Augmented Reality and Inferential Comprehension in Advanced EFL Learners: Disfluency, Metacognitive Reflection, and Productive Struggle

Abstract

Augmented Reality (AR) continues to generate interest as a pedagogical tool in contexts where English is a Foreign Language (EFL). However, its role in developing higher-order cognitive skills, such as inferencing, remains underexplored. This exploratory, mixed-methods study investigates whether AR can scaffold inference-making in advanced EFL learners. Forty-seven university students in XX were assigned to either a control group (CG) or an experimental group (EG). Both groups read Edgar Allan Poe’s “The Tell-Tale Heart” in digital format. The CG received a conventional inference-based comprehension lesson, while the EG engaged with two interactive AR scenes developed using the Onirix Studio platform. Pre- and post-tests assessed inferential comprehension, and qualitative data were gathered through open-ended responses. While the CG demonstrated modest post-test gains and the EG showed a slight decline, neither change reached statistical significance. Notably, qualitative findings revealed that a salient AR element—a ticking clock—likely prompted misinterpretation in the EG, disrupting symbolic reasoning and contributing to schema misalignment. However, some learners exhibited metacognitive insight suggestive of productive struggle. These results suggest that AR may hold untapped potential for developing metacognitive reflection and critical literacy. Directions for future research are outlined.

1. Introduction

Across education systems, the integration of digital technologies has accelerated the use of immersive tools to support complex learning (Conrad et al., 2024; Li et al., 2022; Timotheou et al., 2023). Yet essential skills for academic reading and learning still depend on readers’ capacity to construct meaning that goes beyond the literal text, i.e., inferential comprehension (Kan et al., 2024; Palinscar & Brown, 1984). Augmented reality (AR), which overlays digital content onto the physical environment, represents one of the most recent and promising developments in this trajectory. By layering visual and auditory cues onto print, AR can potentially scaffold the cognitive processes that underpin deep reading; however, its value depends on the extent to which these cues align with the mental operations required for inference-making (Danaei et al., 2020; Şimşek et al., 2024).

Reading comprehension constitutes a complex construct that encompasses literal, inferential, and evaluative skills (Israel, 2017). Among these, inferential comprehension—the integration of textual information with prior knowledge—poses a particular challenge for EFL learners, especially due to linguistic and cultural barriers that hinder access to implicit meanings and contribute to gaps in reading proficiency (Koda, 2007). Research on digital reading comprehension has highlighted how online texts and multimodal environments alter the ways readers process and integrate information, often increasing cognitive demands while providing opportunities for interactive scaffolding (Li et al., 2022; Timotheou et al., 2023). These findings suggest that instructional designs involving digital texts must carefully balance engagement with support for higher-order comprehension.

Recent scholarship has established the benefits of augmented reality (AR) in enhancing vocabulary acquisition, engagement, and motivation (Belda-Medina, 2022, 2025; Belda-Medina & Marrahi-Gomez, 2023; Haoming & Wei, 2024; Lai & Chang, 2021). However, its potential to facilitate higher-order comprehension, such as inferencing, remains underexplored in rigorously controlled experimental research, particularly within EFL classroom settings where cognitive demands are high.

In recent years, AR has garnered scholarly interest in educational research (Al-Ansi et al., 2023; Koumpouros, 2024; Schorr et al., 2024; Singh et al., 2024). A systematic review by Akçayır and Akçayır (2017) identifies several core benefits of using AR in education: improving learning achievement, increasing student motivation, and supporting multimodal learning. These benefits are generally accomplished through dynamic AR environments that help learners process abstract or complex material through multiple sensory channels. Similarly, Zhang et al. (2025) compared AR and VR in vocabulary learning contexts, highlighting how multimodal features can influence engagement and comprehension. While these studies consistently report motivational gains, mixed evidence of long-term cognitive gains suggests that AR’s instructional value depends heavily on how its features align with specific pedagogical goals (Akçayır & Akçayır, 2017; Ibáñez & Delgado-Kloos, 2018; Zhang et al., 2025).

Several recent studies have examined AR’s potential to support reading comprehension. Danaei et al. (2020) found that children who interacted with an AR-enhanced edition of “Where the Wild Things Are” outperformed a control group on general comprehension measures, including a subset of questions requiring inference. Du et al. (2024) echoed these findings in a second-grade EFL cohort in China: although explicit comprehension gains were negligible, AR storybooks yielded substantial improvements in implicit comprehension and narrative reconstruction. Similarly, Şimşek and Direkçi (2023) reported that AR-enhanced texts improved both literal and inferential comprehension in Turkish primary learners and helped sustain attention over successive sessions. Extending this work, Şimşek et al. (2024) examined the role of AR exposure time in reading comprehension among Turkish L1 students aged 11–12. They found that all exposure groups improved relative to print-only baselines, but the 24–27 s condition yielded the highest comprehension scores and lowest cognitive load, suggesting that temporal calibration may also influence AR’s cognitive benefits.

Fewer studies have examined AR’s role in reading comprehension among more advanced EFL learners. Ebadi and Ashrafabadi’s (2022) Reader Buddy project, implemented via CXOCARD and ROAR1, found that upper-intermediate Iranian students achieved significantly higher post-test scores than their peers taught through traditional methods, which is an outcome the authors linked to AR’s ability to help learners draw on prior knowledge, ease unnecessary mental effort, and promote active involvement in reading tasks. More recently, Asadi and Ebadi (2025) investigated intermediate adult EFL learners using the XR Plus platform and reported large effect-size gains in reading comprehension compared with a control group. These gains were attributed to the visual scaffolds, multimodal vocabulary support, and immersive qualities of AR, which instructors also observed encouraged inferencing and critical reading strategies.

To better interpret AR’s potential role in inferential comprehension, it is useful to frame these findings within established theories of cognitive processing in reading. From a cognitivist perspective, reading comprehension develops as learners actively encode, integrate, and reorganize information to construct coherent mental models of a text (Kintsch, 1998). More specifically, inferential comprehension places high demands on working memory because it requires bridging gaps between explicit text and prior knowledge while processing multiple semantic and contextual cues (Graesser et al., 1994). Cognitive disfluency occurs when learners encounter information that is difficult to process, often slowing comprehension but potentially fostering deeper reflection and problem-solving (Bjork & Bjork, 2011). In the context of AR, disfluency may arise when visual and auditory cues challenge learners’ expectations or interpretations, prompting metacognitive monitoring and productive struggle.

Several complementary frameworks illuminate how AR might influence these processes. Dual Coding Theory (Paivio, 1986) and Mayer’s (2009) Cognitive Theory of Multimedia Learning suggest that pairing verbal and visual information can enhance comprehension when modalities are well-aligned, while Cognitive Load Theory (Sweller, 1988) cautions that poorly integrated multimodal cues may overload working memory and fragment schema construction. These perspectives imply that AR’s visual and auditory scaffolds could facilitate inference-making by enriching textual representation—yet they also highlight the risk that misaligned or extraneous elements might distract attention from core interpretive tasks (L. W. Anderson & Krathwohl, 2001; Mayer, 2009; Sweller, 1988).

Despite promising evidence that AR can enhance both literal and inferential comprehension, existing research leaves two critical questions unanswered. First, most prior studies have focused on younger L1 learners or intermediate-level EFL students, often in multi-session interventions, limiting insights into how AR functions in time-constrained higher-education contexts with advanced EFL learners. Second, even in studies involving older learners, comprehension has typically been measured using general instruments (e.g., Dialang) rather than tasks designed to isolate inferencing as a distinct cognitive process. As a result, it remains unclear whether AR’s multimodal scaffolds can meaningfully support the specific demands of inferential comprehension in advanced EFL reading.

To address this two-part gap, the objectives of this study were twofold: (1) to examine whether AR-enhanced literary materials can improve inferential comprehension in advanced EFL learners within a higher-education context, and (2) to explore how learners perceive and interpret these materials in relation to their reading processes. To meet these objectives, the study was conducted with undergraduates at a Spanish university who engaged in close reading of Edgar Allan Poe’s The Tell-Tale Heart (Poe, 1843/2021) a text selected for its rich symbolism, psychological complexity, and high inferential demands. The experimental group interacted with two custom AR scenes created in Onirix Studio, designed to integrate visual and auditory cues with key passages, while the control group completed an equivalent traditional comprehension lesson. This single-session design enabled a controlled comparison of instructional formats under authentic classroom conditions. The research was guided by two questions:

• What effects does the use of AR-enhanced reading materials have on the inferential reading comprehension of advanced EFL learners compared to traditional instruction?
• What learner-reported experiences help explain any observed differences in inferential comprehension outcomes between AR and traditional formats?

2. Materials and Methods

2.1. Research Design

This study employed a quasi-experimental, mixed-methods design to examine whether an AR intervention could support inferential reading comprehension among advanced EFL learners. The intervention was conceived as a pedagogically equivalent alternative to a traditional lesson plan; both conditions targeted the same interpretive goals using different teaching formats. The independent variable was instructional modality (AR-enhanced vs. traditional), and the dependent variable was performance on an inferential reading comprehension assessment administered before and after the intervention.

Participants were assigned to experimental or control groups based on pre-existing class groupings. All learners were provided digital PDF copies of Edgar Allan Poe’s “The Tell-Tale Heart”2. Instructional design was held constant across both groups to ensure both formats scaffolded inferential comprehension through guided reflection, multimodal engagement, and group collaboration. The experimental group interacted with two AR scenes developed in Onirix Studio, while the control group completed parallel close-reading tasks supported by small-group discussion. The study utilized both quantitative and qualitative data sources to examine instructional effects on inferential comprehension.

2.2. Participants

Participants were 60 third-year undergraduate students enrolled in an Applied Linguistics course at a university in XX. All were B2–C2-level English users with prior experience analyzing literary texts in academic settings. Ages ranged from 20 to 40 years (M = 21.6). Most identified as female (71% EG, 81% CG) and reported Spanish as their native language. Reading habits were similar across both groups: nearly all reported reading in English for academic purposes, while a higher percentage in the experimental group (EG) indicated personal enjoyment as a primary reason for reading books in English.

The initial participant pool included 34 students in the EG and 26 in the control group (CG). Of these, 47 students (30 EG, 17 CG) completed all key phases of the study—pre-test, intervention, and post-test—and were included in the main analysis. A slightly smaller subset (n = 45; 29 EG, 16 CG) completed the post-intervention survey. Although the study focused on advanced EFL learners, four native English speakers were enrolled in the same course sections and participated under identical conditions; their small number and advanced academic profile did not significantly affect group outcomes.

Group assignment followed a quasi-random structure: the two intact classes were already formed administratively, and one was randomly designated as the control group while the other was assigned to the experimental group. Participation was voluntary, and informed consent was obtained from all students. Email addresses were temporarily collected in the pre- and post-tests solely to match responses across timepoints; all identifiers were removed during data cleaning before analysis.

2.3. Instruments

To measure inferential comprehension, a researcher-designed assessment was administered before and after the intervention. The instrument consisted of two components:

• Multiple-choice questions: Three items, each accompanied by a required open-ended justification. On the post-test, items were reworded slightly to reduce recall bias while preserving inferential demands. Multiple-choice responses were scored dichotomously (correct/incorrect) in Google Forms for quantitative analysis.
• Short-answer prompts: Five items targeting literary interpretation, symbolic understanding, and narrative evaluation.

The comprehension instrument was purpose-built for this study to isolate inferential comprehension and address a gap in prior research that has often relied on general-purpose reading assessments. Items were modeled on the inferential reasoning levels of Bloom’s taxonomy and targeted higher-order literary analysis skills (L. W. Anderson & Krathwohl, 2001). While the instrument was not adapted from an existing standardized measure, all items were reviewed for clarity and content relevance by the supervising faculty member.

Open-ended justifications and short-answer responses were exported to Atlas.ti for qualitative analysis. Responses were later thematically coded using a hybrid deductive–inductive approach (see Section 2.5, Data Analysis). Following the comprehension assessment, students completed a brief post-intervention survey. This included open-ended items on lesson perceptions for both groups; the experimental group received additional prompts about their AR experience and its perceived impact on comprehension. Surveys were completed in class immediately after the post-test.

The comprehension assessment and post-intervention survey were part of a larger assessment battery that also included additional instruments. Only the components relevant to the present study’s focus on inferential comprehension are reported here. The full text of the comprehension assessment and post-intervention survey items used in this study is provided in Appendix A and Appendix B, respectively.

2.4. Materials

The instructional materials for the experimental condition consisted of two augmented reality (AR) scenes based on Edgar Allan Poe’s “The Tell-Tale Heart”, developed in Onirix Studio v2.73.1. to v2.73.3 (January to March 2025). The project was accessible via mobile devices (Android and iOS) with camera-enabled browsers. File sizes ranged from 5 to 6 MB per scene.

The AR experience depicted two key scenes from the story:

• Scene 1—The nocturnal observation: A dimly lit bedroom with the old man asleep, his “vulture eye” exaggerated as a prominent visual element. See Figure 1 for a representative screenshot captured from this AR scene, showing the exaggerated eye and an embedded reflective prompt activated during interaction.
• Scene 2—The auditory hallucination: A gradual build-up of a heartbeat sound synchronized with visual cues of guilt and psychological collapse. See Figure 2 for a screenshot captured from this AR scene, showing the visualized beating heart accompanied by a direct quotation from the text.

The AR scenes were designed as interpretive adaptations of “The Tell-Tale Heart” and incorporated intentional elements that are not present in the original text to emphasize thematic and symbolic dimensions. Prior to the main study, the AR resources were pilot-tested with a small group of students (n = 6) to ensure usability and content relevance. While these tests confirmed the potential of the materials, they also revealed occasional technical instability and synchronization delays between audio and visual elements. These issues were acknowledged as limitations that could affect participants’ experiences and interpretation of the AR scenes. For example, a 3D animated clock and subtle auditory motifs that reinforced concepts such as time, inevitability, and psychological tension. These visual and auditory additions were purposefully integrated with embedded textual prompts and direct quotations from Poe’s narrative. These features were intended to prompt students to cross-reference the AR representations with the original text, evaluate interpretive choices, and reflect critically on the author’s thematic intentions.

The AR features were designed to align with symbolic and thematic elements in the narrative. For instance, the visual exaggeration of the eye corresponded to the narrator’s fixation in the text, while the combined heartbeat–clock motif represented pacing and inevitability. During the intervention, students accessed the scenes in pairs or small groups, using embedded prompts and quotations to compare the AR representations with the original text and discuss the narrator’s reliability and emotional state.

The control group worked with the same unabridged digital PDF of “The Tell-Tale Heart” provided to the experimental group, without AR enhancements. Their materials included printed task instructions and discussion prompts corresponding to the same narrative segments represented in the AR scenes. These prompts were text-based and designed to elicit close reading, quotation selection, and interpretive discussion of the narrator’s perspective and psychological state. To illustrate, two representative excerpts from the close-reading passages are provided below. These excerpts correspond to the same two narrative moments represented in the AR scenes; the control group received the complete passages during the intervention.

And then, when my head was well in the room, I undid the lantern cautiously—oh, so cautiously—just so much that a single thin ray fell upon the vulture eye. And this I did for seven long nights—every night just at midnight—but I found the eye always closed; and so it was impossible to do the work; for it was not the old man who vexed me, but his Evil Eye.

(Excerpt A)

It was a low, dull, quick sound—much such a sound as a watch makes when enveloped in cotton. … It grew louder—louder—louder! … “Villains!” I shrieked, “dissemble no more! … it is the beating of his hideous heart!”.

(Excerpt B)

2.5. Data Analysis

Multiple-choice responses were analyzed in SPSS (version 29) using paired-sample t-tests to compare pre- and post-test scores within groups and independent-sample t-tests to compare scores between groups. The significance threshold was set at α = 0.05, and effect sizes are reported as Cohen’s d.

Open-ended justifications from the multiple-choice items, short-answer responses, and post-intervention survey comments were imported into Atlas.ti for thematic analysis. Coding followed a hybrid deductive–inductive approach: a preliminary codebook was developed from the research questions and theoretical framework, and additional codes were added to capture emergent patterns. Codes prioritizing inferential comprehension included inference-making, narrator reliability, emotional and sensory engagement, and textual visualization. All coding was conducted by the primary researcher, with iterative review to ensure consistency and analytic rigor. All student responses are reproduced verbatim unless otherwise indicated. Ellipses (“…”) denote minor omissions made to retain only the most relevant portions of a response, without altering its meaning or tone.

2.6. Procedure

The study was conducted during regular class hours in on-campus university classrooms. Each session followed an identical 2-h structure and was facilitated by the researcher. All students received the unabridged digital version of “The Tell-Tale Heart” and were allocated 20 min for silent reading following a brief (~5 min) scripted introduction to the author, the narrative context, and the concepts of narrator reliability and inference-making.

After reading, students completed the pre-test (25 min). The central 30-min instructional segment differed by group. Students in the control group worked in small groups (3–4 members) on two close-reading tasks centered on the same narrative passages adapted for the AR condition (see Section 2.4). These discussions focused on identifying textual evidence, comparing different interpretations, and debating the narrator’s reliability, guided by prompts provided by the instructor. This collaborative exchange likely supported more stable interpretations by encouraging students to justify their reasoning and resolve ambiguities collectively.

The first task examined the nocturnal observation scene, and the second analyzed the closing section depicting the narrator’s psychological collapse. Activities included drawing the old man’s eye as described in the text and selecting a quotation that best represented the narrator’s mental state. The researcher circulated among groups to offer clarification, pose guiding questions, and encourage critical engagement.

In the experimental group, students also worked in small groups (3–4 members) with the AR experience described in Section 2.4, which adapted the same two narrative passages used in the control tasks. Embedded textual prompts within each scene guided real-time analysis, and groups were then given a set of discussion questions comparable to those used in the control condition. These were designed to consolidate interpretation and link the AR experience back to the text. The researcher circulated to provide clarification, guide discussion, and offer technical support as needed.

In both conditions, students were asked to anchor their interpretations in textual evidence and to consider how details from the narrative shaped their understanding of the narrator’s reliability and emotional state. Following the instructional segment, students completed the post-test (25 min), which replicated the pre-test format with minimal wording adjustments. The post-intervention survey (10 min) was then administered. All data were later compiled for analysis.

3. Results

Of the 60 students who completed the pre-survey and pre-test, 47 (30 EG, 17 CG) completed both the pre- and post-test and were included in the main analysis. Forty-five participants (29 EG, 16 CG) also completed the post-intervention survey.

3.1. Quantitative Results

Inferential comprehension was assessed using three multiple-choice items with open-ended justifications. The same questions were used pre- and post-intervention, with minor wording adjustments to reduce memorization. Scores were calculated as the proportion of correct responses per group. At pre-test, the experimental group (EG) averaged 70% correct, compared to 65% in the control group (CG). Following the intervention, CG scores increased modestly to 71%, while the EG experienced a decline to 65%. These results are summarized in Figure 3.

To assess whether prior familiarity with “The Tell-Tale Heart” influenced performance, an independent-samples t-test was conducted on comprehension gain scores. No statistically significant difference was found between students who had previously read the story (n = 6) and those who had not (n = 36), t(40) = 1.30, p = 0.201, d = 0.57. This suggests that prior exposure did not meaningfully affect outcomes. Moreover, paired samples t-tests revealed that neither change was statistically significant, and both effect sizes were small.

A pronounced shift was evident in responses to Question 2 (Q2), which assessed students’ interpretation of the symbolic function of the narrator’s auditory hallucination. On the pre-test, 29 out of 30 EG participants answered correctly (97%), compared to 21 out of 30 (70%) on the post-test. See

Table 1. Mean Pre- and Post-Test Scores by Comprehension Item and Group.

Group	Q1 Pre	Q1 Post	Q2 Pre	Q2 Post	Q3 Pre	Q3 Post
CG	1.00	0.88	0.71	0.94	0.24	0.29
EG	0.87	0.90	0.97	0.70	0.27	0.33

for a breakdown of the proportion of correct responses for each comprehension item by group and test phase.

Item-level analysis revealed that Q2 accounted for 161% of the experimental group’s total decline in comprehension scores, a result that may initially appear counterintuitive. This disproportionate figure reflects slight gains for the other two items; Q2 alone accounts for—and exceeds—the total score decline. Notably, Q2 was the only item to display a substantial disparity between groups: while performance for Q1 remained stable or improved slightly in both groups, and Q3 remained challenging for all participants (with mean scores below 0.35), Q2 alone exhibited a bidirectional shift. Specifically, the experimental group’s performance for Q2 declined from 97% to 70%, whereas the control group’s scores increased from 71% to 94%. This distinct pattern suggests that the experimental condition, rather than item difficulty or prior familiarity, was responsible for the experimental group’s unexpected decline. This finding is explored further in the qualitative analysis.

3.2. Qualitative Results: Thematic Analysis of Open-Ended Responses

Responses from multiple-choice justifications and open-ended post-survey questions were thematically coded for this analysis. Prominent code categories included Inferencing_Strategy, Narrator_Reliability, Emotional_Engagement, and AR_Misinterpretation. Frequency data were normalized by group size to enable equitable comparison.

In the multiple-choice justification items, Inferencing_Strategy appeared in at least 80% of responses across all conditions, including over 96% of post-test EG replies. This prevalence demonstrates that both groups relied heavily on inferential reasoning to justify their selections. Typical CG responses demonstrated structured reasoning grounded in textual analysis, whereas EG responses more often combined inferential logic with psychological or affective framing. For instance, one EG participant noted: “He is trying to convince us that what he did was reasonable and that he is not insane; his paranoia and guilt are consuming him, and he needs to justify his actions”. This comment exemplifies the synthesis of textual interpretation and character psychology characteristic of EG justifications.

Continuing with the multiple-choice justifications, Narrator_Reliability reflected increased attention to the narrator’s subjectivity and psychological instability. This rose from 53.33% to 65.56% in the EG and from 33% to 52.94% in the CG across pre- and post-tests. A representative EG response observed: “The narrator insists on his sanity despite exhibiting irrational behavior… his obsession and distorted perception reveal his paranoia and guilt.” Such responses signal a deepening sensitivity to narrative perspective and internal inconsistency, both of which are essential components of higher-order inferencing. For instance, another participant commented, “At first, I believed the narrator was trustworthy, but after the AR experience I realized how unreliable he was because his vision and hearing were distorted.” Similarly, a CG participant reflected, “Discussing the text with my group helped me understand that the narrator’s words could not be taken at face value.” These examples illustrate recurring patterns of critical interpretation across both groups, demonstrating saturation in the themes identified.

In addition to these inferential and analytical approaches, affective participation emerged as a distinct dimension in participants’ justifications. Emotional_Engagement was more prevalent in the EG (21.11% at post-test) than in the CG (17.64%), though both groups showed increases from pre-test levels. EG responses frequently invoked narrative immersion and symbolic association. One student reflected: “The eye of a vulture works as a symbol for fear and dread… just like ravens and crows since they feed on corpses.” This response demonstrates how EG students connect literary imagery with emotional meaning.

Post-intervention reflections further differentiated the groups. CG participants tended to retain a more analytical focus, while EG responses emphasized experiential immersion. One CG student wrote, “I have to read slowly to understand the text, and I need to highlight ideas… it is very important to work on the text to understand it.” In contrast, an EG participant commented: “The AR experience helped me to visualize better the scene… once you see it, you have a visual reference to it, and it helps you to understand better the story.” These shifts reflect how narrative visualization, particularly when mediated by sensory scaffolds, can modulate the inferential process.

While both groups engaged substantively with inference-making, the EG displayed greater integration of psychological and emotional dimensions. These shifts in interpretive focus were framed experientially rather than analytically, suggesting a cognitive reorientation toward immersive scene construction and the emotional resonance of symbolic imagery.

Among the 30 participants in the experimental group, 11 shifted from a correct to an incorrect interpretation of the heartbeat’s symbolic meaning after the AR experience, while 14 maintained consistent correct interpretations across both tests. The remaining 5 participants remained incorrect on both occasions and provided mixed responses. This variation indicates that the AR intervention did not have a uniform effect: while some students demonstrated deeper symbolic understanding, others experienced interpretive misalignment prompted by the multimodal cues. These differences reflect the complexity of individual cognitive processing in multimodal reading environments and underscore the need for scaffolding strategies that can support diverse learner responses.

Case Study: Multimodal Interference in Question 2

A focused case study centered on Question 2 of the inferential comprehension test, which required students to identify the symbolic meaning of the narrator’s auditory hallucination. In the EG, correct responses dropped from 97% at pre-test to 70% at post-test, while CG scores on the same item increased from 71% to 94%. To better understand this decline, we examined the content of incorrect EG responses. Qualitative analysis revealed that five students—who had previously answered correctly—now cited the AR clock as the literal source of the sound. One participant wrote, “Maybe the reason behind is his guilt but the thing he is hearing in that moment is a clock in the old man’s room.” Another added, “His paranoia makes him mistake the sound of the clock for the old man’s heartbeat.” A particularly illustrative response merged sensory and narrative interpretation: “After seeing the AR, it can also be the sound of the clock what is haunting him and he is feeling guilty so that he think that the old man is nagging him.” This comment integrates auditory cues from the AR interface with the narrator’s psychological state, indicating active meaning-making, though the interpretation diverged from the intended symbolic reading.

Although the AR intervention’s ticking clock and animated wall clock were intended to reinforce Poe’s themes of time and inevitability, this multimodal feature redirected some learners’ attention from symbolic interpretation to literal causality. In these cases, the clock was construed as the acoustic source of the hallucination, rather than as an external representation of internal guilt. This interpretive shift led to incorrect responses, despite sustained engagement with the text and its multimodal scaffolds.

However, not all learners were affected. One EG participant demonstrated critical awareness of the extratextual element: “I think it did not change my approach that much. In the case of Poe, a lot of details are left to the reader’s interpretation and with AR we are assuming things that Poe might have not meant, like the clock on the wall.” This variation in response reveals how some learners engaged with paratextual elements selectively or even skeptically and exemplifies the nuanced interplay between learner interpretation and the visual scaffolding within the AR environment.

4. Discussion

This pilot explored the pedagogical and cognitive impacts of augmented reality (AR)-based scaffolds within literary instruction. Situated in a cognitivist theoretical framework, the premise was that immersive digital prompts might improve inferential comprehension. However, technical instability compromised intervention fidelity for most participants. As such, conclusions about AR’s comparative efficacy remain provisional. Findings are therefore treated as generative: they raise new questions for AR-supported learning and identify priorities for future research.

4.1. RQ1: Quantitative Patterns and Design Fragility

The experimental group (EG), which engaged with AR-enhanced literary materials, demonstrated a modest decline in post-test inferential comprehension. The control group (CG), taught using traditional print-based methods, showed a slight improvement. These results challenge assumptions about AR’s inherent educational benefit and suggest its effectiveness is highly sensitive to implementation conditions.

One theoretical explanation for the EG’s performance decline can be found in Cognitive Load Theory (Sweller et al., 2011), which distinguishes between intrinsic, extraneous, and germane cognitive load. When learners are presented with cues that are ambiguous, redundant, or poorly integrated with the task, extraneous load increases and leaves fewer working memory resources available for germane processing—such as linking textual details to prior knowledge to form inferences. In parallel, Mayer’s (2009) principle of spatial and temporal contiguity in multimedia learning holds that verbal and visual information should be presented in close proximity and with clear conceptual alignment to support schema construction. In this study, several AR features, while conceptually linked to narrative elements, were subject to technical delays and minor desynchronization between audio and visual streams. These disruptions likely impaired contiguity, making it harder for learners to integrate multimodal cues with the unfolding textual representation. The most salient case was the 3D animated clock, which, though designed to reinforce narrative time and suspense, became a perceptual focal point that competed with the symbolic heartbeat motif.

This effect was most evident in post-test item Q2, which assessed the symbolic meaning of the narrator’s auditory hallucination. The pattern of responses indicates that a specific multimodal feature redirected interpretive focus. Viewed through the lens of schema theory (R. C. Anderson & Pearson, 1984), the animated clock may have disrupted the coherence of learners’ mental models. Rather than linking the heartbeat to the narrator’s guilt, some students anchored their interpretation to the more perceptually salient, but narratively peripheral, clock element. This represents a form of schema misalignment, in which attention to an extraneous but vivid cue interferes with the intended construction of symbolic meaning.

While these findings point to limitations in the AR experience, they also clarify its cognitive implications: AR’s potential to enrich literary interpretation depends on its capacity to support, rather than fragment or distort, symbolic reasoning. When design elements overload working memory or compete with textual cues, even advanced learners may experience degraded comprehension. Understanding how learners themselves experienced and interpreted these disruptions offers further insight into the mechanisms behind these outcomes.

4.2. RQ2: Learner Experience and Meta-Cognitive Insight

Analysis of learner-reported data revealed three distinct interpretive orientations in the EG, each shedding light on the comprehension patterns described above. These orientations were: (1) interpretive misdirection linked to a salient AR cue, (2) signs of metacognitive awareness and productive struggle, and (3) a symbolic and affective engagement style distinct from the CG’s more analytic approach.

4.2.1. Orientation 1: Interpretative Misdirection

A prominent pattern emerged in EG participants’ post-test justifications for item Q2: the clock motif. While only one student explicitly mentioned the AR experience as the source of their interpretation, five who shifted from correct to incorrect responses on this item referred to the clock as narratively significant. As one participant explained, “Maybe the reason behind is his guilt but the thing he is hearing in that moment is a clock in the old man’s room.” This form of misdirection exemplifies construct-irrelevant activation—where perceptually salient but pedagogically marginal elements draw attention away from core inferential targets. While the AR clock was designed to scaffold interpretive inference, its salience and ambiguous narrative role, compounded by fragmented AR delivery, likely intensified cognitive load and diverted interpretive focus. The resulting disfluency appeared to remain unresolved for many learners within the compressed instructional window, leading to quantifiable performance dips.

4.2.2. Orientation 2: Metacognitive Awareness and Productive Struggle

Not all learners responded passively to these disruptions. A smaller subset of participants demonstrated early signs of meta-cognitive awareness. One noted, “With AR we are assuming things that Poe might not have meant, like the clock on the wall,” signaling critical distance from the AR content and reflective consideration of authorial intent. These reflections indicate nascent forms of productive struggle, wherein learners grapple with multimodal ambiguity in ways that foster interpretive monitoring. Theoretical work on desirable difficulties (Bjork & Bjork, 2011) and disfluency (D’Mello & Graesser, 2012) suggests that such moments can catalyze deeper learning if accompanied by adequate scaffolding. In the present study, the brevity of the intervention precluded opportunities to resolve this struggle; nevertheless, the emergence of meta-awareness amid confusion offers a promising signal for future designs.

4.2.3. Orientation 3: Symbolic and Affective Immersion

Beyond the clock episode, the overall mode of engagement reported by EG participants differed sharply from that of the CG. While CG participants tended to emphasize explicit comprehension strategies, EG responses more frequently foreground emotional resonance, psychological modeling, and symbolic insight. One student described how “the sound was really disturbing, so it kinda made me feel as if I was the mad man going crazy with guilt and paranoia.” This tendency toward immersive, effectively charged engagement can be interpreted through the lens of cognitive resource allocation. Sensory-rich AR cues often encourage learners to elaborate schemas by linking textual details with vivid mental imagery (Paivio, 1986), which can deepen symbolic interpretation. However, this form of elaboration also increases processing demands; without explicit scaffolding to connect these impressions back to textual evidence, working memory resources are diverted from analytic monitoring (Sweller, 1988; Mayer, 2009). In this sense, AR’s multimodal scaffolds do not simply support inference; they can also change how readers build their interpretations. Instead of relying on step-by-step analytical reasoning, some learners focus more on experiential engagement and symbolic elaboration. In the present study, this shift toward symbolic and experiential engagement did not yield short-term gains in inferential comprehension, but it may represent a form of meaning-making that, with targeted scaffolding, could strengthen comprehension over time.

4.3. Integrative Insights and Implications

Taken together, the findings from RQ1 and RQ2 point to a spectrum of responses shaped by the affordances and disruptions of AR. Misdirection, reflection, and reorientation coexisted within the same instructional event. By reconceptualizing this interpretive disfluency as a potential learning catalyst rather than simply a failure point, this study contributes to theory at the intersection of digital multimodality and literacy pedagogy. While misalignment and instability led to measurable declines in comprehension, these same conditions also prompted, in a minority of learners, critical self-reflection and deeper narrative engagement—outcomes that educational models aspire to cultivate (L. W. Anderson & Krathwohl, 2001). The novelty of this framing becomes clearer when considered alongside prior research on AR-supported reading.

In contrast to our results, the few studies examining AR-supported reading with older EFL learners have reported comprehension gains (Asadi & Ebadi, 2025; Ebadi & Ashrafabadi, 2022). Those longitudinal interventions, using broader comprehension measures, showcase AR’s potential to boost learning over time. Our single-session design, assessed with a targeted inferential comprehension instrument, points in a different direction—one that draws attention to interpretive tensions, technical imperfections, and the continued value of traditional instruction. Together, these complementary findings suggest the potential value of integrating the sustained benefits seen in longitudinal studies with the process-focused insights from shorter, more intensive designs.

This study has several limitations. The results are specific to a single, short-term intervention in a particular instructional setting. This limits the ability to determine whether observed patterns—particularly the shift toward symbolic and affective engagement—would consolidate into measurable comprehension gains over time. Additionally, technical disruptions, while integral to the study’s theoretical contribution, may have masked potential benefits of AR under optimal delivery conditions. To meet these challenges and advance the field, we recommend:

• Technical robustness: Pilot immersive tools under classroom conditions, with contingency plans for asset failure.
• Design for cognitive alignment: Map AR cues to learning targets; sequence multimodal prompts incrementally; minimize thematically irrelevant features.
• Pedagogical scaffolding for struggle: Equip instructors with strategies to recognize, pause, and resolve productive disfluency as it arises.
• Process-oriented, mixed-methods research: Employ design-based cycles combining quantitative and qualitative metrics to track immediate and longitudinal effects.

These steps align with the broader insight that interpretive disfluency—provoked by technology—can become a meaningful site of inquiry, design innovation, and pedagogical growth. Ongoing research should track when and how learners’ trajectories of confusion, engagement, and reflection are best supported. The limited effect of AR observed in this study suggests that the pedagogical value of immersive tools depends not only on their novelty but also on their alignment with instructional goals and cognitive processes. AR should not be assumed to enhance comprehension automatically; rather, it requires careful integration with traditional strategies, explicit scaffolding, and sufficient time for learners to adapt to multimodal cues. When these conditions are not met, AR may function as an engaging supplement but fail to produce measurable comprehension gains.

5. Conclusions

This study investigated the impact of AR-enhanced literary materials on the inferential comprehension of advanced EFL learners in higher education. While the intervention did not yield short-term gains in inferential comprehension, the qualitative findings revealed a more complex picture: AR heightened emotional and symbolic engagement for some learners while simultaneously introducing interpretive misalignments for others.

These results underscore a central tension identified in prior research: immersive technologies can enrich comprehension through multimodal elaboration (Paivio, 1986), but they can also disrupt it when design flaws increase extraneous cognitive load (Sweller et al., 2011) or impair multimedia contiguity (Mayer, 2009). Rather than treating such disruptions as failures, we frame them as potential moments of productive struggle (Bjork & Bjork, 2011; D’Mello & Graesser, 2012), in which learners confront ambiguity, challenge assumptions, and engage in metacognitive reflection.

From this perspective, AR’s instructional value extends beyond guiding textual interpretation to encompass the surfacing of interpretive conflicts that can be leveraged through timely scaffolding, consistent with pedagogical taxonomies of higher-order learning (L. W. Anderson & Krathwohl, 2001). By recognizing and working with these tensions, educators can transform technological imperfections and narrative disfluencies into catalysts for deeper cognitive engagement and critical literacy. These insights complement prior longitudinal findings showing AR’s potential for sustained comprehension gains (Asadi & Ebadi, 2025; Ebadi & Ashrafabadi, 2022) and highlight the importance of integrating process-focused analysis with robust design and implementation strategies.

Nevertheless, this study’s short-term design, the technical instability of the AR implementation, and the restricted focus on inferential comprehension constitute notable limitations. Future research should, therefore, adopt longitudinal and mixed-methods approaches to capture long-term effects, explore design-based improvements to align AR cues with pedagogical targets, and investigate how teachers can scaffold moments of interpretive disfluency to transform them into opportunities for deeper comprehension and critical literacy.