Strategies of text-world consolidation in reviews of Margaret Atwood’s Surfacing

Abstract

Text World Theory (Gavins, 2007; Werth, 1999) offers a cognitive linguistic account of the mental representations created during discourse comprehension. To date, text-world accounts of comprehension have largely focussed on the mental representations created in the moment of discourse processing, and little attention has so far been paid to how text-world representations change over time. However, the comprehension of novel-length fictional narratives requires readers to draw upon large amounts of text-specific information as they read later sections of a text. This paper reports an exploratory study of 100 reviews of Margaret Atwood’s novel Surfacing, posted to the Goodreads website. 50 precis of the novel are isolated, and the text-world conceptual structures of these precis are compared to the original text. Several potential consolidation strategies are identified to account for how text-world mental representations change as the novel is remembered and later recalled. In particular, evidence is presented to show that readers create an evolving mental representation of the fictional world projected by the text which is built and maintained in the long-term memory and remains distinct from the text-world mental representations created in the moment of reading. In the light of these findings, an argument is made for an expanded Text World Theory which accounts for readers’ long-term memories of fictional texts.

1. Introduction

One factor which typically distinguishes literary narratives from conversational narratives is their length. Most short stories are many times longer than narratives told in face-to-face settings, while novels are typically read in more than one sitting over a period of days or weeks. A full account of the comprehension of literary narratives must therefore address processes of memory and recall.

Most theories of human memory (e.g., Atkinson and Shiffrin, 1968; Baddeley, 1999, 2010; Richardson et al., 1996) distinguish two discrete cognitive systems: working memory, with a small capacity and a duration of a few seconds, and long-term memory. Memories of the same events may be structured differently in the two systems. For example, studies have shown that experimental subjects are more likely to confuse similar sounding words if tested immediately after rehearsal, but more likely to confuse words with similar meanings if tested after a longer interval (Baddeley, 1999: 35), suggesting that the precise linguistic formulation of a text is not recalled in the longer term.

Text World Theory (Gavins, 2007; Werth, 1999) is offered as a theory of the mental representations that discourse participants create while processing both spoken and written language. To date, however, theorists have largely avoided making specific claims about the structure of such representations and how they relate to working and long-term memory systems. If Text World Theory is to offer a comprehensive cognitive account of literary narrative comprehension, it must seek to explain not only the mental representations that readers create in the moment of reading, but also how such representations are remembered and later recalled.

Text World Theory draws on the framework of mental spaces, with Werth stating explicitly that text-worlds fall ‘within the definition of “mental space” of Fauconnier’ (Werth, 1999: 20; see also Fauconnier, 1985, 1997). Fauconnier and Turner have claimed that mental spaces ‘operate in the working memory but are built up partly by activating structures available from long-term memory’ (2002: 102). This paper therefore begins with a hypothesis that text-world mental representations might be considered as dynamic structures created in working memory. It then seeks to explore how these dynamic structures might change as a result of processes of consolidation, as information from the working memory is transferred to long-term memory and later retrieved for use.

While little work has been done on the memory of text-worlds, a substantial body of work exists on textual recall more generally. Extensive evidence shows readers do not typically retain precise linguistic structures over time but instead remember the situations that texts describe (Bransford and Franks, 1971; Elfenbein, 2018: 100; Reyna and Kiernan, 1994). When discussing memories of narrative texts, scholars have frequently drawn on theories of ‘schemata’ or ‘frames’ to account for how readers’ prior knowledge shapes comprehension and later recall, building on work by Bartlett who suggested that memories of narratives are likely to become more coherent and contain clearer causal links over time (Bartlett, 1932; for examples of cognitive stylistic discussions of schemata in narrative comprehension see Cook, 1994; Emmott, 1997; Mason, 2019; Stockwell, 2003; Whitt, 2024). From a Text World Theory perspective, Werth (1994, 1999: 323) and Gavins (2007: 151) have explored how ‘extended metaphors’ (also known as ‘megametaphors’) can function across long discourses to shape interpretation. However, few claims have yet been made about what happens to text-world mental representations over time.

One exception to this is a recent study by Gibbons (2023). Gibbons attempts to draw parallels between Text World Theory and psycholinguistic theories of comprehension, and specifically with the situation model account of how readers consolidate long discourses dynamically in the mind (Graesser et al., 1997; Kintsch, 1995; Van Dijk and Kintsch, 1983; Zwaan et al., 1995; Zwaan and Radvansky, 1998). Gibbons proposes the concept of ‘world-retrieval’ to account for ways in which readers consolidate information currently being processed in the working memory with text-worlds that have previously been experienced. Her arguments highlight the benefits to be gained from a more rigorous focus on how text-worlds are recalled and consolidated over time.

This paper examines a sample of precis of Margaret Atwood’s Surfacing (1979), taken from the Goodreads website (https://www.goodreads.com/), focussing on the text-world mental representations prompted by readers’ descriptions of the text. Summaries posted to book review websites cannot be equated with reviewers’ comprehensive memories of a novel. In particular, a significant amount of consolidation is likely to occur at the point where readers recall a text and subsequently craft a response that is appropriate for the Goodreads platform. This is likely to involve not only processes of memory retrieval, but also subjective judgements about the relative importance of different features of the narrative. Nonetheless, I will argue that, while significance cannot be attached to missing information in reader summaries, structural differences between reader summaries and the original text may reveal evidence of cognitive processes that readers employ when reading and recalling narratives. I will then draw on this analysis to offer hypotheses about potential text-world consolidation strategies that could be tested using different methodologies.

2. Atwood’s Surfacing

Surfacing (1979) is Margaret Atwood’s second published novel. The unnamed narrator describes returning with three friends to her childhood home in northern Canada, to search for her father who has been reported missing. The realist tone established in the opening pages is slowly subverted as the narrator’s account of her journey is populated with increasingly outlandish claims, culminating in a description of a dream-like encounter with her dead parents. Many readers of the novel conclude the narrator undergoes some form of mental breakdown, though several also note that Atwood’s narrative style leaves room for multiple interpretations. For example, one reviewer on the Goodreads platform suggests that ‘[t]he protagonist’s madness offers nice opportunities for strategic ambiguity’ while another claims that ‘[i]t is intended that the reader be confused.’

Academic reviewers of Surfacing have noted how the novel reflects many themes common in Atwood’s early fiction, including gender politics and notions of Canadian identity (for introductions to critical approaches to Atwood’s prose see Howells, 2021; Tolan, 2022; Wisker, 2012). Much attention has also been paid to the novel as an example of unreliable narration (Booth, 1961: 158; Phelan, 2005, 2017; Rabinowitz, 1977), a theme which is discussed frequently in relation to Atwood’s oeuvre. For Clark, the narrator is guilty of ‘constantly presenting so-called facts and subsequently subverting and exposing them as fiction’ (1983: 3). Similarly, for Bouson, the narrator’s ‘seeming “realistic” and initially convincing story’ is revealed to be ‘pure fabrication’, meaning ‘what has appeared to be narrative “truth” is exposed as an elaborate fictional construct’ (1993: 41). The extent to which the narrator’s account can be trusted is frequently addressed by contributors to Goodreads. One reviewer noted ‘[t]he narrator is unreliable and so is the end. It leaves everything open’, while another suggested that ‘[w]hat surfaces throughout the narrative [is] the unreliable continuity of memory’. A third was even more blunt in their assessment, claiming that ‘everything the woman has told us at the start of the story is revealed to be a lie’.

For these reasons, the novel provides an ideal case study for an exploration of the recall of fictional narrative. As the testimonies from both professional critics and Goodreads reviewers show, comprehending Surfacing involves a continual process of comparing later claims by the narrator to a growing body of information about the fictional world of the text that has already been read and remembered. In other words, we cannot consider readers’ mental representations of Surfacing to be static entities. A full account of comprehension must explain how states of affairs described in later passages are compared with an evolving model of a fictional world in the long-term memory. One way to embark on a systematic investigation of such processes is to attempt to account for the comprehension of Surfacing using Text World Theory.

3. Text World Theory and unreliable narration

Text World Theory (Gavins, 2007; Werth, 1999) offers a rigorous cognitive linguistic account of discourse comprehension. It suggests that discourse comprehension always involves the creation of at least two distinct mental representations at two distinct ‘conceptual levels’ (Gavins, 2007: 9). Firstly there is the ‘discourse-world’, in which a discourse participant conceptualises themselves and their fellow discourse participants. Distinct from this are one or more ‘text-worlds’ in which they conceptualise states of affairs which the discourse describes. These two conceptual levels remain ontologically discrete, even when participants are discussing the situation they currently find themselves in (Gavins, 2007: 76). As discourse progresses, text-worlds are populated by ‘world-building elements’ (Gavins, 2007: 36), and evolve as a result of ‘function-advancing propositions’ (Gavins, 2007: 56). Subsequent text-worlds may be created if a narrative describes events happening at more than one spatial or temporal location; a phenomenon Gavins describes as a ‘world-switch’ (2007: 48). The comprehension of fictional narratives involves the additional understanding that states of affairs conceptualised in the text-world(s) prompted by the text may have no counterpart in the discourse-world.

Text World Theory also acknowledges that discourses themselves may describe states of affairs at different ontological levels. For example, fictional narratives typically describe beliefs held by characters about the fictional world, which may not be true in that fictional world. They can also describe situations that characters desire to bring about or fear will become reality. In Gavins’ (2007) formulation of Text World Theory, the mental representations required to conceptualise these alternative states of affairs are known as ‘modal-worlds’, and are triggered by specific linguistic features such as modal adverbs or modal lexical verbs (for a more comprehensive list see Norledge, 2022: 64). Indeed, a major strength of Text World Theory is that it allows analysts to produce a replicable list of each discrete mental representation it predicts listeners or readers will form in response to text. Gavins refers to this inventory of text-worlds, modal-worlds and the relationships between them as the ‘conceptual structure’ of a text (2007: 35, 40, 46).

In a homodiegetic narrative such as Surfacing, readers have no access to states of affairs in the world inhabited by the narrator, other than through the narrator’s testimony. This has led Gavins (2007: 133) to claim that the states of affairs described by a narrator are conceptualised within an ‘epistemic modal-world’, while the act of narration is conceptualised in a separate initial text-world. Lahey describes the initial worlds produced during homodiegetic narration as ‘empty’, meaning they contain no world-building elements other than an enactor of the narrator (Lahey, 2004: 27). Gavins describes such worlds as ‘normally text-initial but ultimately immaterial’ (2007: 133).

In Gavins’ Text World Theory, the states of affairs conceptualised in modal-worlds and in the initial text-world remain ontologically discrete. The distinction between these constructs can be significant for understanding a text, as Gavins herself demonstrates in her readings of absurd fiction (2013: 88–90). However, in many homodiegetic narratives, readers take much of what the narrator says at face-value. Moreover, in many fictional narratives, characters and narrators - and, by extension, readers - frequently learn much about a fictional world from other characters. For example, literature abounds with protagonists who gain significant knowledge via the testimony of an archetypal ‘dispatcher’ figure (Propp, 1928) who makes the protagonist aware of particular circumstances and sends them on their quest. Text World Theory does not currently account for the specific processes by which the claims made by a homodiegetic narrator, and the beliefs of, or claims by, fictional characters, initially conceptualised in ontologically discrete representations, may be combined or otherwise consolidated to provide a single coherent representation of the fictional world in which the narrator resides.

This presents particular challenges for accounts of unreliable narration, where some, but not all claims made by a narrator are treated as reliable. From a logical perspective, it is not possible for readers to make claims about the truth of states of affairs in an empty initial text-world, which are not specified by the text. Yet reader accounts of unreliable narratives (such as those quoted in Section 2) suggest that many readers do devote considerable effort to determining how events in the epistemic modal-worlds triggered by the narrator’s testimony differ from those in the ‘empty’ text-world in which the narrator resides. For this reason, Atwood’s Surfacing provides an ideal case study to explore how text-worlds are consolidated over time.

4. Methodology: Analysis of Goodreads reviews

In this section, I report the results of an analysis of Goodreads reviews of Surfacing. Focussing on user-generated book reviews presents both advantages and problems for a study of this kind. Such reviews provide access to responses that are more ‘naturalistic’ (Peplow and Carter, 2017) than those gathered under more controlled conditions. However, in any public discussion group dynamics (as discussed by Peplow et al., 2015; Peplow and Whiteley, 2021) are likely to play a part in shaping reader responses, and in determining which responses are ‘liked’ and hence foregrounded by other readers. In particular, Goodreads contributions are shaped by shared assumptions about what should or should not be included in a book review. For example, it is a convention that reviews should not rely heavily on discussions of plot developments from the end of the text which act as ‘spoilers’. Care must therefore be taken about what can be concluded from such reviews. In this study, I make no claims that information missing from reader reviews can be interpreted as evidence that details have not been remembered, and acknowledge that individual readers would be likely to produce different responses to the same text under different conditions. This notwithstanding, I suggest that when reviewers produce reviews in which information is presented in a different conceptual structure from that in which it is presented in the original text, this can be read as evidence of consolidation processes that have taken place between reading and review writing.

Disadvantages of this approach are offset by the large number of reviews that can be gathered with minimal effort, meaning that such analyses are often suitable for a ‘“framework generating” preliminary study’ (Mason, 2019: 33). Stylisticians frequently draw on such data by quoting from individual reviews to demonstrate that particular readings of a text occur in naturalistic settings (see e.g., Mansworth, 2022; Norledge, 2020; Nuttall, 2015). However, the availability of large numbers of reviews means it is also possible to adopt a quantitative approach to such analyses, as demonstrated by Gavins (2013) and by Nuttall (2018) and Mason (2019), who have piloted systematic approaches to gathering representative collections of reviews for analysis. In this study I adopted a similar approach to Mason. In Section 4.1 I outline how I isolated a sample of 100 Goodreads reviews of Surfacing, using a methodology that could be replicated for other texts. In Section 4.2 I then describe how I isolated a sub-sample of 50 precis of the novel for text-world analysis.

4.1 Assembling a sample of Goodreads reviews of Surfacing

On Goodreads, individual novels often receive many thousands of reviews. Practical constraints prevent an analyst reviewing every contribution. Doing so would also make it hard to compare texts, as some texts receive many more reviews than others. For these reasons, isolating a discrete sample of reviews allows methodical analysis and comparison.

Goodreads reviewers normally accompany their reviews with ‘star’ ratings of between one and five stars. The platform offers no guidance on how to assign star ratings, but there appears to be a consensus that stars should be awarded on the basis of a judgement of the text’s overall quality, with reviewers frequently explaining how they have balanced their personal enjoyment of a text with its technical merit when assigning a score. Furthermore, plenty of reviewers indicate they think carefully before assigning such ratings. As well as submitting their own reviews, readers can also comment on and ‘like’ reviews by other readers. Again, no guidance is given on the platform about how users should assign ‘likes’, but a pilot investigation indicated that posts that attracted more likes tended to be longer and to contain more nuanced descriptions of specific reading experiences. It is noted, however, that no evidence is available to support the idea that a ‘like’ indicates that a reader agrees with the sentiments expressed in a review, so care must be taken when imputing meaning to this metric. In other words, star ratings and likes are both crude measures, but they do provide opportunities for quantitative analysis.

My primary consideration in this study was to assemble a sample containing 20 reviews of each possible star rating, to ensure the widest possible range of responses was represented. As a secondary consideration, reviews were selected on the basis of those the site ranked as ‘popular’ (a metric based heavily but not exclusively on the number of ‘likes’ they had received), as this was judged to be a way of identifying longer (and potentially more useful) reviews while minimising the chance of introducing experimenter bias. It is acknowledged that this approach may generate a sample which does not necessarily reflect the distribution of reviews on the site (which tended to skew towards higher star ratings, at least for Atwood’s texts). However, it has the advantage that, if followed for different texts, it allows comparisons to be drawn between corpora for different texts; a benefit which was judged to outweigh the disadvantages.

When assembling the sample, a small number of short reviews were rejected as being unsuitable, as they were judged to contain no relevant information about the reader’s experience of the text. In such cases, the next-most-popular review with the same star rating was substituted. Following this protocol resulted in a sample of just over 24,000 words; a figure, I judged to be at the upper end of what could be analysed practically.

4.2 Isolating discrete ‘precis’ of Surfacing

An initial survey of my Goodreads sample revealed that a significant number of reviews contained similar summaries of what reviewers presented as the novel’s central premise. The following two extracts are offered as indicative examples. ¹

Our unnamed female narrator brings her lover and their two (married) friends to her childhood lakeside cabin in the woods for a brief getaway from life and for the two men to capture some footage for the amateur film they are producing. She hides her true intentions of returning to this familiar lake however. She is trying to find her father. Long missing, our narrator does not presume him dead but instead believes that he is still alive and living by the lake. (GRS-3)

The story starts out very simply with an unnamed woman who returns to her family home, a cabin in the wilderness of Quebec, Canada. This is not a happy homecoming as she had left home years earlier and has not spoken to anyone in her family since. There is the added layer of her reason for going back. She is hoping to find clues to her father’s mysterious disappearance. She is accompanied by her partner, Joe, and a married couple, Anna and David. (GRS-16)

The degree of similarity between these passages is, I suggest, significant. They share many common features without being identical, and suggest a significant degree of independent creative effort on the part of each reviewer. The fact that common semantic meanings (Jackendoff, 1983) can be identified means that stylistic similarities and differences between the passages may be meaningfully compared.

I attempted to define a set of objective criteria by which such precis could be isolated from longer reviews. Ultimately, I was not able to define a single set of necessary and sufficient conditions, but instead decided upon a working definition. To qualify as a precis, a passage had to describe one or more events or states of affairs in the fictional world which I judged had been included by the reader primarily because they believed them to be of key significance to the narrative. This involved a degree of subjective judgement on my part, but allowed me to identify a usable sub-sample of such passages. These passages were frequently marked by one or more of the following features:

• They occurred as discrete descriptions of events at the text-world level, sandwiched between descriptions of events at the discourse-world level.

• They were separated from other parts of the review by paragraph breaks.

• They were explicitly labelled as summaries or similar by the reviewers.

Analysing the sample of 100 reviews produced exactly 50 such passages. Of these, clear start and end points for 48 precis were identified using one or more of the criteria above. In the other two cases, a more subjective decision was required about where the precis began or ended. Thus, while such a selection process is not entirely free from subjective judgement, it meets a reasonable standard of reproducibility, and could be used to assemble comparable data sets for different texts.

Before presenting further analysis, the following caveats are noted. Firstly, we cannot conclude that details missing from such precis are also missing from the mental representations of the text stored in a reviewer’s long-term memory. Secondly, this point reflects a wider truth that such precis are produced as a result of multiple cognitive processes, including processes that occur during or shortly after reading, processes that occur during storage in and retrieval from the long-term memory, and processes that form part of a period of conscious selection and crafting during review writing, after information about the text has been retrieved from long-term memory. On the basis of this study alone, no claims can be made about where in this sequence of events consolidation takes place. Finally, it is noted that the reduction of the original Goodreads sample to a subset of contributions from reviewers who have foregrounded descriptions of plot in their reviews may introduce a bias towards readers who adopt particular reading strategies; for example those who engage in ‘story driven reading’ rather than those who seek ‘empathy’ or ‘imagery vividness’ (Miall and Kuiken, 1995). These caveats not withstanding, the analysis did reveal evidence of particular consolidation processes. I discuss these in the following section.

5. Reader precis of Surfacing: An analysis

In this section I present the results of a text-world analysis of reviewer precis of Surfacing. Comparing the conceptual structure of these precis with that of the source text, I offer hypotheses about text-world consolidation strategies that readers of lengthy fictional narratives may employ as novels are read and later recalled.

5.1 Content

The assembled precis varied between 12 and 373 words in length, with an average length of 79 words. A full discussion of how we might parse such precis in a rigorous way to determine their ‘semantic content’ (Jackendoff, 1983) is beyond the scope of this paper. However, an initial analysis reveals a high degree of similarity in what reviewers chose to report. 24 of the 50 precis (48%) opened with a sentence that began with a noun phrase identifying the narrator and placing her in the subject position. In addition, several other ‘world-building elements’ and ‘function-advancing propositions’ (Gavins, 2007) occurred frequently:

45 reviewers (90%) reported that the narrator made a journey (e.g. ‘A young woman goes home to rural Canada’, GRS-27)

44 reviewers (88%) reported that the narrator was in the company of friends (e.g. ‘A woman travels in the company of friends’, GRS-2)

42 reviewers (84%) reported that the narrator’s father had gone missing (e.g. ‘She is going there to look for her estranged father who has disappearedʹ, GRS-12)

31 reviewers (62%) reported that the narrator was returning to her childhood home (e.g. ‘An unnamed Canadian woman returns to her isolated home’, GRS-23)

The high incidence of these particular features, selected from the many events narrated in the text, may support Emmott’s suggestion that there is a core of narrative information that most readers consider essential to understanding a narrative:

Presumably, there is a certain minimum threshold of information which the average reader might be expected to retain, such as key facts about the major characters and the locus of a particular stretch of action. (Emmott, 1997: 7)

By contrast, only two reviewers (4%) mentioned the narrator’s brother in their precis, and only one (2%) mentioned her mother, despite references to both characters on the first page of the novel and frequently throughout. Though we cannot conclude that these characters had been forgotten by participants, one possible explanation for the higher incidence of references to the narrator’s father may be that his disappearance fits a ‘detective story’ schema that many reviewers made reference to in their reviews. I discuss the idea of schema application in more detail below.

Despite the similarities between many of the precis, a significant minority of readers reported events that the majority of reviewers failed to mention. For comparison, I include the following short precis, quoted in its entirety, which mentioned none of the most commonly recurring plot elements:

[A] woman who turns into an animal living on a remote island. (GRS-56)

While this summary stood out within the sample, it is far from an idiosyncratic reading of the novel, echoing as it does the ‘fairy-tale transformation of sorts’ that Wilson (1993: 11) has noted as an important feature of the text. The presence of atypical precis reflects the fact that summarising a complex narrative is not a deterministic process, and that individual backgrounds and dispositions will ensure that different readers focus on different elements of the text.

5.2 Unreliable narration and empty text-worlds

Surfacing is a homodiegetic narrative, yet all 50 precis transposed the first person account to a third person report. This is likely to be at least in part a consequence of shared expectations of how book reviews or novel precis should be written. Yet the switch from first person to third has implications for how text-world mental representations are consolidated after reading.

In Section 3, I discussed the empty text-world generated by Atwood’s prose, and noted that function-advancing propositions do not populate an initial text-world but rather the epistemic modal-worlds generated by the narrator’s testimony. Reader precis written in the third person do not replicate this conceptual structure and the epistemic nature of function-advancing propositions is lost. Thus, the language used by reviewers presented events as factual statements about the world in which the narrator resides.

A woman travels in the company of friends to a remote island to find out what happened to her father (GRS-2)

The occurrence of such third person precis in reviews did not, however, mean that all reviewers accepted every statement by the narrator uncritically. Plenty of reviewers whose precis did not recreate the empty text-world structure created by the novel, discussed the implications of the novel’s first person narration elsewhere in their reviews. One, who reported the narrator’s return home and the activities of her companions as if they were established facts, went on to offer a detailed discussion of the narrator’s unreliability:

It becomes obvious early on that we are stuck in the mind of a mentally ill young woman. Her grasp on reality is oneiric and muddied [...] our narrator seemingly slips in and out of her reality (GRS-3)

Thus, while the shift from first to third person in precis is not unexpected, the construction of these precis as discrete sections within reviews appears to be indicative of a process by which readers judge some - but not all - of the claims made by the narrator to be accurate descriptions of states of affairs in the empty text-world projected by Atwood’s novel. Reviewers who did remark on the narrator’s unreliability tended to do so outside of the precis section of their review. This trend was epitomised by reviewer GRS-50, who offered factual-style statements about the fictional world projected by Atwood’s text in a paragraph headed ‘synopsis’, and then used a subsequent section labelled ‘review’ to explore more ambiguous claims made by the narrator:

It all got quite confusing and sometimes the story appeared to be in a stasis, where nothing felt real [...] There’s also the mention of war and its accompanying fear and devastation, memories from the narrator’s childhood. While calling her childhood a ‘good’ one, it’s evident there’s a sense of disassociation from her parents. (GRS-50)

This process of assuming that some, but not all, information presented by a homodiegetic narrator can be taken as fair description of states of affairs in the empty text-world projected by a novel is one that is not currently accounted for in Text World Theory.

5.3 Removal of modality

Critics such as Palmer (2004), Ryan (1991) and Zunshine (2003, 2006, 2007, 2015) have suggested that the ability to track a characters’ beliefs and desires alongside their actions is essential to narrative understanding. For this reason, examining patterns of modality in reader summaries is revealing.

We can explore this by comparing indicative examples, each of which was offered as the opening of a precis by Goodreads reviewers:

• The unnamed narrator returns to her father’s home on an island in a lake in Quebec because her father has gone missing. […] (GRS-30)

• A woman goes back to the small Quebec village of her childhood vacation to look for her estranged father, who was reported missing. She doesn’t really want to see him, but she needs to know he is safe. […] (GRS-21)

These examples describe broadly similar scenarios, but contain significant differences. Both reviews invite the reader to create an initial text-world and populate it with a small number of world-building elements. In this case, ‘her father’s home’ and ‘the small Quebec village’ can be interpreted as alternative descriptions of the same referent. However, the language of review GRS-30 results in a single text-world. (The shift from present to present-perfect tense in the clause ‘because her father has gone missing’ does not result in a world-switch, as the purpose of this is to emphasise the father’s missing status in the present moment). Here, the link between the father’s disappearance and the narrator’s response is presented as a relationship of cause and effect, and no information is given about the narrator’s attitude towards her journey or what she hopes to achieve. By contrast, the language of precis GRS-21 prompts the creation of worlds at different ontological levels; a negated boulomaic modal-world cued by the verb ‘want’ in the clause ‘She doesn’t really want to see him’, and a deontic modal-world cued by the verb ‘need’ in ‘she needs to know he is safe’. In addition, it contains the clause ‘to look for her estranged father’, which uses a verb in its infinitive form to describe an as-yet-unrealised intention of the protagonist. Werth (1999: 53) describes such clauses as ‘purpose clause [s]’, and suggests these create what he describes as ‘purpose worlds’, whose contents does not necessarily reflect states of affairs in the future of the parent text-world. Thus, in my reading, this clause is also boulomaic modal-world forming. Taken as a whole, the conceptual structure prompted by the language of review GRS-21 therefore retains more of the ontological complexity of that prompted by the original text.

In the sub-sample of 50 precis, 24 reviewers (48%) offered language which triggered the creation of a boulomaic modal-world containing a function-advancing proposition encapsulating the narrator’s ambition to find her missing father. This was by far the most common modal-world prompted by the precis: a fact which supports Ryan’s suggestion (1991: 156) that modelling a protagonist’s desires may be an important part of narrative understanding. However, this figure is significantly lower than the total number of precis (84%) that made reference to the missing father. A significant number of respondents presented this information using language which did not result in boulomaic modality. This can be seen by comparing four more responses. The first two trigger boulomaic modal-worlds via the verbs ‘hoping’ and ‘trying’:

• She is hoping to find clues to her father’s mysterious disappearance. (GRS-16)

• She hides her true intentions of returning to this familiar lake however. She is trying to find her father. (GRS-3)

In contrast, the second two present ontologically flat descriptions of the same event:

• [Atwood tells the story of] a young woman who returns to her rural childhood home in northern Quebec after receiving word that her elderly father has gone missing. (GRS-73)

• Learning of the disappearance of her father, [the narrator] efficiently organizes a lift back to the home she fled abruptly decades ago. (GRS-45)

There appears, therefore, to be a tendency for some reviewers to offer precis that are ontologically more simple than the texts which prompted them, focussing only on reporting events at the initial text-world level. This results in an important difference between such precis and the original text. Atwood’s Surfacing requires readers to keep track of both states of affairs at the initial text-world level and of the narrator’s attitude towards states of affairs in as-yet-unrealised worlds. It therefore juxtaposes events in a fictional world with a particular subjective perspective on that world. The language of precis such as GRS-73 and GRS-45 does not provide the same contrast of perspectives. Though this more simplistic point of view may be a feature of many narrative summaries, it is an important one to account for in Text World Theory terms.

5.4 Modal generalising

One more observation was made with respect to boulomaic worlds in this data set. This is exemplified by two more indicative examples:

• While the other three are there to enjoy the scenery, even hoping to shoot some footage for an arthouse film they are making, our protagonist is intent on finding her father who has recently vanished. (GRS-59)

• The main character, her boyfriend, and a dysfunctional married couple travel to her childhood home on an island in Quebec to attempt to locate her missing father. (GRS-47)

Both passages invite the reader to create one or more boulomaic modal-worlds, either through the use of verba sentiendi (‘hoping’), or through the use of an infinitive form (‘to enjoy’, ‘to attempt to locate’) to indicate an unrealised state of affairs a character wishes to bring about. However, GRS-47 prompts just one world, which is offered as a representation of the joint desires of four separate characters. By contrast, GRS-59 triggers two different boulomaic modal-worlds to represent the fact that the narrator’s reasons for returning home are not the same as those of her friends; a conceptual structure which more closely matches that prompted by Atwood’s text, in which David gives a clear indication of his reasons for making the trip:

“I’d like to go down to the lake for a couple of days,” I say to David, because it’s his car. “I’d like to look around, if that’s okay.”

“Great,” says David. “I’m going to get me one of them smart fish.” He brought along a borrowed fishing rod… (Atwood, 1979: 23)

Here, David’s declaration that ‘I’m going to get me one of them smart fish’ uses the verb ‘going’ with the infinitive ‘to get’ to cue the creation of a boulomaic modal-world, representing an as-yet-unrealised state of affairs which David desires to bring about, and which may or may not be realised. This provides a contrast with the narrator’s reasons for returning home. Further evidence that David and the narrator’s other companions do not share the narrator’s motivations is provided elsewhere:

They’re doing me a favour, which they disguised by saying it would be fun, they like to travel. But my reason for being here embarrasses them, they don’t understand it. (Atwood, 1979: 10)

I noted above how the omission of modal-worlds from reader precis could not be taken as evidence that reviewers had forgotten their existence. However, the language of review GRS-47, which does not cue discrete boulomaic modal-worlds to encapsulate the conflicting desires of the non-narrating characters, but instead implies that all four characters shared a common motivation, provides a direct contradiction of the original text rather than a simple omission. In all, five of the 24 reviews whose language triggered a boulomaic modal-world describing the narrator’s desire to find her father attributed this motivation to the entire party rather than to the narrator alone. This behaviour could be seen as a form of ontological consolidation distinct from the tendency to focus only on states of affairs in the world from which the narrator narrates.

Intriguingly, this finding appears to mirror behaviour in two studies reported by Norledge (2021, 2022). In the first, she documents reading group responses to Paolo Bacigalupi’s ‘Pop Squad’, describing one reader attributing a narrator’s mental states to other characters, and in doing so ‘position [ing] the narrator as an extension of collective society’ (2021: 55). In the second study she reviews critical responses to Ray Bradbury’s Fahrenheit 451, and suggests that readers judge the protagonist’s attitudes to arson to be characteristic of those of the society he inhabits (2022: 70). Thus, this pattern of recasting the beliefs of a single character as more widely applicable may be a consolidation process that occurs commonly during or after reading.

5.5 Application of schematic knowledge

Previous studies of narrative recollection (such as Bartlett, 1932) have emphasised the role of schematic knowledge in shaping reader summaries. Goodreads precis frequently reported events from different parts of Surfacing. In Atwood’s text, these events were presented as function-advancing propositions in different text-worlds, often separated by multiple world-switches, and will therefore have been conceptualised by readers of the novel as distinct mental representations. Precis writers frequently reported such discrete events using language that prompted the creation of a single text-world. Often such summaries involved the application of schema knowledge. An example of this can be seen in the following precis, quoted in its entirety:

An unnamed 20-something narrator travels with her lover and another couple back to her childhood home, an isolated cabin on an island in Quebec's bushland, because her father has gone missing. While ostensibly searching for him, they spend their ten days alone on the island doing standard camping-trip activities like fishing, canoeing, and picking wild blueberries. (GRS-34)

This precis triggers a single text-world in which the enactors (the narrator, plus ‘her lover and another couple’) are described using function-advancing propositions which correspond to actions narrated at different points in the original text. The party arrive at the cabin on page 27, go fishing from pages 57–60, and again from pages 112–115, pick blueberries from pages 78–80, and take several trips by canoe. These events represent only a fraction of those recounted by the narrator, and this reviewer makes no reference to function-advancing propositions included by other reviewers; for example ‘The couple and the boyfriend are filming a documentary’ (GRS-12), or ‘David is always making demeaning sexualised jokes’ (GRS-28). Instead, the reviewer appears to have reported these particular events because they conform to a schema of ‘standard camping-trip activities’ which has shaped their summary of the text.

Precis GRS-34 was notable for singling out several individual events, each one corresponding to an identifiable scene or scenes in the original text. By contrast, many precis contained function-advancing propositions which are likely to represent a cumulative understanding of multiple events narrated in Atwood’s text, but did so without listing the individual events themselves. Again, such descriptions emphasised particular kinds of happenings over others, and often revealed the role of schema knowledge in reader interpretations. Reviewer GRS-23’s description of the narrator ‘[g]radually letting herself go back to nature’ represents an alternative construal of the ‘standard camping trip activities’ listed in GRS-34, but conceivably also draws on the narrator’s description of removing her clothes (171), and deciding to eat only raw food and sleep out of doors (172). Thus, different reviewers selected from - and hence interpreted - the events in the text in different ways. In each case, reviewers appear to have amalgamated multiple function-advancing propositions which they will have originally experienced in discrete text-world mental representations when reading Atwood’s novel, and presented them in their precis using language that results in a single function-advancing proposition presented within a single text-world.

Examples such as these hint at an interesting process that appears to be taking place during narrative comprehension. Toolan, in attempting to define the typical characteristics of a narrative, has emphasised the importance of connected events, arguing that narratives do not simply represent a series of random occurrences but instead present ‘consequential’ rather than ‘sequential’ descriptions (2013: 6). The reframing of multiple function-advancing propositions as single events appears to have implications for how readers construe processes of cause and effect in narrative summaries, as the following example demonstrates:

the past haunted her and turned her into a lunatic (GRS-71)

This summary appears to contain two schematic reinterpretations. Firstly, events presented in Surfacing as multiple acts of memory, each conceived through a discrete world-switch, have been reframed as a single function-advancing proposition; ‘the past haunted her’. Secondly, as described above, events presented in multiple text-worlds have been reframed using the single function-advancing proposition ‘turned her into a lunatic’. Crucially, summary GRS-71 presents ‘the past’ as the actor in a material intention process (Halliday, 1985) responsible for the narrator’s change in state. This reading, however, is not inevitable if one considers the configuration of individual text-worlds prompted by Surfacing. One of the final text-worlds the reader is invited to construct is one which contains an enactor of the narrator and an unnamed figure feeding birds in the garden: an activity previously associated with the narrator’s dead mother. This ghost-like presence, which subsequently vanishes, is described as having hair that is ‘long, down to her shoulders, in the style of 30 years before’ (176). This incident, which could fairly be described as an example of the narrator being ‘haunted’ by the past, is experienced after those described above, which I have previously argued invite readers to invoke a ‘madness’ schema, and could conceivably be read as a product of, rather than as a cause of, the narrator’s instability. However, such a reading is not captured in the summary quoted above, which implies an opposite direction of causation.

6. Towards an extended Text World Theory

Having outlined examples of particular text-world consolidation processes involved in constructing reader summaries of Surfacing, I now offer a brief discussion of possible implications for an extended Text World Theory.

If Text World Theory is to account for readers’ long-term recollections of narrative texts, I suggest it needs to adopt some way of describing readers’ evolving mental representation of what I will henceforth refer to as the most ontologically prior world projected by the text. (This terminology is consistent with Galbraith’s (1995) account of POPs and PUSHes between different ontological levels of a narrative, and I provide a more detailed justification for the use of the term in Harvey, 2023: 63–67). This most ontologically prior world corresponds to the world in which the narrator resides in a homodiegetic narrative, and may informally be thought of as the reality of the fictional world projected by the text. It has traditionally been referred to as an ‘empty’ text-world in accounts of homodiegetic narration. However, in this paper I have suggested that readers do seem to accept at least a subset of claims made by a homodiegetic narrator as fair descriptions of states of affairs in the world in which the narrator resides, even when the narrator is judged to be unreliable. In other words, one common consolidation practice which may occur as text-world mental representations are stored in and retrieved from the long-term memory is that information (world-building elements and function-advancing propositions) initially conceptualised in discrete modal-worlds is ‘mapped’ into a readers’ mental representation of this most ontologically prior world. This ontological simplification may occur as part of a process of combining different ‘working models’ of a narrative situation into a single ‘integrated model’, described by Gibbons (2023) as she attempts to reconcile Text World Theory with Zwaan and Radvansky’s (1998) situation model account of discourse comprehension.

The examples I have discussed above have focussed on readers mapping information originally conceptualised in modal-worlds triggered by the narrator’s act of narration. However, the Surfacing sample also contained a notable example of a reviewer reproducing a belief of a non-narrating character as part of their representation of the fictional world. Towards the end of Surfacing, the narrator’s friend David reports his understanding of what happened to the narrator’s father:

‘Then I saw David hurrying, taking the hill steps two at a time. The screen door banged shut behind him. “They found your father,” he said […]

“Oh,” I said. “Where?”

“Some American guys found him in the lake. They were fishing, they hooked him by mistake'

(Atwood, 1979: 151)

In this passage, David’s act of direct speech prompts the creation of a new text-world (Gavins, 2007: 50); a world which has a different ontological status to that of the narrator’s own testimony. Indeed, the narrator herself subsequently refutes David’s claim by informing the reader ‘I knew it was a lie’ (151). However, reviewer GRS-44 presented David’s account as an unqualified statement of fact in their summary of the text, (a description which tallied with my own reading of the text):

Basics seem to include the physical truth that her father who lived there is missing. This does seem to be real since some American fishermen find his body. (GRS-44)

In other words, the mapping of information from epistemic modal-worlds of narration into the initial text world may be a specific example or a more general process of mapping information between ontological worlds at different levels subsequent to reading. Currently, Text World Theory offers no account of such processes. However, Fauconnier’s mental spaces model (1985, 1997) does describe such processes occurring in discourses which lead to the creation of more than one mental space.

Fauconnier defines mental spaces as mental constructs with structure comprising a series of elements and the relationships between them (1985: 16). He then suggests that this structure is transferred by default from ‘parent’ spaces to ontologically subsidiary spaces via a process he calls ‘optimisation’ (1997: 43), so long as it does not contradict explicit structure in the subsidiary space. He also suggests that analogous mappings occur in the reverse direction when presuppositions of linguistic statements defining subsidiary spaces ‘float’ back into parent worlds (61). I argue that Text World Theory might usefully adopt analogues of these processes to account for ways in which world-building elements and function-advancing propositions are mapped between worlds of different ontological status subsequent to their initial conceptualisation. In this way, a growing mental representation of the initial text-world projected by the text can be created and maintained alongside the individual text-world and modal-world representations that are created in the moment of reading.

A rigorous test of this hypothesis using Goodreads data is hard, as it is not practical to map the conceptual structure of an entire novel to which readers are responding, but could be attempted using other methodologies. A study of responses to a shorter text, gathered under more controlled conditions, which provides additional evidence to support this hypothesis is reported in Harvey (2023).

As a result of these analyses, I therefore offer three hypotheses that may lead towards an extended Text World Theory. Firstly, I suggest that text-world mental representations should be regarded as constructs in the working memory. However, since working memory is finite in both capacity and duration, some or all of the information initially conceptualised using text-world mental representations must subsequently by transferred to the long-term memory. My second hypothesis is that text-world conceptual structures are often consolidated when information is stored in and retrieved from the long-term memory. Thus, whereas some long-term memories of fictional narratives may retain information about how different states of affairs were originally conceptualised at discrete ontological levels, other long-term memories may not retain these ontological distinctions. My third hypothesis is that one result of such consolidation processes in the comprehension of fictional narratives is an evolving representation of states of affairs in the most ontologically prior world projected by a fictional text. This evolving representation in the long-term memory exists as an entity discrete from the text-worlds created in the moment of textual comprehension. Parallels between this evolving representation and the integrated models of Zwaan and Radvansky’s (1998) account of situation models would bear further exploration. The formation of such a representation is foregrounded in reader summaries of unreliable homodiegetic narratives such as Surfacing, where readers accept some but not all of a narrator’s testimony. However, given that readers frequently learn about fictional worlds through the beliefs and testimonies of non-narrating characters as well as from a narrator, such representations may result from the comprehension of fictional narratives of all kinds.

Alongside these hypotheses, however, I suggest that is it important not to discount the possibility that information from a single narrative may be stored in more than one format in the long-term memory. In other words, while readers may build a working model of states of affairs in the most ontologically prior fictional world, they may also retain information about how these states of affairs were originally conceptualised while reading the original text. A full text-world account of narrative memory would therefore distinguish between discrete processes of consolidation. Some processes may occur as information in discrete text-worlds is transferred into long-term memory. Other processes may occur as a subset information in the long-term memory is retrieved when readers produce new discourses about fictional worlds with their own text-world conceptual structures. Further work is needed before more precise claims can be made about how text-world mental representations are configured in the long-term memory.