Trust in Stories: A Reader Response Study of (Un)Reliability in Akutagawa’s “In a Grove”

Abstract

For this article, we reviewed and synthesized narratological theories on reliability and unreliability and used them as the basis for an exploratory study, examining how real readers respond to a literary short story that contains several unreliable or conflicting narrative accounts. The story we selected is “In a Grove” by Ryūnosuke Akutagawa (orig. 藪の中/Yabu no naka) from 1922 in the English translation by Jay Rubin from 2007. To investigate how readers evaluate trustworthiness in narrative contexts, we combined quantitative and qualitative methods. We analyzed correlations between reading habits (i.e., Author Recognition Test), cognitive traits (e.g., Need for Cognition; Epistemic Trust), and trust attributions to characters while also examining how narrative sequencing and character-specific reasons for (dis)trust shaped participants’ judgments. This mixed-methods approach allows us to situate narrative trust as a context-sensitive, interpretive process rather than a stable individual disposition.

1. Introduction

When we are reading or viewing stories, how do we know what sources, characters, and voices to trust? How do we know when we should be vigilant towards the possibility of being misled? How do we determine when to read, listen, or view with, and when against the grain? Stories are everywhere: not just in books of fiction, popular series on streaming platforms like Netflix, and newer forms of writing like fan fiction and Instagram captions but also in politics and advertisements. Politicians ‘spin’ their stories to appeal to the public. On social media, people tell stories about their lives, and in the literature, popular genres like autofiction and fan fiction blur the boundaries between literary fiction and memoir (Browse et al. 2019). Stories are important tools for making sense of our personal lives and the world we live in. Yet, because of this proliferation of often contradicting narratives across different media, deciding what sources and voices to trust and pay attention to becomes an increasingly pressing matter. Especially since the last decades, we have seen a decrease in trust when it comes to former sources of authority such as mainstream journalistic media, scientists, and experts (Oreskes and Conway 2010). The question of truth seems to increasingly be replaced by the question ‘who tells the most compelling story?’

Narrative fiction trains our capacity for the attribution of meaning and value and for modularization: our capacity to discern nuances, degrees, and shades of verisimilitude and fictionality in information. Doubts concerning the reliability of narrative communications, or concerning the sincerity and competence of a narrator, are effective for triggering such reflections (Korthals Altes 2015). Narratology, the study of narrative fiction across different media, offers an elaborate toolkit for analyzing discordant or conflicting narrative voices, unreliable narrators, ambiguity and irony, and ‘conspiratorial’ storylines. The attribution of trust is part of the dynamics of communication or of the distribution of information throughout a story.

In fiction, a narrator is generally considered unreliable when they deviate from the norms posed by the text or held by the author (Booth 1961) or the reader (A. Nünning 1999). The unreliability of a narrator can be detected based on textual (grammatical, stylistic, or historical mistakes; internal discrepancies) and paratextual elements. The reliability of a narrator can remain ambiguous (think of the governess in Henry James’ The Turn of the Screw) or a narrator can go from reliable to unreliable and vice versa. Narrators further differ in terms of intentionality: some set out to deceive and manipulate; others aim to tell the truth, yet are deluded or misinformed themselves (for instance child narrators or narrators who are non-neurotypical, like in Mark Haddon’s The Curious Incident of the Dog in the Night-Time). Of course, no narrator can be fully reliable: any act of communication attributes meaning to events, selection, perspectivism, moral positioning, and genre conventions. We rather speak of scales of reliability or a spectrum between the two extremes (Olson 2003).

Theories of narrative (un)reliability enable a detailed analysis of trustworthiness, encompassing various types, degrees, motivations, and the effects these have on readers. What remains to be researched is how actual readers/viewers respond to unreliability and reliability in narrative fiction. More precisely, how do readers of narrative fiction calibrate between trusting and distrusting attitudes in regard to narrative, and how do these affect their reading/viewing strategies and interpretations? This knowledge could not only be valuable to increase our understanding of how trust in a narrator is established but also to better comprehend how people engage with narrative outside of fiction. Here, we can think of fake news websites or podcasts that spread misinformation.

For this article, we reviewed and synthesized narratological theories on reliability and unreliability and used them as the basis for an exploratory study, examining how real readers respond to a literary short story that contains several unreliable or conflicting narrative accounts. The story we selected is “In a Grove” by Ryūnosuke Akutagawa (orig. 藪の中/Yabu no naka) from 1922 in the English translation by Jay Rubin from 2007. This study is explorative in nature, meaning we did not depart from a strict hypothesis. To investigate how readers evaluate trustworthiness in narrative contexts, we combined quantitative and qualitative methods. We analyzed correlations between reading habits (i.e., Author Recognition Test), cognitive traits (e.g., Need for Cognition; Epistemic Trust), and trust attributions to characters, while also examining how narrative sequencing and character-specific reasons for (dis)trust shaped participants’ judgments. This mixed-methods approach allows us to situate narrative trust as a context-sensitive, interpretive process rather than a stable individual disposition.

We argue that trust in literary narrative is not reducible to textual or paratextual cues alone, but emerges from an interpretive interplay between narrative sequencing, character positioning, and readers’ cognitive, cultural, and ethical frameworks. Drawing on empirical reader response data, we show that readers often confirm, but also complicate or subvert narratological expectations about unreliability. Our findings suggest that narrative trust is not a fixed or purely text-internal property, but a dynamic, situated process, formed by the interaction between textual form and reader disposition. In doing so, this study contributes to a more nuanced understanding of how narrative reliability functions in actual reading contexts and proposes a more context-sensitive, empirical addition to prevailing narratological models. The results will therefore contribute to the integration of cognitive sciences and literary studies in its examination of how formal, narratological aspects may influence attention, comprehension, and interpretation.

2. Narratological Approaches to Unreliable Narration

In narrative theory, trust is defined as perceived reliability (Van Lissa et al. 2016). It is informed by an interplay between textual, paratextual, and extratextual factors, and characteristics of the reader. Trust emerges during reading, but also informs our reading: it is both precondition and result of reliability in narrative, suggesting a dynamic process or feedback loop (Korthals Altes 2014). Narratologists have developed myriad theoretical models and analytical frameworks for the analysis of reliability in narrative fiction.1 These have often been divided into three main categories or schools. The rhetorical school (Booth 1961; Phelan 2017) holds that an unreliable narrator expresses values and perceptions that do not align with those of the implied author. Text-centered approaches (A. Nünning 1999) focus on detecting unreliability through textual cues or signals, such as contradictions, narrative gaps, signs of biased or limited perception, confused language, and lapses. Third, the cognitive school shifts the focus from text to reader. Cognitive narratologists, such as Tamar Yacobi (2005), treat the attribution of unreliability primarily as a reading and interpretation strategy, rather than an immanent textual phenomenon. In this view, the reader comes to the text with a variety of mental and conceptual frameworks, shaped by their knowledge of the real world, previous experiences with literature and other media, as well as their values, cultural background, and social norms. This can lead to varied readings of a narrative’s reliability, where a single passage might be perceived as unreliable by one reader and completely credible by another, depending on how each applies their unique set of interpretive lenses. Such an approach highlights how the cognitive process of reading involves not only deciphering the words on the page but also integrating personal cognitive schemas that may lead to conflicting judgments about the truthfulness of the narrative.

In contrast to the often binary thinking in fact-checking approaches to fake news that tend to focus on ‘true versus false,’ narratologists offer a nuanced perspective by distinguishing between different types and degrees of unreliability. James Phelan (2017) developed a typology in which six types of unreliable narrators are categorized along three communication axes: facts and events (mis- or underreporting); understanding (mis- or underinterpreting); and values (mis- or underevaluating). Every unreliable narrator can be located somewhere along these axes.

3. Integrating Narratological Approaches

Integrating these approaches would allow us to understand unreliable narration at the intersections of textual characteristics and cognitive processes, or: bottom-up and top-down processes (Olson 2003). Bottom-up processes start with the text and its clues, while top-down processes occur when readers bring their own conceptual frameworks to the text to determine its reliability. We might then construct the following schematic overview of the reading process (Figure 1).

We start reading this figure at the lemma ‘text,’ but this is already an artificial move. As discussed, trust and reliability in narrative function more accurately as a feedback loop. Long before encountering the text itself, readers often form expectations about an author’s trustworthiness based on paratexts, media presence, or prior knowledge of their work. Genre conventions, such as those in thrillers or mystery fiction, can similarly prime assumptions about (un)reliability. When reading fiction, we typically treat its propositions as true within the diegesis, despite knowing they are fictional. To question each claim would be cognitively inefficient. As Lisa Zunshine puts it, “once we have bracketed off the fictional story as a whole as a metarepresentation with a source tag pointing to its author, we proceed to consider its constituent parts as more or less architecturally true” (Zunshine 2006, p. 147).

Readers become alert when a text presents cues such as inconsistencies, ellipses, or tonal shifts. These trigger what Dan Sperber (Sperber et al. 2010) calls epistemic vigilance: cognitive mechanisms that help us assess the trustworthiness of a communicator by calibrating trust and suspicion. In response, the reader attempts to restore coherence using prior knowledge and narrative experience. If the incongruities persist to the point where a coherent mental model of the story world can no longer be sustained, the reader may interpret the narrator as unreliable—an interpretive hypothesis that, following A. Nünning (1999), can be seen as an integrative hermeneutic device.

Next, the reader assesses the degree of (un)reliability. For skilled readers, this goes beyond a binary of true versus false. No narrator is fully reliable: all narration involves selection, perspective, moral framing, and genre-based constraints. Reliability is better understood as a spectrum (Olson 2003). Readers also consider possible motivations—ranging from deliberate deception, as with Alex in A Clockwork Orange, to honest misperception, as in Huckleberry Finn. At this stage, readers may implicitly position the narrator along the communicative axes proposed by Phelan (2017): facts, interpretation, and evaluation. These judgments often lead to new reading strategies, such as filling narrative gaps or reading against the grain.

In sum, narratology provides tools to analyze narrative unreliability and to distinguish its types and degrees. As we have seen, (un)reliability is largely reader-dependent, leading to wide variation in interpretation. These judgments are shaped not only by textual and medial cues but also by readers’ own beliefs and experiences. To fully understand how unreliability is detected and interpreted, we must therefore attend to actual reader response.

Traditional narratological theories of unreliability often rely on an idealized reader, typically reflecting the scholar’s own expert interpretation, rather than accounting for real reader responses (Alber 2018). This creates the misleading impression of a single ‘correct’ reading. In practice, readers interpret unreliability in diverse and often conflicting ways. As Alison Gibbons observes, rhetorical narratology tends to rely on introspective or hypothetical responses—“stand-ins” for real readers—and privileges authorial intention over actual reception (Gibbons 2022, pp. 313–14). Empirical narratology is beginning to address this gap (Wirag 2020; Alber and Strasen 2020). Innovative approaches like Alice Bell and Astrid Ensslin’s (Bell and Ensslin 2024) medium-conscious reader response methodology provide valuable tools for studying complex narrative phenomena qualitatively and empirically. Reception research can deepen our understanding of narrative experience and also challenge or refine theoretical models (Gibbons 2022, p. 314). While empirical studies of trust and reliability are growing, systematic reader research on unreliability remains scarce.2 To understand how readers perceive and interpret (un)reliability, we must attend to interpretive diversity. How do readers decide which voices to trust, and how do these decisions shape their reading strategies, attentional focus, and interpretations?

4. Reading Material

Ryūnosuke Akutagawa’s modernist short story “In a Grove” (Akutagawa [1922] 2007) was selected for its structured presentation of conflicting testimonies about a single event. The story gained widespread recognition through Akira Kurosawa’s film Rashomon, which combined In a Grove with Akutagawa’s earlier story “Rashōmon”. The film popularized the term ‘Rashomon effect,’ describing the phenomenon where multiple witnesses offer contradictory accounts of the same event, revealing the subjectivity of memory and perception. It highlights the narrative instability of truth, especially in the absence of a verifiable version—famously exemplified by debates surrounding the assassination of John F. Kennedy.

The story is told through a series of seven conflicting firs-person testimonies to a police commissioner, each shedding contradictory light on the murder of a samurai named Takehiko and the possible assault of his wife Masago. His body was found in bamboo forest near Kyoto. His wife disappeared; a Robber called Tajomaru was arrested. First, we read the testimony of a woodcutter who claims to have found the body in a bamboo grove. He says the victim was stabbed in the chest, with no weapon nearby, only a rope and a comb. Second, a Buddhist priest recalls having seen the samurai and his wife on the road the day before. She was veiled, on horseback, and he was armed with a sword and arrows. Third, a policeman reports arresting the notorious bandit Tajōmaru, who was found injured and in possession of the dead man’s horse, bow, and arrows. He implicates Tajōmaru as the suspected killer and warns of his past crimes. Fourth, the samurai’s mother-in-law confirms the victim’s identity and expresses concern for her missing daughter, Masago, whom she describes as spirited and loyal. Fifth, Tajōmaru’s confession follows. He admits to luring the couple into the grove by pretending to sell buried treasures. He says he tied up the husband and then raped Masago but claims she demanded one of them die to avoid the shame of being defiled before two men. Tajōmaru says he fought the samurai in a fair duel and killed him. He claims Masago fled afterward, and he does not know where she went. Sixth, we have Masago’s testimony, from when she was later found. She contradicts Tajōmaru, saying he indeed raped her but then left. When she freed her husband, he looked at her with silent contempt. Shamed by this, she tried to kill herself with her dagger but failed. She later woke up to find her husband dead and fled in terror. Seventh and last, a medium channels the spirit of the dead samurai himself. According to this version, after the rape, Masago urged the two men to fight. Tajōmaru, ashamed, freed the samurai and left. Masago fled as well. The samurai claims he then killed himself with his wife’s dagger in despair.

5. Approach

We asked 154 participants from our faculty’s student pool to read this short story,3 after which we asked them to fill out a questionnaire consisting of a number of open and multiple choice questions about the story, a self-report on personality traits, and reading habits and familiarity with literature (see below). Participants were mostly undergraduate students from the BA or premaster programs Communication and Information Sciences (62) and Cognitive Science and Artificial Intelligence (65), with some outliers from Sociology, Literary Studies, Bsc Global Management of Social Issues, and MA New Media Design.

We checked for potential priming effects—whether earlier fragments might activate interpretive schemas that shape subsequent comprehension. In the context of reading and narrative interpretation, priming suggests that an earlier fragment can activate certain cognitive or interpretive frameworks—such as expectations about characters, genre conventions, or emotional tones—which then shape how a reader interprets what comes next. This effect has been studied in psycholinguistics and literary cognition. For example, Gerrig (1993) explores how narratives can prime readers’ interpretive stances or activate schemas. Similarly, Zwaan and Radvansky (1998) discuss how situation models constructed from earlier narrative content affect comprehension and memory of later parts. Another possibility we considered was a primacy effect, in which information presented earlier in the sequence might disproportionately influence judgments of trust or distrust, consistent with serial-position effects observed in impression formation and narrative persuasion (Nahari and Ben-Shakhar 2013).

In this case, it meant that the order of the narrative fragments would potentially determine which ones were interpreted as more and less reliable. We wanted to see whether the narratives our participants trusted and distrusted, were impacted by the order of the story. We needed to make sure that problems with interpretation stem from unreliable narration and conflicting information and not from factors like word frequency, difficult sentences, or unknown terms, which can be filtered out by mixing up the order. Based on the integrated frame of theories on unreliability above, a reader might be expected, when first confronted with the account of the bandit Tomoko, to accept it as true, and judge the contradictory account by Yuki, encountered later, as unreliable because of that contrast. Presenting readers with different narrative fragments in sequence could entail a form of conceptual or interpretive priming: prior exposure to a narrative element (e.g., a morally ambiguous action or a revealing inner monologue) could make readers more likely to interpret subsequent ambiguous fragments in a particular light (e.g., as trustworthy or untrustworthy).

To control for a priming effect, we made six different versions of the story (1–6) by placing the story of the three main suspect characters in different orders: the robber Tajomaru (“Tomoko”), woman Masago (“Yuki”), murdered samurai Kanazawa no Takehiro (“Hiro”).4 Since the accounts of the woodcutter, traveling Buddhist priest, policeman, and old woman only contain smaller inconsistencies, and since they were not present at the murder, we decided that the differences between these three characters’ account would be the most salient and impactful for readers’ construction of (un)reliability. We randomly assigned each participant to one of the six conditions. Breakdown per condition was 1:28, 2:23, 3:23, 4:22, 5:22, 6:22.

We presented the story in the form of a web page and asked participants to read it in one sitting while scrolling. Because the Japanese names could potentially be confusing for Western readers and impede reading, we decided to use simpler names with different first letter to avoid confusion, while retaining the Japanese sound for realism. The robber Tajomaru became “Tomoko”, the wife, Masago, became “Yuki,” and the murdered samurai Kanazawa no Takehiro became “Hiro.” After reading, we asked the participants to fill out a questionnaire with questions about the text and about their personality.5

The Questionnaire

We collected participants’ age and gender. We asked whether they read the story before or had seen the film adaptations from 1950 or 2011. Three participants had prior familiarity with the written story; two had seen one of the adaptations. We asked three multiple choice control questions to test basic comprehension (Who was the victim?; What did the bandit steal?; and ‘Did the woman and the bandit leave together’). The answer to the last question is actually ‘it is impossible to know,’ which was one of the options. We excluded participants who answered all three comprehension questions wrongly, as well as the ones that were otherwise incomplete, which left us with 148 remaining participants (F: 80; M: 66; Other: 2). We also included three multiple choice questions about details, to see if they had read attentively (what was the victim wearing; what was the woman wearing; and was there a comb at the crime scene?). We asked them to rank the seven character narrators in terms of whom they found the most to least trustworthy, and for the characters they put on top (trust the most) and bottom (trust the least), to briefly explain their rating in writing.

Reading frequency was assessed with a self-report item: “How often do you read literary fiction (such as novels and short stories)?” Responses were given on a ten-point scale, with options ranging from never, a couple of days a year, a couple of days a month, to 1–7 days a week (scale from De Mulder et al. 2021).

Print exposure was measured using an adapted version of the Author Recognition Test (ART; Stanovich and West 1989). The ART is widely regarded as a reliable proxy for lifetime print exposure to fiction, outperforming other measures such as self-reports or book counts (Wimmer and Ferguson 2022). In this task, participants are shown a list of names that includes both actual authors and invented foils. They are instructed to mark only those they are certain are real authors. To discourage guessing, each false name selected is subtracted from the total score.

The English-language survey ART contained 42 names: 31 author names and 11 foils. The ART was scored by adding up the number of correctly identified names and subtracting the number of chosen foils. The range of the ART was 20 (M = 4.35, SD = 3.79. Distribution is positively skewed (skewness = 1.364, SE = 0.20), indicating more lower scores and a long right tail. We also asked used a Shortened Need for Cognition Scale–6 (NCS-6) (Lins de Holanda Coelho et al. 2020) and the Epistemic Trust, Mistrust and Credulity Questionnaire (Campbell et al. 2021) with three subscales: trust (5 items), mistrust (4 items), and credulity (5 items). We chose these scales because previous research relates them to skills of detecting and interpreting misinformation (Tanzer et al. 2021; Leding and Antonio 2019).

6. Qualitative Analysis and Discussion

To conduct a thematic analysis of the open questions, we used the open source qualitative research software Taguette (Rampin and Rampin 2021). Data were coded on the go through a bottom-up approach. The prompts (based on the participants’ ranking of characters from 1 to 7 in terms of trustworthiness) were “Why did you trust the person you put on top of the list the most? Please specify your reason(s)” and “Why did you trust the person you put on the bottom of the list the least? Please specify your reason(s).” This resulted in 17 labels and 35 Sublabels (see Appendix A for an overview). We will now discuss our main findings.

6.1. Dis/Trust Based on Role or Profession

Our data show that, for some participants (N = 20), trust was grounded in an a priori belief linked to a character’s social role or identity, particularly the victim role. Participants expressed a presumption that victims should be trusted based on their experiential position, e.g.,: “whenever everyone is giving highly contradictory statements, it makes sense to me to believe the victim” and “I think Hiro will be honest because he is the one who got murdered.” This reflects an ethical or moral stance towards testimony associated with classical notions of epistemic trust, where vulnerability or victimhood can confer presumed credibility. This presumption may also reflect broader cultural scripts around victimhood and the moral imperative to believe survivors, formed by recent public discourses around testimony and justice. Such a stance exemplifies Fludernik’s (1996) concept of the “natural attitude” towards narrative truth, whereby readers adopt a default trust in first-person experiential accounts. This readerly disposition is remarkable given that the story they are responding to is an unnatural narrative: it resists synthesis, frustrates conventional truth-seeking, and features a dead man who speaks through a medium. In rarer cases, a similar logic was applied to argue for Yuki’s trustworthiness: “I trust the account of Yuki (the woman at the temple) because of her experience. She went through a lot ….”. The underlying idea here is that victims should be believed by virtue of their experience, which can be considered an ethical reading stance. Yet the belief also seems selectively applied here, choosing some victim accounts over others, which suggests that this ethical disposition might be moderated by gendered stereotypes or narrative salience.

In a related vein, a small number of participants (N = 5) trusted maternal figures by default, by virtue of their perceived altruistic motivations, e.g., “a mother always wants what’s best for her daughter” and “she is a mother figure, she seems to be trustworthy.” This means that she would have no incentive to lie (“Her only concern was that her daughter was found!”). Even though this was a relatively small group of comments, they were so specific that we decided to make a separate label (‘Believe mothers’). Such responses draw on cultural narratives about motherhood as selfless and virtuous, invoking a stereotypical association between the maternal and trustworthiness. This an example of how readers patch interpretive gaps by importing familiar schemas to sustain coherence (Yacobi 1981). The presumed purity of the mother’s motivation is a reflection of genre tropes and broader cultural beliefs about family roles, which are culturally contingent.

Conversely, the mediation of a character’s account through a supernatural figure (the medium relating Hiro’s story) generated distrust for some readers (N = 13). While the murdered Samurai Hiro was first on the list of most-trusted characters, he was therefore also mentioned as the least trusted character by those readers: “I don’t trust the murdered man because his story was explained through a medium and i can not trust for sure this source as trustworthy”; “his story isn’t told by himself but by a spirital [sic] person and I don’t really trust that.” For those reading the story in a realistic vein, the use of a medium to represent his story was therefore automatically discredited as unbelievable (“The murdered man was inwoked [sic] through a medium, this is not a reliable source. This is just what the medium thinks that happened, not what really happened. Its an imagination from someone who wasnt [sic] there”) or confusing (“he is dead? I’m a bit confused as to how we even got his side of the story.”). This suggests that the narrative frame and its genre conventions significantly influence trust calibration. For readers adhering to a realist interpretive stance, supernatural mediation may signal unreliability or fictionalized “imagination,” highlighting the impact of interpretive frames on narrative trust. For readers adhering to a realist interpretive stance, supernatural mediation may point to unreliability or fictionalized ‘imagination,’ highlighting the impact of interpretive frames on narrative trust. However, such distrust also invites the kind of patching that Yacobi (1981) describes, whereby readers attempt to resolve discrepancies or reconcile improbable elements in order to restore coherence and maintain epistemic trust.

Professional roles were also frequently invoked as proxies for trustworthiness. The policeman (N = 19), samurai (N = 3), and Buddhist priest (N = 13) were all often trusted due to associated social values such as law enforcement, honor codes, and spiritual integrity, respectively. This mirrors findings in narratology that readers often use real-world schemata and cultural ideologies in their trust evaluations (Caracciolo 2020). The policeman, for instance, is associated with objectivity and social authority, which were frequently cited as reasons to trust (e.g., “he enforces the law and should be the most trustworthy person on the list”; “the policeman has always an objective view on the situation, about the persons and is not influenced by the minds/ideas/principles of the characters”; and “we are taught to trust police since we are little kids”). Similarly, the samurai’s code of honor and the priest’s religious commitment are seen as guarantors of honesty: “what I know from swordsman in those stories they tend to have a very high respect for pride and honesty”; “the victim is a samurai which seems like a title of honor and trust.” Third and last, the Buddhist Priest was often considered trustworthy by default as he is “sworn to certain beliefs” (N = 13): “a Buddhist priest wouldn’t lie”; “A priest is least likely to be biased”; “A buddhist priest is a [sic] honest human being that is my biased [sic] of religion related figures”. This category in particular shows us how readers’ societal and spiritual beliefs play an important role in their assessments of narrators’ reliability, leading to divergent interpretations.

On the flip side, distrust was sometimes similarly attributed to social roles or professions. Policemen were automatically distrusted by some participants (“I just do not trust police”), and the bandit Tomoko was widely distrusted (N = 21) simply because of his criminal label which was interpreted as an obvious sign of his untrustworthiness, e.g.,: “He is the bandit. And people who do bad things are more likely to lie” and “The bandit is used to performing illegal actions. It is believed that lying is a normal trait of him.” This could be considered remarkable, considering the fact that Tomoko’s confession is self-incriminating and might otherwise generate trust. This persistence of distrust in the face of apparent sincerity indicates that cultural labels like ‘bandit’ serve as strong genre cues that prevail over textual nuance. These responses are a reflection of culturally embedded moral hierarchies and binary ideas of good and evil which are, reinforced by popular culture and media.

We see that in this category of reader response, which centralizes narrative roles and professions of characters as trust indicators, there is no emphasis on the implied author or textual elements. Rather, these responses are understandable through cognitive narratology as they are based on readers’ mental and conceptual frameworks shaped by real-world knowledge, previous reading experiences values, cultural background, and social norms.

6.2. Characters: Motivational Cues, Behavior, Personality Traits, Personal Stakes

In considering individual characteristics, our participants referenced factors such as age (N = 3), altruism versus egoism (N = 5), biased perspectives (N = 11), and intentions (N = 4). Here, one can think of statements like “Yuki, Hiros wife was selfish and wanted only herself to survive.” We conceptualize these as motivational cues rather than stable personality traits, highlighting the situational and inferred nature of these reliability assessments. This distinction corresponds with the notion that trustworthiness judgments often rest on mental state attributions rather than fixed traits (Zunshine 2006). Readerly inferences about motivation exemplify Korthals Altes’ (2014) notion of trust as both a precondition and an evolving product of narrative engagement. Rather than passive reception, readers actively calibrate their trust using epistemic vigilance, attending to discrepancies between a character’s expressed intentions and the situational context. This also shows how narrative interpretation draws on schema-based reasoning, consistent with Yacobi’s (2005) framework: readers apply extra-textual knowledge (e.g., cultural models of honor and shame) to explain otherwise ambiguous behavior.

In the category of characters’ behavior, behavioral consistency (N = 6) and immorality (N = 29) were important factors for (dis)trust. Yuki was often mentioned for perceived inconsistent behavior, for instance “not following suit on her promise to her husband,” namely: “promising that she would kill herself after, but she couldn’t”. Here, postulating that Yuki is an unreliable narrator functions as a naturalization strategy (Fludernik 1996), where readers try to integrate conflicting accounts into a coherent story-world. Immorality was most often attributed to the bandit: “it seems like he does not really have any morals so lying would come natural to him” and “Someone that rapes a women and kills her husband doesn’t especially qualify as trustworthy in my textbook.” This is consistent with traditional narrative reliability theory (V. Nünning 2015), which emphasizes coherence and moral evaluation.

These findings confirm Phelan’s notion that unreliability is not solely a matter of factual distortion, but also of value-based judgments, of scrutinizing a character’s ethical stance and reliability in evaluating the narrative world. The reader’s attribution of egoistic or altruistic motives, for instance, can be mapped onto Phelan’s value axis: when readers suspect biased or self-serving motivation, they infer under-evaluation or distorted moral framing. This reflects how individual character traits trigger epistemic vigilance (Sperber et al. 2010), here understood as cognitive monitoring processes that try to align a character’s intentions with the narrative’s implicit norms. Readers attempt to reconcile moral cues in the text with their own interpretive schemata and expectations, especially when character motivations appear unclear or contradictory (Yacobi 1981).

A number of participant mentioned personality traits that made the characters reliable or the reverse, but here, there does not seem to be a lot of agreement as to the precise labels, as most of these were only mentioned one to three times (e.g., empathetic, fair, (un)kind, plain, rational, spontaneous, stable). Honesty (N = 10) and impulsivity (N = 4) also played notable roles, illustrating how perceived character stability and ethical comportment shape trust judgments (e.g., Yuki was deemed “unpredictable, impulsive and untrustworthy,” whereas the Priest “seemed like an honest person with an eye for detail”). Psychopathology was mentioned three times. That readers diverge in labelling characters as ‘empathetic’ or ‘rational’ illustrates Olson’s (2003) spectrum of reliability and also underscores the role of subjective interpretation in narrative comprehension. As Yacobi (2005) argues, readers bring to bear diverse interpretive schemata, which shape not only how they assess the truth-value of a character’s account, but also how they assign personality traits.

A particularly salient theme was what we chose to label “personal stakes”. This category described whether a character-narrator had something to gain or lose by lying. Under this label, we see that distance from or lack of involvement (N = 19) was often reason for trust, reflecting the belief that detached witnesses are more credible (“His description was as if he was just a bystander, someone who does not really have a lot to do with this case”). Being present at the crime scene was another reason for trusting the objective eyewitness accounts of character-narrators (15). Reputation factors such as pride and honor (N = 5) were linked to personal stakes, showing that moral codes influence trust calibration: Hiro, for instance, allegedly suffered the humiliation of his wife’s infidelity, which might compromise his trustworthiness (“the incident was humiliating, the bandit wanted to take his wife, so for me that is a good reason to not tell the truth”); Yuki’s “honour would be tainted and she would be labeled as a murderer when the truth came out. Whereas Tomoko did not have any honour to lose anymore.”

Readers thus draw on culturally specific schemata related to shame, honor, and social standing when interpreting characters’ motivations to deceive. These intuitive mappings align with Phelan’s motivation axis, as readers evaluate not just what characters report, but the social imperatives that might influence their testimony. These responses seem intuitive and affectively charged, and they exemplify what Yacobi (2005) describes as reader-driven hypotheses grounded in real-world frameworks, in which unreliability is inferred by projecting familiar social scripts. Readers map culturally specific notions of shame, honor, and social standing on their interpretations of character motivation. The assumption that a loss of face motivates lying draws on longstanding cultural scripts and narrative tropes that link pride with credibility.

Last, characters who stood to gain from lying (N = 27) or had nothing to lose (N = 15) were judged accordingly. For example, Hiro’s death meant he had “nothing to lose” (N = 27), bolstering his credibility, while Tomoko’s arrest imminent death sentence (N = 12) similarly implied a lack of incentives to lie, a classic example of the “disclosure against interest” principle: “In his statement he was already defeated, he was ready for his punishment. So, he had nothing to lose but to tell the ‘real’ story of what happened.”

6.3. Narration

In addition to social roles and character traits and behavior, narrative form also played a critical role. Participants cited clarity (N = 8), completeness (N = 13), consistency (N = 5), level of detail (N = 18), embellishment (which included making self or others look better than they are, N = 19), emotional expression (N = 23), objectivity (N = 7), and tone (N = 12) as factors affecting trust. Often, the same aspect could signify both reliability and unreliability, for instance in the case of emotion, which could indicate honesty—“she was in such a pain because her daughter was gone that she would not lie about it”—as well as its opposite: “She only wants the best for her daughter and therefore says anything to protect her. This is not trustworthy when emotion play [sic] a big role.” Similarly, too little detail was cited reasons for distrust: “The policeman didn’t have much information and his testimony didn’t seem detailed enough,” but too much detail equally led to contradictory assessments of the same narrators: “I felt like the policeman was talking a lot and came up with a lot of unnecessary details to support his story. I felt like he was rambling a lot in his statement.”

Besides narrative characteristics of individual accounts, readers also made comparisons between accounts. Additionally, participants compared accounts for convergence (N = 10) or divergence (N = 23). An example of a statement based on divergence is “As 1 of the 3 person who were involved in the crime, Yuki’s confession regarding the death of Hiro shared no common details with the confession made by Hiro and Tomoko,” whereas the following assessment was based on convergence:

I trusted Tomoko the most, because he said that Yuki clinged [sic] onto his arms and that he said to him to kill Hiro and that Yuki said that she can only marry the man who survived. These sentences came back in story’s [sic] of other people so that makes it thrustworthy [sic].

Descriptions by other characters were also used as corroboration to assess a character’s reliability: “I read a lot of negative opinions of him and also I read about the crimes and potential crimes he committed.” Such assessments indicate that inter-narrative consistency is an important heuristic for trustworthiness. This directly invokes Sperber et al.’s (2010) notion of epistemic vigilance, here understood as a cognitive mechanism through which readers evaluate conflicting testimony to form coherent mental models. When discrepancies persisted or exceeded a certain threshold, readers inferred unreliability, echoing A. Nünning’s (1999) idea of unreliability as a hermeneutic strategy for restoring coherence. The attention to inter-narrative consistency, corroboration by other characters, and subjective bias (for instance due to trauma) also maps onto Phelan’s (2017) communication model. Here, judgments often hinged on all three axes: the factual accuracy of events (misreporting), the narrator’s interpretive competence (misunderstanding), and the value framework behind their evaluations (misevaluation).

Last, related to objectivity, but not so much narration as perspective, thirteen participants named subjective perspective as a reason for distrust. Subjectivity of perspective could for instance stem from traumatic experiences, e.g.,: “She undergoes trauma, potentially clouding her memory and narrative.” This aligns with Phelan’s (2017) typology, where readers distinguished between narrators perceived as intentionally misleading and those interpreted as fallible or morally biased. We see that readers often judge trustworthiness based on surface cues like tone, consistency, and completeness. While this evokes Booth’s and Nünning’s text-centered models, these judgments are always shaped by readers’ own perspectives and by the broader ecology of competing voices.

6.4. Reader-Related Reasons and Subjectivity

A smaller subset of participants (11) mentioned reasons for (dis)trust that were not so much text-related, as related to themselves as readers. They acknowledged their own subjectivity as influencing trust judgments: some admitted bias, others relied on gut feelings, and some confessed to limited attention (“the policeman did not strike my attention in anyway”) or recall (“I don’t really remember his story meaning I possibly didn’t find him trustworthy”; “it’s the one I remember the best”). This self-reflexivity highlights the co-constructive nature of trust, where reader disposition, attention, and memory all shape reliability assessments.

Together, these findings demonstrate how narrative trust judgments emerge from a complex interplay of social roles, character motivation, narratorial form, and reader subjectivity. They echo and extend longstanding narratological theories of reliability (Booth 1961; Olson 2003; Phelan 2017) by foregrounding the situated, ethical, and interpretive work readers perform when navigating multi-perspectival narrative worlds. Cognitive narratological theories (Yacobi 2005; Tobin 2022) that foreground reader perspectives and cultural backgrounds explain how readers interpret the reliability of narrators differently: we see variation in which testimony each reader found most credible, and how moral or cognitive biases influenced their judgments. Different readers bring in diverging real-world knowledge and personal values to make sense of the story’s conflicting narratives.

7. Quantitative Analysis and Discussion

7.1. Reading: Frequency and Familiarity

Spearman’s rank-order correlation was conducted to assess the relationship between Author Recognition Test (ART) scores and reported reading frequency. Results revealed a moderate, positive correlation, ρ(78) = 0.42, p < 0.001. This indicates that higher frequency of reading literary texts is associated with greater familiarity with authors, as measured by ART scores—interpreted here as a proxy for print exposure.

7.2. Character Trust Rankings and Gender

In trustworthiness rankings, Hiro emerged as the most frequently selected “most trusted” character (41 mentions; 29.3% of responses), while Yuki was least often selected as most trusted (5.4%). Regarding “least trusted” characters, the Bandit was most frequently selected (50 mentions), accounting for 31.5% of all responses and 33.3% of valid responses (excluding missing data). Yuki followed closely with 47 mentions (31.5% of all responses; 33.3% of valid ones). To examine gender differences, a chi-square test of independence was conducted to test the association between participant gender and selecting the Wife as the least trusted character. The test was not statistically significant, χ²(2, N = 141) = 4.95, p = 0.084. While not conclusive, this result may suggest a trend worth exploring further in a larger sample.

7.3. Character Trust Rankings and Narrative Order

To investigate whether narrative order influenced trust attribution, we randomized the sequence of the three central testimonies (Wife, Bandit, Samurai) across six conditions. A chi-square test of independence examined the effect of narrative order on the selection of the most trusted character. Results showed no significant association, χ²(30, N = X) = 26.84, p = 0.632, suggesting narrative order did not meaningfully influence whom participants trusted most. However, a second chi-square test assessing the association between narrative order and the least trusted character revealed a statistically significant effect, χ²(30, N = X) = 44.99, p = 0.039. This suggests that narrative sequencing did influence distrust judgments.

A closer inspection of the contingency table showed that Bandit (Tomoko) was most often selected as least trusted in orders 1 and 2 (which presented his account first after minor characters). In contrast, the wife Yuki’s distrust frequency increased in conditions 5 and 6 (where her testimony appeared earlier among the main testimonies). The samurai Hiro was more evenly distributed across conditions. This discrepancy may be explained by the fact that many participants selected a minor character like the woodcutter, policeman, or Priest as ‘most trusted,’ since these were uninvolved in the murder and considered objective eyewitnesses with no stakes in lying, as we saw in the qualitative analysis. For least trusted, the majority of participants chose one of the main characters, who were present at, or victim of, the murder, and whose versions contradicted each other dramatically.

These patterns may also suggest a primacy effect in the assignment of distrust: characters whose testimonies appeared earlier in the sequence were more likely to be judged untrustworthy. Interestingly, the same effect did not emerge for trust, suggesting that narrative order may more strongly influence suspicion than belief. This asymmetry is likely due to the epistemic framing of “In a Grove,” which primes readers for skepticism in the face of conflicting testimony. This is consistent with (empirical) findings in narrative psychology and cognitive science (Nahari and Ben-Shakhar 2013; Demorest and Gleckel 2014; Tuckett and Nikolic 2017). Early testimonies may establish a suspicion anchor or frame subsequent narratives as potentially unreliable. Later accounts may be scrutinized more critically if they diverge significantly. This echoes recency effects in suspicion attribution and supports findings in epistemic vigilance (Sperber et al. 2010), where initial narratives set interpretive expectations that shape trust and reliability allocation. It shows the construct of narrative (dis)trust is not a stable trait-based evaluation, but a dynamic, context-sensitive judgment.

Our finding that narrative order disproportionately shaped suspicion but not ultimate belief also resonates with research on negativity bias and trust asymmetry in cognitive psychology. This line of research shows that distrust is more readily triggered and more durable than trust (Baumeister et al. 2001; Lewicki et al. 1998). In line with primacy effects in impression formation, early testimonies may serve as suspicion anchors (Anderson 1965), but subsequent information can still re-open interpretive possibilities, producing the observed tension between anchoring and epistemic openness. More recent cognitive-psychological work shows that distrust, unlike trust, enhances cognitive vigilance and memory performance (Schul et al. 2008; Posten and Gino 2021). In addition, experience-sampling research reveals how affectively driven trust judgments arise quick, but remain contextually flexible (Legood et al. 2023; Weiss et al. 2022), which could account for the noted tension between interpretive anchoring and later epistemic openness.

7.4. Reasons for (Dis)trust and Personality Traits

To explore connections between readers’ cognitive profiles and their narrative trust judgments, we converted participants’ qualitative responses into numerical codes based on thematic labels. Thirteen broader categories were used to reflect reasons for trusting or distrusting characters.6 These were then compared to scores on ART, Need for Cognition (NFC), and trust-related measures to examine whether participants who prioritize certain types of reasons for trust differ in epistemic trust or cognitive style. For example, do people high in need for cognition mention different reasons than those low in need for cognition?

A one-way ANOVA tested whether Need for Cognition scores differed across groups defined by their most trusted character (N = 73). Results showed no significant differences, F(23, 49) = 1.23, p = 0.265. A similar ANOVA for least trusted characters also found no significant effect, F(22, 52) = 1.29, p = 0.22. Likewise, neither ART nor NFC scores showed meaningful differences by gender or by trust/distrust rankings. These findings suggest that individual cognitive traits and literary expertise (as measured by ART and NFC) do not significantly predict which characters participants trust or distrust.

This absence of significant correlations between cognitive profiles (as measured by NFC and ART) and participants’ trust judgments could point us to the limitations of trait-based explanations in capturing the complexities of narrative evaluation. Trust in narrators does not appear to stem from stable dispositions such as cognitive effort or print exposure. It is context-sensitive and shaped by local textual cues, genre expectations, and readers’ interpretive frameworks. Readers who are familiar with modernist ambiguity may for instance be more attuned to narrative indeterminacy and therefore less inclined to base their judgments on character coherence, regardless of their cognitive style. Latent variables such as genre literacy or cultural familiarity with interpretive norms may thus moderate the relationship between cognitive traits and trust. Interaction effects are also plausible: a high need for cognition might influence trust judgments only in combination with genre-specific expertise or certain reading habits. Without direct measures of genre familiarity or interpretive training, such interactions remain undetected. Future research could address this by incorporating measures of media exposure, prior experience with unreliable narration, or interpretive styles.

All in all, the heterogeneous and sometimes contradictory reasons participants gave for trusting or distrusting characters resist clustering along trait-based lines like credulity or epistemic trust. These findings challenge the assumption that narrative trust is shaped primarily by cognitive style or cultural capital, and support recent interventions in cognitive narratology that emphasize the affective and situated nature of trust in narrative comprehension (cf. Kukkonen 2018). Though subtle patterns may have been missed due to the modest sample size and the granularity of qualitative codes, the overall result points to the limits of predictive power in trait-based approaches to narrative trust.

8. Conclusions

While narratological theory offers robust conceptual frameworks for understanding narrative (un)reliability, empirical studies of how actual readers negotiate trust and interpret unreliability remain sparse. This article contributes to bridging this gap by examining reader responses to a canonical example of multiple, conflicting narrators: Ryūnosuke Akutagawa’s “In a Grove” (Akutagawa [1922] 2007). The story’s layered narrative voices provide an ideal testing ground to observe how readers engage with unreliable narration in practice and how such engagement modulates reading strategies and interpretive outcomes. The theoretical framework for unreliable narrators emphasized that trust and reliability are dynamic and involve feedback between textual signals and reader interpretation (Korthals Altes 2014).

Our qualitative findings support this by showing how readers actively shift between trusting and distrusting the narrators as they encounter conflicting testimonies in “In a Grove”. We see that readers use a complex interplay of personal stakes, social roles, narrative qualities, and cultural expectations to evaluate the reliability of narrators and witnesses. Trust judgments are rarely neutral or purely epistemic but are embedded in social and emotional contexts, mediated by genre conventions, and shaped by implicit cultural scripts about authority, victimhood, and confession. This aligns with recent empirical narratology work that underscores the situatedness and contextuality of reliability judgments (Van de Ven 2025). Readers reported changes in their trust as new, contradictory accounts emerge, reflecting this feedback loop rather than a fixed judgment. We found instances where readers initially accept a narrator but revise their judgment after noticing textual inconsistencies or paratextual cues. Participants naturally made distinctions between unreliable narrators who lie with the intent to deceive, immortal narrators, and fallible ones who are confused. Many participants reported varying degrees of trust towards different narrators, with some explicitly labeling certain accounts as deceptive or self-serving. Such findings support the notion of an interplay of bottom-up (textual clues like contradictions or style) and top-down (readers’ prior knowledge, cultural schemas) processes in shaping trust shifts. This links to the integrated approach from Olson (2003) and the framework presented in Figure 1. These preliminary findings underscore the dynamic interplay between text-immanent cues and reader-specific cognitive frameworks in shaping narrative trust. They illustrate that the reception of unreliable narration is neither uniform nor fixed, but contingent on individual epistemic strategies and prior knowledge, confirming the cognitive school’s emphasis on reader variability (Yacobi 2005; Fludernik 1996).

Quantitative analysis revealed that narrative order did not influence whom participants trusted most, but did shape whom they distrusted, suggesting a primacy effect in suspicion attribution. This asymmetry implies that narrative order may function more as a suspicion anchor than a trust scaffold, aligning with theories of epistemic vigilance and interpretive anchoring (Sperber et al. 2010).

Individual cognitive traits such as literary expertise (ART) or need for cognition (NFC) did not significantly predict trust judgments, reinforcing that such judgments are less a function of stable traits than of interpretive strategies and contextual framing (Yacobi 2005). This supports a model of trust as the result of both bottom-up textual cues and top-down reader expectations. These results seems to indicate that, in this context, neither greater literary familiarity or reading frequency, nor higher cognitive engagement predisposes readers toward trusting particular characters in the narrative. This aligns with prior findings that trust judgments in narrative reception are not simply reducible to cognitive traits or domain expertise but may be influenced more by interpretive strategies and narrative context (Phelan 2007). The lack of clear quantitative relationships suggests that trustworthiness assessments might better be understood through qualitative inquiry into readers’ reasons and interpretive practices, rather than through demographic or cognitive trait variables alone.

In future research, eyetracking data will be triangulated with the qualitative data in order to find out more about patterns of reading and attention in relation to trust and reliability. This way, we can examine whether reading behavior and attention modulation varies for the narrators the participants deemed (un)reliable. Unreliable elements are expected to lead to longer reading times, higher recall and recognition of textual elements, and better reproduction and reconstruction. Increased vigilance causes narrative engagement and is therefore expected to elicit longer and fewer fixations (cf. Bruys 2013). Eyetracking can thus offer more information about cognitive processes: where readers allocate their attention, how they modulate, and what strategies of reading result from these processes. How do readers calibrate trusting and distrusting reading attitudes in reading narrative fiction and what he modes and strategies of reading correspond to these attitudes? Analyzing these data and triangulating them with those on personality traits and interpretations could increase our understanding of how trust in a narrator is established.

While “In a Grove” offers a particularly rich site for examining reader responses to unreliable narration due to its polyvocality and undecidability, the findings are necessarily limited by the choice to use a single, canonical short story. Future research will need to test whether the interpretive dynamics and trust calibrations that were found also hold across other narrative forms and genres such as autofiction, diary novels, or thrillers, which mobilize unreliability in different ways. In addition, it is important to consider that participants were predominantly undergraduate students from programs in communication studies and cognitive science. As such, their interpretive strategies may be shaped by disciplinary training that emphasizes analytical, information-processing, or media literacy frameworks. This could limit the generalizability of our findings to a broader reading public with different educational or literary backgrounds.

Beyond literary studies, our insights offer potential relevance for media literacy and information consumption in the postdigital era, where distinguishing trustworthy from unreliable narratives is crucial across diverse platforms—be it in journalism, social media, or political discourse. Trust and reliability are at the core of social life and have become urgent issues in our mediatized culture. In the ‘post-truth era,’ fostering critical media consumers is of paramount importance for the liberal democracies in which we live. By offering insights into how readers come to their assessments of trustworthiness, empirical reader research could contribute to building resilience against misleading and contradictory narratives. Narratological investigations provide a foundation for developing narrative competence, which helps individuals understand different types and degrees of trustworthiness and enhances their ability to understand others’ motivations. This competence equips us with the necessary tools to navigate the complexities of today’s information-rich society, especially in the face of a reported ‘reading crisis’ in the U.S. and many European countries, where reading comprehension scores have been steadily declining. In today’s challenging media landscape, students must cultivate narrative competence to become critical media consumers. This includes recognizing the strategies media use to evoke trust and adopting appropriate reading strategies. By elucidating the mechanisms through which readers detect and respond to unreliability, this research contributes to a broader understanding of narrative trust that extends to real-world epistemic challenges.