept with Marla.

But of course the flashback shows us Tyler, not Jack. The discrepancy is not a problem for viewers, who realize that the flashback features Tyler because his personality dominates at the time. It is a problem for critics, though, who would explain how this film manages to show one image while communicating another.² Films such as Fight Club, films in which what we thought was true turns out not to be true, have much to teach us about filmic narration, and the scenes between Tyler and Marla, twice embedded in frame narratives, are an intriguing place to look. Indeed, not only is Fight Club an example par excellence of films that interrogate perspective and interpretation, but these issues inform its story and discourse to such an extent that the film becomes nothing less than an allegory of cinematic storytelling. Understanding this allegory will require that we revisit our definition of "narrator" and perhaps of "narration" itself, insofar as these terms currently account only for textual narrative fiction. There are, certainly, aspects of textual narrators that translate easily to cinematic narrators, but Fight Club raises questions about whether these aspects can explain narrators qua narrators across media. I will argue that while we need a conventional communication model to understand narrative film, a film such as Fight Club suggests that our conception of a narrator must be substantively different for films than for verbal texts.

For one thing, the entities in a film that might communicate are rather harder to pin down than those in a novel. Since the Cahierists introduced the notion of the auteur, critics have been inclined to refer to the director as though he were the sole creative force behind a film. This usage has become, of course, shorthand—one cannot possible list the hundreds of people actually responsible for creating a film. But auteurism does highlight our need to ascribe intentionality to someone or something, be it a director, a producer, or a studio. Because in film we have no single author, we create an entity—usually identified with the director's name—to which we can attribute intention, the source of meaning. In order for us to receive a message, even a message as banal as "it's a wonderful life," there must be a "Frank Capra" out there somewhere, even if he is encased in scare quotes. Furthermore, when we say "Capra," we can't simply mean the man, as he was only one of many responsible for the film, and we can't mean the narrating voice, as we might in a novel, because there isn't one.

So are we left with the implied author? Is that the only agent left who can communicate with a film's viewer? Seymour Chatman thinks not, as long as we agree that a narrator need not be human.³ If we describe films as if they do things, as if they have agency, Chatman suggests, we may as well ascribe that agency to a narrator. Definitions of cinematic narrators are too few to say there is a consensus, but the best is Chatman's own: A film's narrator is the combination of mise en scène, cinematography, editing, and sound. Other accounts are limited either to one of these—usually cinematography—or to acts of literal narration—such as voiceover or the reading aloud of letters.⁴ The virtue of Chatman's broader definition is that it includes all the means by which a film tells a story, much as our definition of a novel's narrator would include all of the means by which he or she tells a story.

Chatman's definition of the cinematic narrator obviously relies upon a conventional communication model, with the real author, the implied author, and the narrator functioning as distinct entities. As the narrator of a film, though, is so clearly different from the narrator of a novel, not least in its having no persona, we might be forgiven for wondering whether this model is equally applicable to both media. And wondering such, we could do worse than consult Wayne Booth in the hopes that his distinctions regarding novels might shed light on films. Booth, of course, argues that the easiest way to establish the characteristics and locations of implied authors and narrators is to examine cases of irony. In an unreliable verbal narrative, specifically, the narrator does not speak "for the norms of the work," which the implied author establishes and which the reader understands; the implied author winks at the reader behind the narrator's back, as it were (158). These narrators, misaligned with their implied authors, misinterpret or misevaluate the events they relate. In order to construct a coherent narrative out of flawed data, then, the reader must be able to differentiate between the narrator's voice and the agent behind it. We might suppose, then, that unreliable films will provide the most useful examples for differentiating between their agents.

In verbal narratives, there are of course various kinds of unreliable narrators, and James Phelan has recently described many of them: "Narrators perform three main roles," he writes, "reporting, interpreting, and evaluating.… They may, therefore, deviate from the implied author's views in one or more of these roles.… Unreliable reporting occurs along the axis of characters, facts, and events; unreliable reading (or interpreting) occurs along the axis of knowledge and perception; and unreliable regarding (or evaluating) occurs along the axis of ethics and evaluation" (50).⁵ In other words, narrators may misrepresent or fail to represent part of the story, misinterpret or fail to interpret part of the story, or misjudge or fail to judge part of the story. Phelan calls these six types of unreliability "misreporting, misreading, … misregarding—and underreporting, underreading, and underregarding" (51). One benefit of having these axes laid out is that we can speak more precisely about the different types of distance a reader finds between narrator and implied author, particularly as these many types of unreliability often work in tandem.

Unsurprisingly, it turns out that not every kind of textual narrator can be unreliable in every way.⁶ Homodiegetic narrators are able to manifest every type of unreliability, and all heterodiegetic narrators are able to underreport, underread, or underregard—to commit sins of omission. But heterodiegetic narrators are able to misread or misregard only if they have a persona. Furthermore, no heterodiegetic narrator is able to misreport because what a heterodiegetic narrator says automatically becomes true. Most instructive is the difference between heterodiegetic and homodiegetic narrators, and the difference between personified heterodiegetic narrators—who can misread and misregard—and nonpersonified heterodiegetic narrators—who cannot. Bruno Zerweck points out that, in order for a reader to explain away inconsistencies in a text, "it must be mediated by a strongly anthropomorphized narrator-character" (155).⁷ That is, when a reader concludes that a narrator is reporting, reading, and regarding badly, the narrator must be an agent capable of doing things badly, a fallible persona. Heterodiegetic narrators with a persona cannot misrepresent the story, but can present a flawed interpretation or evaluation of the story. Heterodiegetic narrators without a persona can do none of these.⁸

If Phelan's categories describe the range of unreliability available to verbal narrators, they should presumably accommodate cinematic narrators, which have far fewer categories into which they can fall. Specifically, they cannot be homodiegetic and cannot be anthropomorphized.⁹ The combination of cinematography, editing, mise en scène, and sound—Chatman's definition of a film's narrator—cannot be made into a persona, and it cannot arise from a character within the diegesis. And if cinematic narrators are extra- heterodiegetic non-personae, they should be unable to misreport, misread, or misregard, as their textual counterparts are.

But they are able to do these things. Films that are unreliable on the level of reporting, those in which the viewer constructs a coherent story despite the cinematic narrator, are the most intriguing departures from literary narrative. They fall into two broad categories: films that mislead the viewer by underreporting the story, and films that lie to the viewer by misreporting the story. Critics often group these films together—calling them puzzle films or twist movies—but there are crucial differences between films that mislead an audience by encouraging it to draw false conclusions, and films that offer false data and thus demand misinterpretations. It is unsurprising that cinematic narrators can underreport diegetic events—every narrator can do that, heterodiegetic or otherwise—but cinematic narrators are also able to misreport, something that would be impossible for the heterodiegetic narrator of a novel. That is, films can show us things that never occurred as if they had occurred; they can manifestly lie to the viewer about the diegetic world. The unreliability arises in the distance between what the heterodiegetic narrator reports and what is actually the case, and this unreliability is unique to film.

Chatman argues that in films that present events that have never occurred as if they had, an unreliable character narrates the unreliable section. Alfred Hitchcock's Stage Fright (1950), the prototypical example, horrified audiences upon its initial release, not because of the murder plot but because of the notorious false flashback. Early in the film, Johnny Cooper (Richard Todd) tells Eve Gill (Jane Wyman) that he is unjustly accused of murder, and as he begins to describe the sequence of events leading up to his flight, the camera moves into what appears to be a flashback, cued by a dissolve. We see Johnny arrive at his lover's home, discover a dead body, and flee the police. When he finds Eve and tells her what has happened, a representation of his version appears onscreen. What a shock, then, when he confesses to the murder at the end of the film! If he has betrayed Eve, the film has betrayed the spectator. By invoking the conventions for a flashback, but presenting instead a dramatization of Johnny's lie, the film misreports diegetic events—the film lies to us. According to Chatman, Johnny is narrating the false flashback—he is "an unreliable homodiegetic narrator.… Clearly, Johnny narrates the first, untrue version of the story," Chatman writes, "not only in his dialogue with Eve in the car but also by means of the ensuing visual sequence" (131, emphasis in original). Robert Burgoyne agrees with Chatman, that Johnny narrates the lying flashback using "images as well as words," and suggests that the cinematic narrator later invalidates this account by presenting us with Johnny's confession (7). Burgoyne argues that a cinematic narrator's presentation cannot be false or the diegetic world would be incoherent. Any false presentation, then, must arise from a character, in this case Johnny.¹⁰

I must agree with Gregory Currie, however, when he writes, pace Chatman and Burgoyne, that characters cannot narrate portions of a film because it would require that they be ontologically superior to the story, as the cinematic narrator is, and as characters clearly are not. As Currie notes of Stage Fright, "Johnny, like the other characters, exists within the story, and it is no part of that story that he produced and edited cinematic images in order to convince his fictional fellows (and us?) of his innocence.… Rather, the deceptive images and their juxtaposition must be thought of as representations of Johnny's account, though we begin by taking them also to be representations of what is real with the fiction itself" (27, emphasis in original). A viewer might understand a flashback as arising from a character's point of view, but would never assume that the character had actually put the clip together. Johnny does not assemble a crew, script and record a false version of the murder, and sit Eve down to watch it. And even if he had, he would be the author of the clip and not its narrator. No character can, at will, leap across diegetic boundaries to replace or even to manipulate the narrator, who alone can present events to the viewer.

In attributing narration to Johnny, Chatman abandons his own notion of the cinematic narrator by making its components available to a character. Moreover, Burgoyne's conclusion relies on circular reasoning: The viewer knows to believe Johnny's confession at the end of the film, that it's authentic, because the cinematic narrator presents it; and the viewer knows that the cinematic narrator presents it because it's authentic, it's trustworthy. Burgoyne assumes, not just that the cinematic narrator presents the diegetic world, but that the cinematic narrator's illocutionary act creates the fictional world.¹¹ "The 'truth of the text,'" he writes, "and the reliability of characters purporting to speak that truth, can be measured only against the authentic facts of the fictional universe, which are a priori constructed by the anonymous or impersonal narrator" (10). But this argument makes it impossible to identify the cinematic narrator's discourse and authentic discourse without reference to each other. We might more reasonably conclude from Burgoyne's analysis of the cinematic narrator's discourse, not that everything the cinematic narrator presents is true, but that everything it presents has a truth value—that it is either true or false. Such an argument, though, would require another entity responsible for creating the diegesis and thus capable of authenticating the narrator's discourse: the implied author.

Currie approaches the problem by embracing the implied author, but then suggesting that we need only the implied author to explain unreliable films, that we need not posit any narrator at all. For Currie, an unreliable narrative results from an implied author's "complex intentions." In such a film, the implied author offers us "clues at two levels, at level one where the clues are more obvious but only superficially persuasive, and at level two where they are less obvious but more weighty when reflected upon" (25). The more obvious clues ultimately suggest a less persuasive story; the less obvious clues, a more persuasive one. Such a narrative, argues Currie, does not require that we establish distance between any two entities—an implied author and a narrator, say—if we only acknowledge that implied authors can be devious, can purposely trick their audiences.

M. Night Shyamalan's The Sixth Sense (1999) is an excellent example of how Currie's argument might work, as in the last few minutes it manages to elicit an interpretation opposite to what most of the film suggests. In the first scene, psychologist Malcolm Crowe (Bruce Willis) is shot in the abdomen, and the screen fades to black. The next scene begins with an inter-title indicating that it is now fall, and we find Crowe watching a young boy (Haley Joel Osment). We are relieved that he has survived the shooting. The film follows Crowe and his frustrations with this boy, who claims to see dead people, and with his failing marriage. So far, the story is entirely coherent. But in the final moments, the viewers and Crowe simultaneously realize that he himself is dead—that he was killed in the shooting at the beginning of the film—and the story changes entirely. It is now about a young boy helping a man to accept his death, a man who watches his widow grieve. Both narratives are coherent, but the latter replaces the former. Because it requires that viewers reject a previous interpretation of events and adopt a contradictory one, it is the strongest manifestation of unreliability a lá Currie, who would argue that the film's implied author intends to fool us by suggesting through conventional editing and cinematography that Crowe has survived.

The Sixth Sense, however, is not parallel to Stage Fright. The first film misleads us—underreports events—while the latter lies to us—misreports events. And films that lie to us, that present events that have never actually occurred as if they have occurred, are deceptive in a fundamentally different way. Currie argues that, when watching an unreliable film, "we find peculiarities, incongruities, and apparently unmotivated elements that start to fall into place when we see that it can be interpreted in another [way]" (23). But in Stage Fright there are no peculiarities, no incongruities, and no unmotivated elements. The viewer's flawed interpretations are not merely the obvious ones but the only possible ones because the data are coherent, but false. A cautious viewer might consider The Sixth Sense ambiguous until the last few scenes, might keep in mind that the shooting at the beginning of the film is never satisfactorily concluded. In films that misreport diegetic events, though, even the most cautious of viewers could not anticipate that parts of the story will be retracted because there is nothing to suggest that they need be. These films are not ambiguous or even misleading, as they must be misunderstood on a first viewing.

It is particularly clear in films like Stage Fright, films that lie to us, that we cannot do without the construct of a cinematic narrator. For one thing, implied authors cannot be responsible for lying to us because they cannot contradict themselves. When the implied author intends something toward the diegetic world, it becomes true. For an implied author to intend X makes X so.¹² The possibility of a subsequent counter-intention, an "It is now not so that X was so," would render the diegetic world incomprehensible. Without a narrator, the scenes above would be the logical equivalent of "This sentence is a lie"—impossible to comprehend both rationally and according to narrative conventions. In other words, the implied author cannot require distance between himself and his audience without destroying coherence. And as we do find these narratives coherent, do detect a real story behind the deception, we must attribute the coherence of that story to an implied author whose intentions are revealed not throughout the course of a multivalent presentation, but despite the presentation.

Moreover, it cannot be the implied author who misreports because implied authors cannot report at all. They are, by definition, behind the report. An implied author allows us to talk about what we believe a text means or does; it is an anthropomorphic construct in which we locate intent. To say that this anthropomorphic construct reports is to make a category error.¹³ If the viewer is deceived, then, we must assume a communication model that has one agent creating, intending, a diegesis and another agent presenting "diegetic events" that it will later renounce. Tradition tells us to call this second agent a narrator. If we accept such a model, we need not make the preposterous claim that the anthropomorphic construct in which we locate intent lies to us in Stage Fright. And we need not claim that Johnny lies to us, when he is unaware not only of us but of the world we inhabit. Instead, we can say that it is the cinematic narrator who lies to us by reporting events that never occurred.

Of course, this narrator cannot renounce without restraint. After all, there must be a narrative logic that supersedes the lie or the film would become ambiguous. Spike Jonze's Adaptation, for example, dispenses with storytelling conventions, and its viewer quickly learns that events are not necessarily related, that characters and places are unstable, and that diegetic levels were made to be traversed. However frustrated a viewer might be by the end, she has no reason to be surprised when the narrative remains incoherent. We would be wrong, though, to say Adaptation is unreliable, as it does not mean other than what it presents—the meaning is only in the presentation. That is, we need not posit a distance between any two entities. But films that do maintain a coherent narrative, films such as Stage Fright, can be understood only in terms of unreliability, and thus require both a narrator and an implied author.

What is the specific mechanism, then, by which these films misreport without destroying coherence? If we could call Johnny Cooper and Malcolm Crowe narrators as Chatman and Burgoyne do—they do speak and can lie—our problem would be solved: they would be unreliable homodiegetic narrators, misreporting and misinterpreting respectively. But they are not narrators. Theirs may be the points of view to which we are privy, but they cannot communicate to us. They may see, but it is the cinematic narrator who speaks, as it were, who presents. As we would never, since Genette, confuse these roles in a novel, perhaps we ought not forget the distinction in film. If we conclude that cinematic narrators are heterodiegetic, though, we are able to think of these characters as focalizers, or in Chatman's terms, filters. In unreliable films, then, the cinematic narrator would be focalizing, filtering the story, through the characters whose versions we're following. Moreover, in films that lie to or mislead the viewer, one character is almost always the explicit focalizer. If we take Johnny Cooper as an example, we would say that he relates his version of events to Eve Gill, while the cinematic narrator focalizes through him, presenting to the viewer what Johnny describes to Eve. The cinematic narrator is unreliable, then, in that it adopts an unreliable focalizer's point of view as if it were accurate, and it is therefore the cinematic narrator whom we must blame for misreporting.¹⁴

This explanation accounts for the critical tendency to group The Sixth Sense and Stage Fright together: both have narrators that focalize through unreliable characters. True, one underreports and the other misreports, but in both the unreliability is on the axis of reporting. Still, the real benefit in calling Malcolm Crowe and Johnny Cooper focalizers is that it allows us to locate the films' unreliability. The narrator in The Sixth Sense underreports events—the event of Crowe's death—and then focalizes through Crowe as he misinterprets his experiences. In presenting the story through his skewed perspective, the film misleads the viewer without actually lying. Unlike non-personified heterodiegetic narrators in novels, cinematic narrators can also misreport events. By focalizing through a character who, intentionally or otherwise, mischaracterizes diegetic reality, the cinematic narrator can present a false version of the story. Johnny, for example, tells Eve that he's innocent, and the cinematic narrator corroborates his version for the audience. The viewer then misinterprets the story because only one interpretation is possible—that Johnny is innocent. We might also note that, while the term "focalize" applies to novels only metaphorically, it explains almost literally what happens in these films. These characters are "focalizers" in that we see what they would have us see. Events are filtered through them as through a lens.

I have been arguing that films have extra- heterodiegetic narrators who, by focalizing through unreliable characters, are able to misreport events—something that narrators in novels cannot do. In order to play this argument out, let us return to Fight Club. In this film we have a protagonist, his alternate personality, and his alternate personality's girlfriend, all of whom understand the world very differently. There are surprisingly few analyses of this film's narrative perversions, but readings of other puzzle films suggest a starting place. We can assume Chatman and Burgoyne would argue that Jack narrates the bulk of the film, as his delusions dominate it for well over two hours (and in fact the protagonist is identified in the credits as "the narrator"). And we can assume Currie would argue that the implied author misleads us by suggesting that Tyler exists, though careful attention to discrepancies would reveal him to be a figment of Jack's imagination. Neither of these explanations, though, is persuasive. The first would require that Jack narrate events of which he is unaware, and the second would require that there be discrepancies to misinterpret, which there are not. Nevertheless, the film is definitively unreliable, so where does the unreliability lie?

A first-time viewer would not suspect the film of any unreliability until the conclusion. Based on the frame at the beginning and the subsequent two hours, a viewer concludes that Jack befriends Tyler Durden, with whom he founds a club for men who like to fight. Soon, Tyler starts seeing a woman Jack knows, Marla of the surreal sex scene, and plots to blow up several buildings, eventually holding Jack at gunpoint to keep him from interfering. There are no contradictions here, no parts of the story left unexplained. In short, there are no markers for unreliability.¹⁵ Once we learn that Jack and Tyler are the same person, however, we must significantly revise the story. In this version, Jack founds Fight Club, starts seeing Marla, and plots to blow up buildings. It is Jack who holds the gun in his own mouth. As in the first version, there are no contradictions, no parts of the story left unexplained. We have merely a different story—one we choose over the first because otherwise the last twenty minutes of the film would make no sense. We do, though, in this second version, need to posit a distance between what the implied author intends—that Jack has another personality—and what the narrator presents for the first two hours—that Jack and Tyler are separate people. We must now recognize the film's unreliability, though we can do so only in retrospect.

On a first viewing, it is impossible to consider Tyler anything but another character, particularly because the film's presentation of other non-diegetic events are clearly marked and entirely compatible with our first interpretation of the story. The film explicitly introduces fantasies or dreams according to cinematic convention, and they look quite different onscreen from the rest of the film. For example, at one of Jack's support groups, the leader invites the participants to imagine entering their "caves." We immediately see Jack in a cave of ice. And there, the leader says, you will find your power animal. Jack sees a penguin. It's clear to the viewer that this scene is not meant to represent diegetic reality. For one thing, it's absurd. For another, it's explicitly marked as a fantasy by the dialogue, cinematography, and mise en scène (not only to the extent that a penguin appears, but due to the jump cuts and diffuse, blue light). Because the film introduces these fantasies according to conventional cinematic language, a viewer understands just how to interpret them. The viewer would then be justified in expecting other non-diegetic events to be marked likewise. In other words, the film gives us every reason to believe that our first interpretation of the story is correct.

Now, it is true that some moments in the film become more meaningful on a second viewing. For example, early on Jack says in voiceover, "I know this because Tyler knows this," and later, "Tyler's words coming out of my mouth." Even more significant is an airport scene in which Jack, again in voiceover, asks, "If you wake up at a different time, in a different place, could you wake up as a different person?" Just as he utters "different person," Tyler moves into frame, and the camera leaves Jack to follow Tyler. Jack, of course, has been waking up as Tyler for quite some time, so the answer to his question is yes. Once we know they are the same person, we understand these lines literally, not metaphorically. But none of these moments confuses a first-time viewer, who would assume that Jack knows what Tyler knows because they talk to each other, that Jack repeats something he has heard Tyler say, and that the near meeting in the airport is coincidence. These moments do not unsettle the viewer's original interpretation.

The same is true of the flashbacks to Tyler and Marla with which we began. They are coherent and conventional on a first viewing no less than on a second. Marla swallows a bottle of pills and calls Jack, who has no interest in saving her or listening to her die, and he sets the phone down. The film cuts to a dreamy sex scene, and then to Jack waking up. He stumbles upon Marla in the kitchen and asks her to leave. Tyler then enters and explains what happened: "So, I come in last night. Phone's off the hook. Guess who's on the other end." We cut to Jack's face, and in voiceover Jack says, "I already knew the story before he told it to me." The film flashes back to the telephone, and we see Tyler picking it up to listen. It then cuts to Tyler walking up the stairs to Marla's apartment, and we hear Jack in voiceover asking, "Now, how could Tyler, of all people, think it was a bad thing that Marla Singer was about to die?" Tyler leads Marla down the stairs and brings her home. The flashback ends with a cut to Tyler in the kitchen, saying something like "Unbelievable." Some conversation ensues, and then we get another flashback—a cut from Tyler's face to a shot of Tyler and Marla in bed. Marla says something to suggest that she's had a good time, and then we cut back to Tyler in his kitchen.¹⁶"How could Tyler not go for that?" the voiceover asks. Tyler and Jack continue talking.

The film is misreporting diegetic events, here, by adhering to cinematic conventions for storytelling while presenting Tyler as though he exists. A viewer is not inclined to doubt his existence any more than Jack's or Marla's. Indeed, the viewer does not doubt it, not only because there is no reason to—there is nothing that needs explaining—but because there is direct evidence that he does exist. We see him walking around, talking to people, and punching them in the face. This adherence is never more evident than in the flashbacks to scenes between Tyler and Marla. These scenes are (with one exception) framed by shots of Tyler, and are formally identical to every other flashback in the film. A first-time viewer cannot but take them as representing diegetic reality—Tyler telling Jack what happened the night before, and a corresponding presentation of events. It is true that, on some level, the film is underreporting events (such as might occur in a novel) and not misreporting them (such as could not). The film is underreporting, in the scene above, that Tyler is Jack's alternate personality. But what the film doesn't report, then, is that it is lying, that events are occurring differently than they appear to be. Our definition of underreporting would then be "failing to report that 'it was not thus,'" and such a definition of underreporting is far too broad to be useful. In fact it describes every kind of unreliability, making distinctions between different axes impossible. (Moreover, if this were an example of underreporting, it would still be of a kind unavailable to literary narrators.) If we want to distinguish between different manifestations of unreliability, then, we must find this an example of misreporting.

In order to identify the unreliable narrator responsible for misreporting these scenes, we must situate them within the rest of the film, within the film's narrative frames. Such an effort reveals a number of distinct and embedded narrating situations, even if we exclude several minor fantasies and dreams. (See figure one.) The complexity in these embeddings increases as the film progresses, almost to the point of absurdity. But the measure of the absurdity will be the degree to which existing narrative models are inadequate. To that end: we begin with a frame narrative, which we identify as the diegetic present, in which Tyler holds a gun in Jack's mouth (A). The film then moves into its primary flashback (B) to explain how Jack has found himself in this position. Inside of this flashback is another flashback (C), which eventually catches up to the first. The primary flashback (D) continues, interrupted by the flashback depicting Tyler rescuing Marla (E), the brief return to Tyler and Jack (F), the second flashback to Tyler and Marla (G), and then another return (H). Some time later, the film flashes back to a shooting (I), an event that profoundly upsets Jack (J). Eventually, there are several brief flashbacks in which Jack remembers events properly, remembers doing what Tyler had seemed to do (K). We then return to the primary flashback (L), which lasts until the film catches up to the frame narrative (M). The film continues, then, for three minutes after Tyler's personality has disappeared (N). Identifying the point of view we're sharing in situation (E), which is twice embedded in

Figure 1.

frame narratives, is a particularly vexed endeavor. Among other things, the answer will depend on what kind of narrators we are prepared to consider.

If we allow the possibility of homodiegetic character narration in films, as Chatman and Burgoyne do, we must establish criteria for identifying it. For simplicity's sake, let us say that a character narrates a scene in which he seems to generate a flashback or an imaginary event, such as a fantasy, fabrication, or delusion. These rather limited criteria would exclude the broadest, less persuasive descriptions of homodiegetic narration that include any scene with voiceover or any scene limited to a particular character's knowledge. We must likewise consider one other difficulty—that Jack could be narrating in retrospect, from the position he occupies in the frame narrative, or narrating as events occur, from the position he occupies within the primary flashback. At these different moments in time, Jack has very different information and might tell very different stories. To clarify matters (I hope), I shall whenever necessary respect chronology and call Jack as he exists for the bulk of the film, inside of the primary flashback, Jack₁. As he exists in the frame narrative, I shall call him Jack₂.

Despite these fairly limited criteria for identifying homodiegetic narrators, the film is an impossible one. (See figure two.¹⁷) A viewer willing to find such a narrator must be prepared for confusion on every narrative level. Even the frame is problematic because a first-time viewer misunderstands it. Tyler appears to be holding a gun in Jack's mouth. Without knowing that Tyler and Jack are the same person, a viewer would assume that (A) has a reliable heterodiegetic narrator. Upon seeing the film a second time, however, a viewer would understand that Jack₂ is narrating (A). Tyler is, after all, his hallucination. (A) and (M), then, could be attributed to Jack₂, even if part of the attribution must be retrospective. Still, once Tyler is banished (N), the delusion disappears, and there is thus no reason to call Jack₂ a narrator. At this point—also the point at which the voiceover disappears—a heterodiegetic narrator would take over, narrating the last few minutes of the film. True, there is not even a cut in the film to mark this transition. So the move is odd, but not impossible.

A viewer would likewise be inclined to attribute the primary flashback, as it begins in (B), to Jack₂, who would be showing us how he ends up with a gun in his mouth. "And suddenly, I realize that all of this," Jack says just before the film jumps back in time, "has got something to do with a girl named Marla Singer." If voiceover, while not sufficient for establishing homodiegetic narration, might be allowed to support it, it is Jack₂'s retrospective commentary that accompanies segment (B). Moreover, when the film catches up to the frame more than two hours later, Jack₂ refers to everything that has come before as a flashback. (In the opening scene, Tyler asks Jack whether he has any last words, and Jack responds with, "I can't think of anything." In the concluding scene, which overlaps the first, Jack responds with, "I still can't think of anything." "Ah," Tyler says, "flashback humor.")

Figure 2.

The primary flashback (B) includes embedded flashbacks of its own, the first of which (C) seems also to arise from Jack₂. "No, wait," he says in voiceover before the film moves to (C), "Back up. Let me start earlier." The film then backs up another six months.¹⁸ When we catch up to the primary flashback (D), though, it's unclear whether Jack₁ or Jack₂ takes over. True, the return to the primary flashback is marked by the voiceover's retrospective comment—"And she ruined everything"—a sentence that could only come from Jack₂, as Jack₁ is seeing Marla for the first time. A minute later, though, Jack says again in voiceover, "Next time [I see her,] I'm gonna grab that little bitch Marla Singer." This sentence, in present tense, must arise from Jack₁, who expects to see Marla again soon.¹⁹ Eventually, and in retrospect, we would be forced to conclude that Jack₁, Jack₂, and a heterodiegetic narrator are alternately responsible for narrating segment (D), despite the fact that there are no consistent markers for identifying shifts among narrators. And it turns out that we must conclude the same for the entire primary flashback (D, F, H, J, and L), as each segment includes moments that must be attributed to Jack in the present, Jack in the past, and an objective heterodiegetic narrator.

This explanation of how homodiegetic narration might work in the film is convoluted but not untenable. Attributing narration in the scenes with Tyler and Marla (E, G), however, proves impossible for any viewer committed to homodiegetic narration. At first, Tyler seems to generate these flashbacks, according to both filmic convention and the norms of this film. So the obvious conclusion is that Tyler narrates these scenes, which is just what a first-time viewer would assume. But can an alternate personality narrate? Let us say, for the sake of argument, that one can. After all, Tyler talks to many people during the film, and there may be no reason to claim that he can't speak here. In this context, he narrates reliably, describing events as they have actually occurred in the world. He rescues Marla, and he lies in bed talking to her afterward. While according to the diegetic reality it is Jack who does these things, it makes no sense to say that Tyler exists to the extent that he can narrate but not to the extent that he can act. We thus find ourselves, in attributing the narration to Tyler, in the difficult position of claiming that the presentation is reliable. However, these flashbacks are decidedly unreliable, and thus Tyler cannot be narrating them.

The obvious alternative to Tyler is Jack₁. Attributing the narration of (E) and (G) to him seems reasonable, as these flashbacks actually occur when Jack is in his kitchen talking to himself. We might even call these, not flashbacks, but representations of what Jack imagines. Jack₁ would thus be visualizing the story that Tyler tells him and reliably narrating this fantasy for the viewer. But that explanation does not accord with the formal conventions of the film. Every other scene introduced as this one is, is a true flashback, and if this were Jack's imagination, we would expect it to look like the film's other representations of his imagination, which it clearly does not. Moreover, the details from these scenes—Marla's apartment, Marla's dress—correspond perfectly to what the heterodiegetic narrator has presented moments before when Jack and Marla talk on the phone. This correspondence is inexplicable if Jack merely imagines these scenes. Another alternative is to argue that Jack₁ narrates these scenes based on his own memory of them, and of course Jack must remember bringing Marla home, else he would be unable to tell himself about it the next morning. Furthermore, if he and Tyler are the same person, we could even claim that the narrator here is a conflation of both personalities. Perhaps Jack₁ narrates what he remembers doing as Tyler. But if that were true, Jack₁ must have at least some access to memories acquired as Tyler, and the rest of the film has Jack totally unaware of Tyler's activities, unaware to the extent that it takes him a significant amount of detective work to uncover them. The closest we get to a parallel situation is near the end when Jack realizes that he and Tyler are the same person and is able to recall himself doing what Tyler had seemed to do (K). He does, suddenly, have access to "Tyler's" memories once he understands what Tyler is. Even so, when the events in these memories took place both personalities were present—Jack believed he was watching Tyler act and he now remembers himself acting in Tyler's role. That he could in this one early instance have access to Tyler's memories, while he does not in any other instance, upsets the otherwise stable relationship between the two personalities and thus undermines the film's coherence, a coherence that no viewer would otherwise doubt. If we call Jack₁ the narrator of (E) and (G), then, we have him narrating events of which he is unaware.

It makes more sense to say that these flashbacks are meant to be, not representations of memories or fantasies, but representations of what has really happened—or misrepresentations in this case. From whom, then, do these misrepresentations arise? It can be only Jack₂, who knows the whole story and is thus in a position to misrepresent past events. But concluding that Jack₂ narrates, or even that Jack₁ and Jack₂ narrate together, requires giving up our criteria for homodiegetic narration—that the character generating flashbacks or fantasies narrates. Jack₂ is nowhere in these scenes. The only place to look is in the voiceover, which could reflect the hindsight that allows us to say Jack₂ narrates (C). During these flashbacks (E, G), however, the voiceover belongs to Jack₁: "Now, how could Tyler, of all people, think it was a bad thing that Marla Singer was about to die?" and "How could Tyler not go for that?" These are the thoughts of Jack₁, not Jack₂, and there is no alternate space for Jack₂ to inhabit. To attribute the narration of these flashbacks to Jack₂ or simply to Jack, we would have to argue that a homodiegetic narrator is not he who generates a flashback or a fantasy, but he who knows enough to lie to us. Such a broad criterion for identifying a narrator is clearly useless. Not only does it fail to distinguish between Jack₂ and a heterodiegetic narrator, but it allows us to claim that, say, Marla narrates these flashbacks. Thus, if we admit the possibility of homodiegetic narrators for (E) and (G), we find ourselves in the impossible position of choosing among Tyler, Jack₁, and Jack₂—none of whom can actually be responsible for what we see onscreen.

By now it should be clear that such an approach to this film's narrative turns out to be not only absurdly complex but untenable. Our difficulties disappear, however, if we merely give up the possibility of homodiegetic character narration. Each narrative situation falls into place if the flashbacks have a heterodiegetic narrator who can represent or misrepresent at will. This narrator could depict Tyler rescuing Marla in order to keep up the illusion that he and Jack are separate people, as the film still suggests at this point. To account for this depiction, we can think of these scenes as focalized through Tyler; they correspond to his point of view. It is, after all, Tyler's personality that is operative, here. If we preferred to be more specific, though, we might say that the narrator focalizes through that-aspect-of-Jack's-personality-that-is-Tyler. We need not differentiate between Jack and Tyler, here, as there is no need to claim that one knows and another speaks. And one advantage of identifying a focalizer in this scene is that we can locate the unreliability. The film misreports events in that what we see onscreen is different from what any character would have seen: Jack, not Tyler, rescuing Marla. The film is able to deceive us because the focalizer—Tyler, or Jack-as-Tyler—unreliably characterizes the diegesis, and the cinematic narrator corroborates the characterization.

If these flashbacks have a heterodiegetic narrator, there is no reason to claim that the other scenes do not. As in (E) and (G), our narrator may focalize through different characters at different points in the film, or it may not. Fight Club would, then, have one cinematic narrator that sometimes focalizes through Jack₂, Jack₁, or Tyler. (See figure three.) Not only does this view relieve us of insuperable difficulties in attributing narration to a particular character in (E) and (G), but it falls within the constraints of intuition and reason. Instead of requiring that a character traverse diegetic boundaries to control what appears onscreen, we can say that a heterodiegetic, cinematic narrator presents the story, sometimes as a particular character sees it, sometimes not. The film would begin, then, with a heterodiegetic narrator that focalizes through Jack₂ (A). The narrator would then present the bulk of the film, either reliably (as in B, C, I, K, L, and N) or unreliably when focalizing through unreliable characters (as in D, E, F, G, H, J, and M). The frame's conclusion would be focalized through Jack₂ (M) until Tyler disappears, and then it would continue for the last few minutes as reliable (N).

Figure 3.

We end up, then, with a heterodiegetic narrator who misreports events while focalizing through different characters. Indeed, I would suggest that there is no other way to account for the film's narration. And once we understand the film as having a single, heterodiegetic narrator, we can consider some larger claims Fight Club makes about storytelling and narration. This is, after all, a film that foregrounds a complex self-reflexivity from beginning to end. Jack's reference to the extended flashback is more than clever. It confirms for us that Jack's story is also an allegory of the film's narration.

There are multiple moments in the film that establish the relationship between the narrative we construct and the narrative Jack constructs. Most significantly, Tyler has a night job as a theatre's projectionist. When one reel of film ends and the next begins—the changeover—Tyler turns one projector off and another on. In between these reels, however, he splices in one frame from a pornographic film. Jack tells us that no one in the audience notices the image, or even that the changeover has taken place. But the image affects the audience nevertheless—children start to cry, lovers start to cuddle—because Tyler subtly disrupts the narrative. Tellingly, several single-frame images of Tyler appear in Fight Club; he stands near Jack in four different shots very early in the film. These images, lasting only one twenty-fourth of a second, are formally identical to the images Tyler splices into films and virtually undetectable. Tyler (Jack?) and our cinematic narrator thus play parallel roles. When Jack realizes that he and Tyler are the same person, this similarity becomes supremely important: "It's called a changeover," Jack says in voiceover, "The movie goes on, and nobody in the audience has any idea." Jack thinks of the shifts between him and Tyler as akin to a changeover—a shift in identity that no one notices, but that is nevertheless enough to disrupt and propel the narrative.

This disruption is not a move toward any true version, however. Both before and after the changeover, the cinematic narrator and Jack construct their own narratives, and these narratives are always based on a fluidity of perspective and identity. The narrator focalizes through any number of characters, and Jack's perspective is occasionally in the present, occasionally retrospective, and occasionally Tyler's. But whatever the data—whatever the points of view—Jack and our narrator shape coherent narratives out of them. Conversely, Tyler is unable to construct a narrative, as he exists only as an aspect of Jack's point of view. That is, Jack sometimes sees through Tyler just as our cinematic narrator sometimes sees through various focalizers. And both Jack and our narrator build their stories upon a variety of perspectives.

It is the multiple, shifting points of view that allow coherent meaning to emerge. Indeed, without a variety of perspectives, no narrative is possible. The film suggests as much throughout, but it is most evident toward the end, when the film presents us with what an objective account would look like. When Jack finally understands his delusion and attempts to shut down Project Mayhem, he races to a building's garage to disable a bomb. Tyler attempts to stop him, and even though we know now that Tyler is a figment of Jack's imagination, we see them fighting each other—except for a few crucial moments when we don't see them fighting each other. The film cuts to the tape a security camera is recording, and we see Jack fighting himself: punching himself, dragging himself by the neck, throwing himself downstairs. Anyone watching this tape without a narrative context would be baffled. A man beating himself up in a garage—what is one to make of such a thing? The security camera, devoid of a cinematic narrator and its focalizers, is here rendered unable to communicate anything meaningful. This camera is entirely objective, and its recorded data are meaningless. Particularly in a film that depends upon inverting spectatorial expectations, the security camera can signify only spectatorial insecurity—the absence of narrative. The meaning, we might say, of both Jack's life and Fight Club is not only communicated through subjective points of view, but in the necessity of subjective and mediated points of view for constructing any meaningful narrative at all.

While Fight Club may be a particulalry rewarding example of this argument, many films lie to us to some extent—present a dream or fantasy only to reveal a moment later that it was a dream or fantasy. We need only think of any romantic comedy in which the heroine imagines her happily ever after. Some films mark these presentations, often with wavy dissolves, but some don't. In any case, we understand exactly how to interpret these scenes once the lie has been revealed. The films I've been discussing, then, are unusual in the scope of their deceptions, but not in their basic technique. We cannot say that Fight Club is unique, or even that its methods are peculiar to certain kinds of films. In fact, a cinematic narrator who misreports—albeit usually for just a moment or two—is too common to note. And if unreliability in films works fundamentally differently than it does in novels—and I hope this analysis has demonstrated that is does—it is because cinematic narrators are fundamentally different from the narrators of novels.

That is, a film has techniques available to it, specifically the ability of its heterodiegetic narrator to misreport, that a verbal narrative does not. This distance between the cinematic narrator and the story it tells is possible, I suspect, because cinematic narrators and their characters use different languages. Films may quote characters, literally or through voiceover, but the cinematic narrator presents its story through moving images—a character's actions, facial expressions, or point of view—something a literary narrator cannot replicate. This difference makes it possible for films to establish a discrepancy between what they present and what they communicate.

These observations are not entirely new. George Wilson wrote, in 1976, that there is a correspondence between the narration of a film and of a novel, but that it must "be handled with care. The kinds of, e.g., unreliability and their aesthetic aims in film narration are quite different from the ways a literary narrator may be unreliable.… The differences are as important as the similarities" (1041–42). Perhaps narrative theory has finally caught up with the full implications of Wilson's observation, as we extend our analyses to narrative across media. And the consequence is that the text or message around which a communication model revolves can no longer be defined by literary narrative, as it has so often been in the past. Narrators are different than we thought they were, stranger too, and our theories of them perhaps too limited. Broadening our inquiry, we must hope, will encourage us to redefine not only these terms but the terms of narrative theory itself.