Narrative unreliability and metarepresentation in Ian McEwan’s Atonement; or, why Robbie might be guilty and why nobody seems to notice

ABSTRACT Unreliability and uncertainty are at the centre of Ian McEwan’s Atonement, and discussions of the novel often emphasise this aspect of the text, arguing that it demands to be reread in light of its surprise ending. Yet such readings do not entertain a more radical doubt as to the reliability of the novel’s protagonist, Briony Tallis, and the accuracy of her adult attempt to atone for a childhood mistake. This article asks why this is the case and offers an alternative reading. It suggests that the force of the concluding section, as well as the way in which the preceding sections build to suggest that Briony’s revised version of events is correct, have effectively shut down a more far-reaching discussion of reliability and the ways in which meaning is revealed and withheld. Drawing on recent developments in narrative theory, it does not argue for an alternative orthodoxy about what ‘really’ happened and what Briony did or did not see, but rather suggests that the consensus that has accreted around her revised version of events raises important questions about the nature of narrative (un)reliability, the role of the implied author, and the ways in which narratives are read, and reread.

When twin cousins, who have been staying with the family, run away that evening, a search party is organised. The adolescent girl searches the grounds of the estate alone and interrupts a sexual assault on her older cousin, the sister of the twins. She does not get a clear view of the attacker, but is immediately convinced it is the family friend, the 'maniac' she saw 'attacking' her sister. The girl tells the police that she can identify him beyond doubt and her testimony is vital to his conviction.
As she grows older and reassesses the events she witnessed, the girl realises she has made a terrible, criminal mistake. It is too late for legal redress, so instead she writes a novel, setting the story straight and identifying the true attacker, an industrialist friend of the girls' older brother who was staying with the family that weekend.
(B) Over the course of a day at a family gathering on a country estate, the son of the family's cleaner, who has become a family friend and whose education has been paid for by the patriarch of the family, comes to realise his attraction to the elder of two sisters. He writes her a sexually explicit note and has the younger sister deliver it. The note forces the elder sister to recognise that she shares his desire, and the couple have sex in the library of the house. The sex is interrupted by the younger sister, who interprets it as an act of violence.
When twin cousins, who have been staying with the family, run away that evening, a search party is organised. The family friend, sexually frustrated and fantasising about a further tryst outdoors, searches the grounds alone. When he encounters the older sister of the twins he sexually assaults her. He is again interrupted by the younger sister, who does not get a clear view of the attacker but is convinced it is him. She tells the police that she can identify him beyond doubt and her testimony is vital to his conviction.
Over the years that follow, the elder sister maintains that her lover is innocent of the crime, and in time the younger sister revises her opinion and comes to believe that she has been mistaken. She writes a novel, giving her revised view of the events and with equal certainty identifies the true attacker as a friend of her older brother, an industrialist who was staying with the family for the weekend.
(C) Over the course of the day at a family gathering on a country estate, the teenaged son of the family's gardener is observed leering at the young women of the party. When twin cousins, who have been staying with the family, run away that evening, a search party is organised. The older sister of the twins searches alone and is sexually assaulted by the gardener's son. The daughter of the family interrupts the assault and misidentifies the attacker as the son of the family's cleaner, who has become a family friend and whom she had earlier interrupted having consensual sex with her sister in the library.
The gardener provides his son with a false alibi, and the family friendthe cleaner's sonis convicted of the assault. The daughter's testimony is vital to his conviction, but in time she comes to believe that she was mistaken. She writes a novel, giving a revised view of the events and with equal certainty identifies the true attacker as a friend of the girls' older brother, an industrialist who was staying with the family for the weekend. Synopsis (A) will likely be familiar to anyone who has read Ian McEwan's Atonement (2001) or seen the 2007 film version directed by Joe Wright. Synopses (B) and (C) will likely strike an off-key note. It may be clear that they take place in the same storyworld as Synopsis (A) and that they centre on the same crime(s), but they do not sit comfortably with the way in which the novel has been read and interpreted. They may seem jarring and do not tally with the orthodoxy that McEwan's novel is a story of youthful naivety and misadventure followed by a subsequent attempt at, well, atonement. This article asks why this is the case given the revelation in the closing section of the novel that Briony Tallis (the 'adolescent girl') is the 'author' of the preceding narrative, and furthermore that she reveals herself as so clearly unreliable and uncertain as to what she did and did not witness on that evening in 1935. It suggests that the force of the concluding section, in which Briony is revealed as the author-narrator of the novel, as well as the way in which the preceding sections build to suggest that Briony's revised version of events is the correct one, have effectively shut down a more nuanced discussion of reliability and the ways in which meaning is revealed and withheld. It does not argue for an alternative orthodoxy about what 'really' happened on that summer day and what Briony did or did not see, but rather suggests that the strength of the orthodoxy that has accreted around Synopsis (A) raises some important questions about the nature of narrative (un)reliability, the role of the implied author, and the ways in which narratives are read, and reread. In this respect, recent work in narrative theory, principally the cognitivist strand of what is often referred to as 'postclassical narratology, 1 suggests a number of fruitful approaches to exploring the question of Briony's testimony and the credibility ascribed to her mature version of events. This is particularly true of ongoing debates concerning the continued relevance, or not, of Wayne Booth's categories of the implied author and unreliable narrator, as well as work on reading, rereading, and the metarepresentational capacities of readers. This essay begins by exploring what Briony did and did not see, suggesting some elisions in her account, before moving on to examine the category of the implied author (particularly in relation to characterisation), and the extent to which Briony can and should be read as an unreliable narrator. It concludes with a discussion of 'source-monitoring' and the way in which the structure of Atonement, with its surprise reveal at the end, disrupts the reader's ability to read a more fundamental and radical unreliability back into the text. 2

What Briony saw
The synopses above focus mainly on the opening section of Atonement, in which the catalysing events of the novel take place, and in which, as it is later presented, Briony makes her fateful mistake. As Lynn Wells has argued, this self-consciously modernism-inspired 3 section, made up of conflicting points of view, should not be read in isolation from the shorter sections which follow, 4 but it is the lynchpin for what Briony later calls her 'fifty-nine-year assignment' to make sense of and atone for her 'crime'. 5 It is also the section which is most overtly and self-consciously writtena product, it is suggested in the frame narrative, of multiple revisions over many years. In the final version of the scene in which Briony interrupts the sexual assault on her cousin, there is uncertainty from the outset as to what and whom she has seen: As early as the week that followed, the glazed surface of conviction was not without its blemishes and hairline cracks. Whenever she was conscious of them, which was not often, she was driven back, with a little swooping sensation in her stomach, to the understanding that what she knew was not literally, or not only, based on the visible. It was not simply her eyes that told her the truth. It was too dark for that. Even Lola's face at eighteen inches was an empty oval, and this figure was many feet away, and turned from her as it moved back around the clearing. But nor was this figure invisible, and its size and manner of moving were familiar to her. Her eyes confirmed the sum of all she knew and had recently experienced. The truth was in the symmetry, which was to say, it was founded in common sense. 6 The 'glazed surface' evoked in the opening line is both metaphor for Briony's testimony and a reference to the Meissen vase broken during her sister Cecelia's earlier tussle with Robbiethe cleaning lady's sonby the fountain in front of the house. Briony witnesses the scene from an upstairs window and her partial understanding of the dynamic between the pair is fundamental to her later interpretation of events, particularly when Cecelia's decision to strip off and retrieve a fragment of the vase from the fountain is read as a response to 'a command which Cecelia dare not disobey'. 7 Cecelia later glues the vase together, but its superficial integrity belies a structural weakness, and years later it shatters when it is moved to the cellar for safekeeping on the outbreak of war. 8 Like Briony's story, the vase's integrity is only superficial. Yet despite this and other literal and metaphorical caveats about the fragility of Briony's story, critical responses as to what she did or did not see have been limited to a relatively few core points. One readingthe dominant readingemphasises that Briony could not see her cousin Lola's attacker, and in that sense lied about being able to identify him with any certainty; it places the emphasis on the fact that she was relying on 'the sum of all she knew and had recently experienced' and that her understanding of that knowledge and experience was fundamentally flawed. The adolescent Briony viewed Robbie as a predator and a threat to her sister, and she moulded her perceptions to fit this view, only later realising her error and attempting to atone by giving a more truthful account. However, an alternative reading would place the emphasis on the fact that Briony notes the familiarity of the attacker's 'size and manner of moving', and that if she is familiar with the physicality of this silhouette then it is unlikely to be Paul Marshall, whom she had met for the first time that day and who was hardly familiar. Briony also remains constant in her belief that the person was tall, which discounts Danny Hardman, the gardener's teenage son and suspect in Synopsis (C): though sixteen at the time, Hardman is frequently described as a 'boy'. This leaves Robbie, whose laterinventeddialogue with Briony reveals him to be the same height as Marshall. 9 In this and similar descriptions of her memories of the attack, what Briony saw, or did not see, neither proves nor disproves her youthful certainty that it was Robbie or her adult certainty that it was Paul Marshall. 10 Similarly, the fact of Marshall and Lola's marriage, which Briony presents to Cecelia and Robbie as conclusive proof that Marshall is the guilty party, could be interpreted as anything but: is it not possible that rather than marrying her abuser, Lola married a man who was there that weekend and could, to some degree at least, understand her trauma without explanation? It is in fact possible to find a counterargument for every piece of evidence used in support of either Robbie's or Marshall's guilt, and to cast the other as the guilty party. The intention here is not to enumerate each piece of evidence and the counterargument, but rather to open up the possibility of such discordant readings within a novel that centralises the ethical power of storytelling and from the outset emphasises Briony's unreliability and inventiveness. Each interpretation stands or falls not on the strength of the evidence but on the strength of Briony's convictions, and yet the tendency is to believe her story second time around after it has been reworked following decades of reflection. That Briony is taken at her word even after the metanarrative reveal 11 at the novel's conclusion is a function not simply of the greater credulity ascribed to the mature, worldly Briony, but also of the ways in which narratives and narrators are read and interpreted.
The death of the implied author? 12 The term 'implied author' has been contested since Wayne Booth introduced it in The Rhetoric of Fiction (1961) to distinguish between the biographical author and his or her implied image in the text. 13 Ansgar Nünning, for example, takes a 'total text structure' approach and argues that the concept of the implied author introduces a redundant and unsustainable distinction between a disembodied author figure and other narrative features such as the narrator. 14 Mieke Bal, Alan Palmer, Richard Walsh, and David Herman have also found it a redundant or unsustainable term due to the difficulty of maintaining this distinction, and in the case of Herman, due to the concept's inherent rejection of intentionalism. 15 Others argue that it continues to provide a useful and necessary distinction, 16 perhaps most prominently James Phelan, who redefines the implied author as a streamlined version of the real author, an actual or purported subset of the real author's capacities, traits, attitudes, beliefs, values, and other properties that play a role in the construction of the particular text that provides an important critical distinction between the flesh-and-blood author and the author-image constructed by readers [italics in original]. 17 In this sense, Atonement is a valuable test of the stability of this concept because in Briony it introduces an embodied author-image who is also a character in the narrative, but does not do so until the closing sections. Parts one, two and three of Atonement are seemingly narrated by an extradiegetic thirdperson narrator, and it is only at the end of Part Three that the initials 'BT' (Briony Tallis) reveal Briony as the 'author' of the preceding narrative. In the 'London, 1999' section which follows, the implied author, whose perspective has been indexed throughout the earlier sections, is embodied in a personified narrator: extradiegetic narration becomes homodiegetic, but only retrospectively so. 18 There have been earlier clues to the identity of the author, most notably the fictional rejection letter from Cyril Connolly included in Part Three, but it is only in the closing pages of the novel that Briony is revealed as the author-narrator, and it is only here that the reader becomes fully aware that events have been narrated from her perspective and her ethical standpoint. 19 Whilst the arguments of Booth, Phelan and others are in different ways persuasive, it is my contention that whilst Briony is, in a literal sense, the implied author within Atonement's storyworld, her role is best described as that of narrator, a role which encompasses a range of functions to which the category of implied author would represent an unnecessary addition between Briony-as-narrator and McEwan-as-author. Briony would not conventionally be described as the narrator of parts one, two and three of the novel, but it is she, it is revealed at the close, whose judgements, opinions and inventions have shaped the way in which the story was told. As Phelan argues in Experiencing Fiction (2007), from the opening scenes at the Tallises' house, 'McEwan, not surprisingly, clearly signals that each of [Briony's] judgements is erroneous. Through several scenes dominated by the technique of internal focalization, McEwan presents Robbie as an admirable young man.' 20 It is, of course, McEwan who has included these signals as to Robbie's good character and therefore his innocence, but within the storyworld it is Briony who has reconstructed these scenes and who has so thoroughly revised her opinion of Robbie as to turn him from villain to an almost saintly figure. Therefore, whilst it could be argued that the category of the implied author serves a useful function in identifying an authorimage with an explicit agenda, or what Booth would refer to as a set of 'norms' against which characters and events are judged, 21 this role is accounted for by the revelation that Briony is the author-narrator, separate from McEwan but still an identifiable presence, whose thoughts, feelings, perceptions and prejudices shape the way in which characters and events are recounted. This approach to the novel supports Walsh's claim that 'fictions are narrated by their authors, or by characters' and that the categories of narrator and implied author are unnecessary additions (after all, Briony is an undeclared intradiegetic narrator). 22 Although the novel does demand attentiveness to an authorial presence who is not the flesh-and-blood author, in a manoeuvre that could be read as a literalisation of the relationship between author and character/narrator, this is embodied in the character of Briony, who is surely also the narrator.
The move from anonymous implied author to embodied author-narrator in Atonement might, and I argue should, call for a revised understanding not only of Briony's perceptions as a child but also of the revision and reworking of these perceptions over the decades between 1935 and 1999. Phelan's reading identifies an important aspect of the developing case for Robbie's innocenceand Briony's atonementbut at the same time neglects to account for the presence of the adult Briony's hand in constructing a case for Robbie's innocence, as well as her naivety and to some degree vindictiveness. The same applies to the characterisation of Paul Marshall, who over the course of Briony's years of revision and reassessment is recast from the role of inconsequential bystander to that of cold and manipulative villain. If, as Seymour Chatman suggests in Coming to Terms (1990), '[t]he implied author says nothing', meaning any deviation from manifest meaning must occur 'between the lines', 23 Briony's role as both author and character presents a challenge to the neutrality of this position and to habits of reading which ascribe a degree of stability and reliability to heterodiegetic narration. Indeed, the presence of Briony-as-author is evident not only in the depictions of character behaviour -Robbie teaches Briony to swim and saves her when she jumps in to the river; Marshall is a war profiteer who manufactures substandard chocolate to sell to the troops; Hardman leers at the young women at the housebut also in persistent but more subtle distinctions in characterisation. As John Frow notes in Character and Person (2015), names are important markers in narratives, and the most fundamental questions to ask about the names of fictional characters […] have to do with how they are pinned to the quasi-persons of a text (how we come to recognize a name as an index of the textual existence of a character), the kind of information and the quality of coherence attaching to the name, and their relation to a modal world. 24 Proper names define and locate characters in relation to one another and in relation to the storyworld, and it is significant that Robbie Turner is known as 'Robbie' throughout, whereas Paul Marshall and Danny Hardman are always 'Marshall' and 'Hardman'. This distinction is related to questions of class and familiarity, because as Frow again points out, they represent an 'index in power and status', 25 but it is also related to questions of distance and proximity, suspicion and sympathy. Readers are allowed to be on first name terms with Robbie, even to experience his interior life, but Marshall and Hardman remain Marshall and Hardman, always at a remove, always under suspicion. 26 It is important to note at this point that Briony is the mediating presence throughout and that these names index her relation to the characters, and by implication the reader's. The same applies to physical descriptions, whereby Robbie is described as having a 'rugged, friendly face' and as being 'startlingly handsome', 27 while Marshall's face is described variously as 'comically brooding', sprouting tufts of hair 'kinked like pubic hair', 'absurd', 'like an over-furnished bedroom', 28 and most commonly as 'cruel'. 29 Hardman is a persistent, uncomfortable presence, and the single physical description of him notes how 'the childish bow of his lips had become elongated and innocently cruel'. 30 'Cruel' is a recurring adjective for these characters and is also applied to Lola, the victim of the assault, when in old age Briony describes her as having 'a touch of the stage villain', suggesting that with the addition of a couple of props 'she could have been Cruella de Vil.' This last description comes immediately after Briony's admission that she finds Lola's heavy makeup distasteful and that in this respect she is an 'unreliable witness', unable to disentangle her writing from her biases and judgements. 31 Briony's admission of unreliability emphasises the fact that Lola, Briony's childhood rival and the victim of the assault, has been side-lined throughout the novel and is largely treated as a catalyst for the plot: Robbie is the true victim.
Throughout Atonement, Briony's narration casts Robbie as the wronged hero of her story, while Marshall and to a lesser extent Hardman are constructed as shifty and untrustworthy, a pattern introduced in Part One but which continues through Part Two, which details Robbie's wartime service (as researched, reconstructed and reimagined by Briony), and Part Three, which details Briony's own wartime experiences as a nurse, a self-imposed penance after she defers university to follow in her estranged sister's footsteps. Part Three concludes with a semi-hopeful meeting between Briony, Robbie and Cecelia, in the latter's shared flat in Balham, South London. Following this meeting, Briony realises what is required to make amends: 'Not simply a letter, but a new draft, an atonement, and she was ready to begin.' 32 This realisation is swiftly followed by the revealing coda 'BT, London 1999'. In an example of what Brian Richardson describes as 'denarration', 'in which a narrator denies significant aspects of his or her narrative that had earlier been presented as given', 33 the final 'London, 1999' diary entry reveals that the meeting in Balham did not happen and that Robbie died of septicaemia in 1940 during the Dunkirk evacuations. Cecelia died in the same year, when Balham Underground station was bombed. This further twist, in which Briony admits to favouring the narratively satisfying over the historically true, has been described by Peter Boxall as a way of testing 'the relationship between narrative form and a historical "actuality" -Einstein's "real factual situation"to which narrative struggles and fails to bear adequate witness', a test which, crucially, 'demands that we undertake a second reading, in which we become conscious at all times of a double focus, and of contradictory drives running though the novel'. 34 This is a persuasive reading, which makes an important argument for an understanding of Atonement that goes beyond the postmodern focus on narrative in/as history. James Phelan and others have taken a similar approach, reading the novel as among other things an exploration of the complex relationship between fiction and history, and an examination of the implications of this for the ethics of storytelling. 35 Yet although many of these readings insist on the necessity of rereading the novel with what Boxall describes as a 'double focus', they have not so far inspired a more radical rereading which concludes not simply that Briony has 'made it all up' but which probes some of the fundamental orthodoxies that Briony's persuasive and self-justificatory narrative has entrenched. As Richardson argues of this type of denarrated novel: Most readers will probably grant these narrators the benefit of the doubt, and may even find a token of authenticity in the corrected representation of the narrator's erroneous path; the storyworld will remain stable as the transformation is limited to the narrator's act of storytelling. 36 I would argue, however, that the revelations at the end of the novel require a rereading that engages in a more wholesale revision of one's understanding of the preceding narrative, a revision which asks fundamental questions not simply about the factual accuracy of this version of events within McEwan's storyworld, but also the manner in which those events have been narrated and the motivations of the author-narrator. Whilst acknowledging Briony's unreliabilityas she herself doesthere is still a sense in which the seductiveness of her version of events holds sway. Alan Palmer gestures towards this when, in an otherwise positive review of Phelan's Experiencing Fiction, he questions Phelan's conclusion that Briony's revelation marks 'the seam between history and fiction', 37 noting that he 'does not take sufficient account of the tendentious nature of the whole of Briony's novel, not just those parts that are eventually found to be fantasy'; he concludes that 'to refer to Briony's novel up to her fantasy meeting with Cecelia and Robbie as history is to underestimate […] the substantial epistemological, ontological, and discursive difficulties that are present within the whole text of Briony's novel.' 38 This is an astute reading, which identifies the extent to which the revelations in the concluding section require a rereading not only of the elements revealed to have been fictionalised but also a deeper engagement with the fundamental 'epistemological, ontological, and discursive difficulties' introduced in this moment. Central to this is the question of (un)reliability and the ways in which the final section of Atonement introduces instabilities into aspects of the narrative not initially subjected to such a fundamental degree of suspicion.

Narrative (un)reliability and discordant reading
Like the implied author, the concept of the unreliable narrator comes from Booth's The Rhetoric of Fiction and his well-known statement that 'a narrator [is] reliable when he [sic] speaks for or acts in accordance with the norms of the work (which is to say, the implied author's norms), unreliable when he does not'. 39 Booth's formulation remains central to these discussions, but the definition and typology of unreliability is a continued subject of debate, to the extent that it has become what Elke D'hoker and Gunther Martins describe as 'a theoretical touchstone for the distinction between story and discourse in narratology as well as one of the (very few) defining signposts of fictionality.' 40 More recent work has tended, in line with broader tendencies in postclassical narratology, to place greater emphasis on the role of readers in constructing meaning, as in Tamar Yacobi's and Ansgar Nünning's work from the 1980s onwards. 41 Indeed, as Bruno Zerweck suggests in 'Historicizing Unreliable Narration' (2001), unreliability has become naturalised in contemporary fiction, even contemporary realist fiction, because 'subjectivity and unreliability are accepted as realities, and reliability is regarded as an impossibility', to the extent that apparent reliability is suspect, 'whereas a narrator who exposes his [sic] cognitive or epistemological limitations is arguably much more in tune with our notions of 'normality' and of the possibilities of its fictional representation.' 42 This is why, in the opening pages of Atonement, Briony's belief in the type of old-fashioned romance represented by her play The Trials of Arabella, and her notion that 'an unruly world could be made just so' 43that stories have their own orderly logicis likely to be read with a degree of irony, even first time around. Moreover, although unreliable narration is most usually applied to first-person narratives, as Dorrit Cohn argues in her reading of 'discordant narration' in Thomas Mann's Death in Venice (1912), 'third-person novels, no less than firstperson novels, allow for the separation of the narrator from the author, and […] a discordant reading of such novels can have major interpretative implications'; 44 that is, by being attentive to the interplay of voices and points of view, one can discover moments of discord or disagreement between what is said and what is implied. This requires, Cohn suggests, not a choice between interpretations, but rather 'a self-conscious reading that understands the choices involved, a reading aware of the fact that there are choices involved, that the problems created by certain types of narratorsnarrators in whom one can spot incongruities in their evaluations of the events and characters of the story they tellcan be resolved in different ways'. 45 This category of discordant narration, and the reading choices it demands, works for Atonement, though again McEwan's novel is an unusual case. The separation between author and narrator that takes place during a discordant reading is in fact enacted in Atonement during the metanarrative reveal in the final section, contradicting Dieter Meindl's conclusion that 'third-person fiction's narrative agency is basically reliable because it does not figure as a person or subject'. 46 In this instance, there is an implied and ultimately embodied narrator, albeit one whose presence is not fully apparent until the final pages of a first reading. Meindl's schema does allow for 'narrative embedding' of potentially unreliable, limited-scope narration marked by third-person pronouns (within free indirect discourse, for example), 47 but whilst this may account for the manifest subjectivity and unreliability of Part One of Atonement, it does not account for unreliability in the superficially more objective and reliable parts two and three. As Naomi Booth argues in a recent article on Atonement, the conclusion of the novel has more far-reaching implications because 'McEwan structurally undermines the authority of omniscient narration by foregrounding Briony's construction of this perspective'; 48 the apparent objectivity of third-person narration is no such thing, and a discussion of this very fact is embedded throughout the novel, not only in the closing pages.
In Part Three of Atonement, Briony-as-author ('BT') recalls Briony-ascharacter writing her journal and fictionalising the lives of staff and patients at the hospital: 'having changed the names, it became easier to transform the circumstances and invent. She liked to write out what she imagined to be their rambling thoughts. She was under no obligation to the truth, she had promised no one a chronicle'. This passage describes a private journal, a writing exercise, 'the only place [Briony] could be free' during her time on the ward, and a chance to continue her writing in the knowledge that 'she might look and behave like and live the life of a trainee nurse, but she was really an important writer in disguise'. And while this passage does not necessarily constitute an admission of a more fundamental degree of fabrication, a short instance of prolepsis embedded in the same passage does gesture towards Briony's later role as author, with the implication that her account is partial and inaccurate. 'In later years', the passage continues, 'she regretted not being more factual, not providing herself with a store of raw material'; 49 this moment, an example of what Mark Currie calls 'teleological retrospect', both acknowledges the later perspective from which this section is narrated and, as Currie suggests of this type of narration, confers on it a significance it did not possess at the time. 50 Though easily overlooked, moments such as these occur throughout Atonement and emphasise the inherent unreliability of Briony's account and its retrospective reconstruction from memory and later research: there are revisions and refinements that, given the novel's repeated references to the limitations of subjectivity and memory, cannot be read as a gradual winnowing away of falsehood to reveal the hard grain of truth. Similarly, whilst the final 'London, 1999' section reveals that Briony has consulted archival material (including the wartime letters between Cecelia and Robbie) to reconstruct her account of the 1940 British retreat to Dunkirk, she makes a significant creative leap from these documents to the reimagination of Robbie's interior life that takes place in Part Two. 51 Throughout the novel, particularly when read in the knowledge of the surprise conclusion, there are markers of a fundamental uncertainty and unreliability, all pointing towards Briony as the source. Richardson describes her as an example of the '"duplicitous" narrator […] who deliberately provides information that is later falsified by subsequent statements in the narration', and it is Briony's double duplicityfirst as a child and then as an adultwhich remains unchallenged. 52 Why, then, is Briony's account taken at face value second time around, even by professional readers? In the final section I want to suggest that research in reading practices and cognitive narratology offers one framework for understanding why Robbie might be guilty and why nobody seems to notice.

Source-monitoring and metarepresentation
In Why We Read Fiction (2006), Lisa Zunshine draws on work by the evolutionary psychologists Leda Cosmides and John Tooby (2000) to discuss the ways in which metarepresentational 'source tagging' informs the construction of meaning in narrative. Source tags are time-specifying, placespecifying or agent-specifying markers that structure memories and, Zunshine argues, meaning in literary narratives. Narratives are 'perennially stored with either variously implicit source tags, such as 'folk' in the case of Little Red Riding Hood and as 'Anglo-Saxon bard(s)' in the case of Beowulf, or explicit source tags, such as 'Jane Austen' in the case of Pride and Prejudice'. 53 Markers such as, 'as she stood in the nursery' (place-specifying), 'she sensed' (agent-specifying) and 'during a heatwave in 1935' (time-specifying) enable readers to locate agents within narratives and structure the way in which characters and events are interpreted and remembered. 54 Furthermore, these source tags have an important function in identifying (un)reliability. In Kazuo Ishiguro's The Remains of the Day (1989), for example, '[w]e come into the awareness of the missing tag' in Stevens's narration when he notes that 'to any objective observer […] the English landscape is the most deeply satisfying in the world'. 55 Drawing on Phelan's reading of Ishiguro, Zunshine argues that in passages such as this: We contemplate various ramifications of the difference between the two representations ('the English landscape is the most deeply satisfying in the world' vs. 'Stevens thinks that the English landscape is the most deeply satisfying in the world') that jostle against each other in our readerly consciousness. We begin to wonder what other representations within the story may also be missing their source tags.
Such uncertainty means that we 'continue guessing which representations within the story deserve to be treated as "true" and which have to remain metarepresentations with a source tag pointing to the first-person narrator', creating a tension between the 'facts' as they are narrated and the facts as they are inferred and interpreted. 56 Zunshine's work is provocative and has been challenged from a number of quarters, not least by critics such as Brian Boyd who argue that her adoption of Theory of Mind models is too selective and not sufficiently rigorous to support her far-reaching claims. 57 Indeed, the field of cognitive literary studies is still relatively young and is fraught with challenges concerning, for example, the compatibility of scientific and literary critical models, and a frequent tendency to emphasise the universal and transhistorical at the expense of the particular. 58 Despite sharing some of the reservations expressed elsewhere about this work, to my mind Zunshine's argument about the relationship between source tags and reading practices is persuasive and helps account for some of the elisions in critical readings of Atonement that have been identified throughout this essay. Most significantly, the agent-specifying source tags pointing to Briony are missing from a first reading of Atonement, in which Briony's role as the author-narrator is largely concealed: the source tag 'according to Briony' can only be applied retrospectively following the revelations towards the end of the novel. Indeed, even at the paratextual level, the title Atonement has an authoritative weight not apparent in Henry James's What Maisie Knew (1897) or Beryl Bainbridge's According to Queeney (2001), other novels of youthful misapprehension and confusion whose titles suggest a more subjective focus. 59 Beyond the title, in each of the three longest sections of Atonement there are a series of source tags which must be read back in to the text in order to account for the levels of subjectivity and revisionism in the novel. The occasional examples of prolepsis in the opening section ('Briony was hardly to know it then') hint at a future authorial perspective beyond the events being narrated, but subsequent readings require further levels of framing, and for an implied 'Decades later Briony would recall that her younger self felt/saw/said … ' or 'Decades later Briony would recall/interpret/imagine that X felt/saw/said … '. The levels of intentionality accrete, meaning that a whole series of frames and caveats must be applied, based on knowledge gained from the previous reading. There is a curious parallel here with Jane Austen's Emma (1816), in which Zunshine suggests the relationship between Frank Churchill and Jane Fairfax is signalled with 'relatively weak metarepresentational framing' 60 so that, unlike with a detective novel, the reader is not expecting a sudden reversal at the end and has to embark on a more fundamental and unexpected process of reassessment and rereading. David Herman describes this as a process of 'unworlding', in which 'readers are likely to bracket or "derealize" the world they have constructed up to this point', interpreting the storyworld in the preceding chapters 'as a fiction within the fiction, rather than the baseline reality or "textual actual world"'. 61 Yet the first reading, in which Briony attempts to atone for her past mistakes, identifies the true culprit, and exonerates a wronged man, to a greater extent seems to survive this process of unworlding. That is to say, so compelling is the story and so unexpected is the reveal that a more fundamental revision of Briony's account has yet to take place: despite suspicion over the details of Briony's story, Synopsis (A) remains broadly intact, whereas Synopses (B) and (C), as well as other possible interpretations, are dismissed or overlooked. Recent work in narrative theory has emphasised the distinction between reading and rereading, and the ways in which rereading can bring to bear a new understanding of a text which may be significantly at odds with the first reading, 62 and critics writing about Atonement have similarly emphasised the need to reread the novel to comprehend the implications of Briony's authorship and invention. Yet no discussion of the novel has so far accounted for the force of the first reading, and the ways in which Briony's attempt at atonement and revision of the historical record is so firmly embedded in the text that despite her admission of inventionsand despite the novel's pre-occupation with uncertaintyher final version of the story remains intact in its essentials. The source tags are both too weak, in that for the majority of the novel it is not clear that Briony is the author-narrator, and too strong, in that Briony presents a compelling case for the truthfulness and accuracy of her narrative of atonement. Briony is imbricated in the text as character and author-narrator, and her confidence in her final act of atonement is difficult to challenge despite the many questions that remain. Richard Robinson puts it well when he describes how, like Joyce's Stephen Dedalus: Briony […] is 'within' the work, but her perspective is surrounded and undercut by the competing viewpoints of the other characters. She is 'behind' it, in the sense that she is covertly responsible for it. She is 'beyond' it in two senses: temporally, she waits for us in the epilogue, outside of the novel proper that she has completed; the publication of the previously libellous text also confirms she has now died and is beyond reach. 63

Concluding remarks
The 'truth' about who assaulted Lola is also beyond reach, and the intention here has not been to confirm an old suspect or to identify a new one. Rather, it has been to identify an aporia in critical discussions of Atonement and to identify both how recent narrative theory can help to identify and explain this aporia, and how this critical rereading of Atonement might contribute to debates concerning the implied author, narrative unreliability, and metarepresentation. In a famous passage from Writing Degree Zero (1953), Roland Barthes uses the example of Agatha Christie's The Murder of Roger Ackroyd (1926) to comment on 'the ambiguous function disclosed in the preterite', the defining tense form of the realist novel and, Barthes argues, 'the very act by which society affirms its possession of its past and its possibility.' In Christie's novel, 'all the invention consisted in concealing the murderer beneath the use of the first person of the narrative. The reader looked for him behind every "he" in the plot: he was all the time hidden under the "I".' 64 Atonement represents almost the inversion of this, in which the unreliable 'I' hides behind the seemingly more objective 'he' or 'she' and is only unmasked at the end. A truly discordant reading, which takes into account the complexities of narration in this novel, must consider this movement and entertain a more radical doubt as to what Briony saw, how it is narrated, and why Robbie might be guilty.