Visual Literacy in Teacher Education: Narrative Reconstruction and AI Mediation in Warja Lavater’s Cendrillon

Abstract

This study examines the development of visual literacy in teacher education through a classroom experience based on Cendrillon, the coded picturebook by Warja Lavater. The activity required students to reconstruct the sequence of the book’s twenty scenes, assign each scene to a narrative phase, provide a verbal description, and generate a new version of the story using generative artificial intelligence. The participants were 234 students from undergraduate teacher education degrees and the Master’s Degree in Educational Research, organized into 52 subgroups and classified according to their reading habits and self-reported familiarity with traditional fairy tales. Data were analyzed using indicators of sequential reconstruction, distribution of narrative phases, verbal description, and AI-mediated narrative reformulation. The results showed that, descriptively, groups with stronger reading habits achieved higher overall scores in the sequential reconstruction task; however, these differences did not reach statistical significance in the exploratory inferential models, and self-reported familiarity with traditional folktales alone did not guarantee greater accuracy. Narrative phase assignment was predominantly concentrated in the development of the story, while verbal descriptions followed a different pattern from the sequential ordering. The AI-generated stories retained part of the students’ original vocabulary but incorporated a greater proportion of newly generated lexical items. The findings suggest that coded picturebooks provide a valuable framework for examining visual literacy as a process involving narrative reconstruction, verbal transduction, and critical engagement with generative artificial intelligence.

1. Introduction

The study presented here is based on a classroom experience conducted in several courses taught at the Faculty of Education of the University of Alicante, using a particular genre of picturebook: the coded picturebook. The instructional proposal combined work on traditional folktales, the sequencing of visual narratives, creative writing, and, finally, mediation through generative artificial intelligence (GenAI). The main objective was to examine how pre-service teachers interpreted the sequence of images in one of Warja Lavater’s imageries, how they verbally described the action depicted in each scene, and how they reflected on the role of GenAI in generating a version of a traditional folktale based on their own descriptions.

The instructional design was based on the twenty visual sequences that make up Cendrillon, Warja Lavater’s leporello (accordion-fold book), originally published in 1976 by the Parisian publisher Adrien Maeght. Each sequence, available in digital format through the Broward County Library Digital Archives (https://digitalarchives.broward.org/digital/collection/Fontaneda/id/688/rec/2 (accessed on 12 July 2026)), was laminated and assembled into twelve sets of cards, which were distributed among groups specifically formed for this activity in each course. The primary aim was to examine whether students’ reading habits and their familiarity with traditional folktales influenced their ability to reconstruct the correct sequence of images, assign each scene to the narrative functions proposed by Paul Larivaille, and produce rich and accurate verbal descriptions.

In this study, visual literacy constitutes the central focus, while narrative reconstruction, reading habits, familiarity with traditional folktales, and GenAI-assisted writing are examined as interconnected dimensions of the same interpretive framework. Narrative reconstruction is conceived as an indicator of participants’ ability to decode and organize visual narratives; reading habits and familiarity with traditional folktales are considered potential factors influencing this interpretive process; and GenAI-assisted story generation is analyzed as a subsequent stage of verbal transduction, through which visual interpretations are transformed into coherent written narratives. On this basis, the study was guided by the following research questions:

RQ1. To what extent are students’ reading habits and familiarity with traditional folktales associated with the reconstruction of the visual sequence of Warja Lavater’s Cendrillon?

RQ2. How do the participant profiles differ in their assignment of narrative phases and in the verbal elaboration of their descriptions of the visual scenes?

RQ3. How does GenAI mediate the transformation of students’ visual descriptions into a coherent narrative, and what kinds of lexical and narrative reformulation does it introduce?

The following sections present the theoretical framework underpinning the concepts of visual literacy, coded picturebooks, and GenAI-mediated learning. The methodology employed in the classroom experience is then described in detail, followed by the presentation and discussion of the main findings.

2. Theoretical Framework

2.1. Visual Literacy and Multimodal Reading

The implications of the concept of visual literacy, first coined by John Debes in the late 1960s (Arizpe & Styles, 2016), have become increasingly broad and significant. The shared visual culture that transcends age groups and cultural backgrounds is a clear consequence of globalization. While this common visual repertoire provides considerable opportunities through a shared understanding of narrative conventions and the communicative functions of images (Arizpe et al., 2014), it also requires a specific form of literacy that enables individuals to decode, analyze, and interpret the information conveyed visually (Gomes-Franco-e-Silva, 2019; Nodelman, 2017). Although visual communication may appear to operate with a certain degree of “transparency,” the reading of images requires explicit knowledge of the implicit visual competencies that readers already possess in order to articulate and reflect upon them (Kress & van Leeuwen, 2020). In this regard, much of visual communication depends on an understanding of the visual codes that underpin the transmission of meaning (Kümmerling-Meibauer & Meibauer, 2013). Picturebooks constitute a representative example of these contemporary multimodal texts, demanding visual literacy that extends beyond recognizing the relationship between words and images to include an understanding of visual codes such as figure–ground distinction, the functions of line and colour, perspective, composition, and other graphic conventions (Kümmerling-Meibauer & Meibauer, 2013).

Meaning-making in picturebooks arises from the dynamic interplay of multiple semiotic resources, each of which contributes distinct affordances that readers must recognize and interpret to construct coherent meaning (Pantaleo, 2020). Although the present study focuses primarily on the visual mode, other semiotic resources (including verbal language, narrative sequencing, and the materiality of the book) also play an essential role in this process. While each mode can be examined independently according to its own representational resources (Lewis, 2006; Pantaleo, 2020), a comprehensive interpretation of a picturebook requires an integrated reading that considers how these modes interact both within individual visual sequences and across the narrative as a whole (Serafini & Reid, 2024). Sipe (1998) emphasized the complementary narrative functions of the visual and verbal modes, arguing that images primarily organize spatial relationships, whereas written language structures narrative temporality. However, the interplay of these semiotic systems alone is insufficient to generate meaning. Rather, meaning emerges through the transactional relationship between the multimodal text and the reader, whose cultural knowledge, prior experiences, and intertextual understandings actively shape the interpretive process. It is within this dialogic space that the meanings of the picturebook are negotiated and ultimately constructed (Lewis, 2006).

From this perspective, the relationship between text and image cannot be understood as a straightforward correspondence between two semiotic systems that merely coexist by explaining one another. Rather, it should be conceived as a dynamic interaction in which each mode complements, extends, qualifies, or even challenges the meanings conveyed by the other. This interdependence also gives rise to interpretive gaps that require readers to engage in active inferential reasoning. Consequently, picturebook reading demands sustained attention to meanings that are implied rather than explicitly represented (Albers, 2008). Interpretation therefore relies on readers’ ability to identify the visual and verbal cues that shape the interaction between modes and to formulate hypotheses about the complex ways in which these multimodal relationships generate meaning.

This distinctive nature of the picturebook has led several scholars to emphasize the slow and deliberate reading it requires (Arizpe et al., 2014; Pantaleo, 2020). Without careful attention to the details, variations, and interrelationships embedded within the visual and verbal narrative, readers cannot achieve the holistic understanding that these multimodal texts demand. Accordingly, picturebook reading involves two complementary interpretive processes. The first concerns the construction of the narrative’s referential meaning; identifying what happens, how events are connected, and how characters and situations evolve throughout the story. The second entails the reader’s subjective meaning-making, encompassing the associations, hypotheses, and inferences elicited through engagement with the text (Martínez-Carratalá et al., 2024). This interpretive dimension becomes particularly significant when images themselves call for verbal articulation (Arizpe et al., 2014; Nodelman, 2010), requiring readers to organize, narrate, and verbalize what is communicated visually. Such a process is especially evident in wordless picturebooks, where visual interpretation assumes the primary role in constructing the narrative.

These interpretive demands become even more pronounced in the case of the wordless picturebook, where readers are required to reconstruct an implicit verbal narrative from visual representations alone (Kümmerling-Meibauer & Meibauer, 2013). Rather than simplifying the reading process, the absence of written text places greater interpretive demands on readers (Martínez-Carratalá & Rovira-Collado, 2022), who must reconstruct the narrative by interpreting a complex visual code. This process involves identifying relationships between scenes and sequences, as well as interpreting spatial arrangements, compositional variations, and other visual features that guide narrative comprehension (Painter et al., 2014). It also requires readers to draw upon their cultural knowledge and intertextual experiences (Arizpe et al., 2014) to complete a narrative that unfolds through a visual discourse offering only partial guidance (Beckett, 2014). This active process of narrative reconstruction has led scholars to identify wordless picturebooks as particularly well suited to the collaborative construction of meaning (Arizpe et al., 2014). Within this context, shared discussion enables readers to compare interpretations, negotiate inferences, and make their meaning-making processes explicit.

Wordless picturebooks further foreground the process of transduction (Pantaleo, 2020), insofar as they require readers to transform meanings constructed in the visual mode into verbal representations of an implicitly unfolding narrative. This process demands that readers establish causal relationships, discriminate among salient visual cues, and hierarchize semiotic resources in order to produce a coherent verbal narrative. Consequently, meaning is not simply retrieved from the visual text but actively reconstructed through the translation of information across semiotic modes. Such transductive activity underpins the considerable educational value of wordless picturebooks across different educational contexts (Martínez-Carratalá et al., 2024), since the verbal reconstruction of visual sequences can be adapted to readers’ levels of literary competence, making these works equally suitable for novice and expert readers alike. Furthermore, many of these picturebooks incorporate interpretive scaffolds, such as maps, legends, or visual keys, that facilitate readers’ navigation of increasingly complex visual systems (Kümmerling-Meibauer & Meibauer, 2013). This is particularly evident in the coded picturebook, a format closely related to the artist’s book (Beckett, 2014), in which narrative meaning emerges through the interplay of symbolic signs, spatial organization, and culturally shared visual conventions.

2.2. Coded Picturebooks and Warja Lavater’s Work

The narrative and affective potential of visual forms, colour, and composition, discussed in the previous section, provides a useful starting point for understanding the distinctive characteristics of the coded picturebook. This format is based on the premise that meaning in picturebooks does not necessarily depend on figurative representation. Instead, meaning can be constructed through the arrangement of simple geometric forms, such as circles, squares, and triangles, as well as through chromatic contrasts and the spatial organization of these minimal visual elements. From this perspective, even subtle variations in shape, colour, or composition can produce significant differences in readers’ emotional and narrative responses, as demonstrated by Bang (2023). The effectiveness of geometric forms in early childhood has been associated with children’s sensitivity to symmetry, contrast, and basic shapes. Pankenier Weld (2015) argues that these visual preferences constitute an early visual grammar that was subsequently explored by the Russian avant-garde, whose artists recognized the expressive and communicative potential of elementary geometric forms.

The coded picturebook can therefore be understood as a distinctive form of visual narrative: a wordless—or almost wordless—picturebook composed of symbolic signs and accompanied by an introductory visual key that establishes the correspondence between those signs and the characters, settings, and objects that constitute the narrative world (Hamaide-Jager, 2019). This format demonstrates that images do not merely support verbal comprehension but operate through their own semiotic system, which readers must actively decode and reconstruct. Consequently, engaging with coded picturebooks requires a form of visual literacy that extends beyond image recognition to encompass the interpretation of semiotic systems, recurring visual patterns, compositional variations, and internal relationships through which meaning is constructed. Although these narratives contain little or no written text beyond the initial visual key, their interpretation depends on the existence of a recognizable hypotext, signalled by both the title of the picturebook and the identification of its characters. Consequently, coded picturebooks function as rewritings of traditional folktales rooted in the oral tradition (Beckett, 2015). Readers therefore rely not only on the visual code presented in the picturebook but also on their intertextual knowledge to reconstruct the underlying narrative. Narrative meaning thus remains implicit within the visual sequence, and it is through the interaction between the visual code and the hypotext that readers actively construct meaning.

Strictly speaking, the corpus of coded picturebooks remains relatively small, although broader classifications have been proposed that encompass a wider range of visually encoded narratives. Nevertheless, the work of the Swiss artist Warja Lavater occupies a central position within this format, both because of its pioneering character and the lasting influence of its visual language. Trained in graphic design in Zurich under the influence of Ernst Keller, the Bauhaus, and the Swiss Style, Lavater developed a visual esthetic grounded in the communicative potential of typography, composition, and elementary geometric forms (Beckett, 2012). Her stay in New York between 1958 and 1960 further strengthened her interest in public visual communication systems, including signage, pictograms, and graphic symbols associated with urban environments (García Blanco, 2022; Meunier, 2013). These influences culminated in the creation of her Folded Stories, also known as her Imageries, which reconceptualized the leporello as an artistic and narrative medium.

A defining characteristic of the coded picturebook, and particularly of Lavater’s work, is that it does not simply translate a verbal narrative into a visual system, nor does it merely replace recognizable figurative representations with abstract forms. Rather, it establishes a system of visual codes that reinterprets the hypotext through a distinct semiotic framework. Meaning emerges from pictograms capable of conveying actions, movement, transformation, and emotional states. Although the relationship between the visual signs and the elements of the source narrative is initially established through the visual key, Lavater’s reinterpretation extends far beyond this one-to-one correspondence. Instead, readers are invited to construct a second-order interpretation by actively engaging with the semiotic system itself. Consequently, the narrative functions as a rewriting of the original tale, in which meaning is co-constructed through the interaction between the visual code and the reader’s intertextual knowledge. It is the reader’s active engagement with the proposed visual code that ultimately transforms the system of pictograms into a coherent verbal narrative.

Although the classroom activity described in this study required dismantling the accordion-fold structure of the book, it is important to acknowledge that Lavater’s artistic conception is fundamentally grounded in its material form. The leporello allows the narrative to unfold along a continuous horizontal axis, transforming the story into a spatial journey that can be apprehended in its entirety at a single glance. Consequently, Lavater’s Imageries support two complementary modes of reading: a sequential reading that follows the progression of individual scenes and a global reading made possible when the leporello is fully unfolded. In this configuration, the continuous strip of paper functions simultaneously as a temporal axis, guiding the narrative sequence, and as a unified visual surface that invites readers to perceive the work as a whole.

Another defining characteristic of Lavater’s work is its close adherence to the structural organization of the hypotext, a feature that aligns her visual narratives with the narrative models developed within the formalist tradition. Accordingly, the stories preserve the canonical narrative progression of initial situation, conflict, development, resolution, and final situation. In a further connection with formalist narratology, Lavater’s work also reflects the distinction between fabula (story) and syuzhet (plot). Fabula refers to the sequence of events that constitutes the source narrative, whereas syuzhet denotes the way in which those events are selected, organized, and reconfigured within the visual narrative.

This section concludes by emphasizing the pivotal role of the reader in constructing meaning from these Imageries. Narratives composed through pictograms do not give rise to a single verbal realization of the hypotext; rather, each reading reconstructs the narrative in light of the reader’s prior knowledge, cultural background, and personal experience. In this respect, Lavater’s Imageries preserve the oral and variable nature of traditional folktales, as each reading generates a new reconstruction of the story through the interpretation of its visual signs (Montanaro Staples, 2016). The experimental nature of Lavater’s work also situates it within the tradition of the artist’s book, as reflected both in its formal characteristics and in its publication history through institutions such as the Museum of Modern Art (MoMA) and Adrien Maeght (Beckett, 2013). At the same time, these works have been appropriated by child readers through playful exploration and interpretive experimentation, as Lavater herself observed (Gromer & Lavater, 1991). In the case of Cendrillon, discussed in the following sections, the greater complexity of its coding system, compared with Lavater’s earlier works, further intensifies readers’ interpretive engagement (Beckett, 2010), making it a particularly valuable resource for investigating visual literacy. Its educational value lies precisely in the demands it places on readers to observe, infer, and verbally articulate the narrative encoded within the visual sequence. This pedagogical potential provides the rationale for the present study, in which visual reading is conceptualized as a process of narrative mediation through which meanings constructed in the visual mode are transduced into verbal discourse. Within this framework, generative artificial intelligence (GenAI) is examined as a mediating tool to support this process of verbalization.

2.3. Generative AI as a Tool for Narrative Mediation

The integration of generative artificial intelligence (GenAI) into educational contexts has prompted growing interest in its potential to support innovative teaching practices, while also generating substantial research and critical debate regarding its role in creative and academic writing. Current scholarship consistently argues that the educational value of GenAI should not be understood in terms of the immediate production of texts, but rather in its capacity to support the successive stages of the writing process. In the specific context of creative writing, this process involves generating ideas, selecting and transforming narrative elements, and organizing them into a coherent narrative through the interconnected phases of planning, drafting, and revision (Onbasili, 2026; Oruç et al., 2026). Within this framework, GenAI can support writers by facilitating idea generation, character development, plot construction, and the improvement of narrative coherence (Kızıltaş, 2025). Among its most frequently reported educational benefits are increased learner motivation and writing self-confidence, as well as a reduction in the anxiety and creative blocks that often accompany creative writing tasks (Behrens et al., 2025; Oruç et al., 2026).

Nevertheless, the benefits associated with GenAI are frequently accompanied by important pedagogical concerns, some of which appear to contradict the advantages reported in the literature (Bruch & Brenden, 2026; Niloy et al., 2024). For example, AI-assisted writing may become superficial when expressive and rhetorical decision-making is delegated to technology. Previous research suggests that, although GenAI can improve the organization, fluency, and presentation of written texts, its use may ultimately constrain creativity if the processes of idea generation, revision, and critical reflection are transferred from the writer to the AI system. Another frequently discussed concern is the risk of shifting the educational focus from the development of writing competence to the acquisition of technical proficiency in using AI tools (Botturi et al., 2025). Consequently, the educational value of GenAI should be assessed not primarily in terms of the texts it produces, but rather according to the cognitive and pedagogical processes it supports throughout the writing process.

These considerations make it essential to situate the use of GenAI within an explicit and carefully designed pedagogical framework. Its integration into educational practice should not rely on the spontaneous use of AI tools but rather be embedded within instructional designs that clearly define learning objectives, learning activities, and the pedagogical role of the teacher (Jiang et al., 2026). To ensure that GenAI contributes meaningfully to learning, students must possess sufficient prior knowledge of the subject matter and receive appropriate guidance to formulate effective prompts and critically evaluate the relevance and quality of AI-generated responses (Borge et al., 2024; Rovira-Collado et al., 2025). These considerations place teachers at the centre of the educational process. Their role extends beyond facilitating access to the technology to mediating learning, fostering critical reflection, and supporting students in making informed pedagogical and communicative decisions throughout the writing process (Kızıltaş, 2025; Muñoz Martínez et al., 2025).

The integration of GenAI into educational settings also requires a broader understanding of literacy that extends beyond conventional notions of reading and writing as exclusively verbal practices. The mediation of algorithmic technologies highlights that literacy also involves the selection, orchestration, and transformation of multiple semiotic resources operating within specific communicative contexts (Kladaki et al., 2025; Lim, 2026; Mishra, 2025). This perspective is particularly relevant to the instructional activity presented in this study, in which students are required to transform an abstract visual code into a coherent verbal narrative. Although GenAI can mediate this process, its effective use depends on students’ ability to formulate appropriate prompts, identify potential biases, omissions, or distortions in AI-generated outputs, critically evaluate the resulting texts, and recognize that these technologies are not neutral but are shaped by the social practices through which meaning is constructed (Burriss & Leander, 2024; Thomson et al., 2025). Conceived in this way, GenAI has the potential to foster metacognitive reflection on both visual reading and narrative writing while making explicit the processes through which visual interpretation is transduced into verbal discourse. Accordingly, the classroom experience based on Lavater’s Cendrillon was designed as a sequence of visual interpretation, verbalization, AI-assisted story generation, and critical revision, with the aim of examining how pre-service teachers interpret and narratively reconstruct the meanings conveyed by a coded picturebook.

3. Materials and Method

3.1. Previous Educational Approaches to Lavater’s Work and Contribution of the Present Study

Although relatively few educational experiences have been reported using Warja Lavater’s Imageries, the existing literature provides a useful context for situating the instructional proposal developed in the present study. Meunier (2013) notes that the presentation of Le Petit Chaperon Rouge during a children’s workshop at the Centre Georges Pompidou in 1976 generated considerable interest among participants, who were particularly attracted by the possibility of “playing” with the book. Although this observation is largely anecdotal, it highlights several characteristics of Lavater’s work that make it especially valuable in educational settings, including its materiality as a manipulable book object, its capacity to encourage visual exploration, and the active reconstruction of narrative meaning.

More systematic educational approaches have been reported by Plath and Richter (2017) and Hamaide-Jager (2019). Plath and Richter’s (2017) instructional proposal, based on Blanche Neige, demonstrates how visual narratives can serve as a gateway to more complex literary texts. Designed for primary school students, the intervention aims to foster learners’ ability to progressively decode the visual code, develop their own visual key, and produce oral and dramatized retellings of the story. Hamaide-Jager (2019), in turn, describes a classroom experience with pre-service teachers that explores literary genres through coded picturebooks, including Cendrillon. The activity requires participants to collaboratively construct an interpretation of the visual narrative. The findings highlight both the challenges posed by interpreting abstract visual representations and the crucial role of the hypotext in supporting narrative reconstruction.

The present study builds upon these previous contributions while introducing several distinctive features. First, it was conducted within the context of teacher education and involved students enrolled in different university courses. Second, participants were profiled in advance through a questionnaire assessing their reading habits, familiarity with traditional folktales, and perceptions of these narratives. Based on these data, students were organized into four analytical groups, enabling the study to examine the potential influence of these variables on the interpretation of the coded picturebook. Third, the leporello was transformed into a manipulative learning resource by converting its twenty visual sequences into laminated cards, allowing participants to collaboratively reconstruct and sequence the narrative. Finally, the instructional design incorporated a GenAI-based web application through which each group recorded the sequence of the scenes, identified their corresponding narrative phases, and produced verbal descriptions of each visual sequence. These descriptions were subsequently used to generate a complete narrative with the support of GenAI. Consequently, the study extends beyond examining students’ ability to decode a visual semiotic system by investigating the transition from visual reading to verbalization and AI-mediated narrative construction. The objective was not merely to determine whether participants could accurately reconstruct the narrative sequence of Cendrillon, but also to examine how they interpreted a coded visual narrative, recognized and related its visual signs, activated the hypotext, collaboratively negotiated the narrative sequence, and ultimately transduced visual meaning into verbal discourse. From this perspective, the instructional design was conceived as a multimodal reading task in which participants’ understanding of the picturebook was reflected in both the organization of the narrative sequence and the quality of their verbal descriptions.

3.2. Participants and Educational Context

The study was conducted at the Faculty of Education of the University of Alicante in three university courses related to language and literature education. Participants were enrolled in Teaching Spanish Language and Literature in Early Childhood Education, part of the Bachelor’s Degree in Early Childhood Education; Teaching Reading and Writing, part of the Bachelor’s Degree in Primary Education; and Research on the Development of Literary Literacy, Children’s and Young Adult Literature, and the Literary Canon, part of the Master’s Degree in Educational Research. The final sample comprised 234 students distributed across five class groups: two from the Bachelor’s Degree in Early Childhood Education, two from the Bachelor’s Degree in Primary Education, and one from the Master’s Degree in Educational Research.

Within each of the four undergraduate class groups, students were organized into 12 working groups. These groups were distributed across the four participant profiles identified through the initial questionnaire, with three working groups assigned to each profile within every undergraduate class. In contrast, because the master’s cohort was considerably smaller, students were organized into four working groups, with one working group representing each participant profile. Overall, the study comprised 52 working groups: 48 from the four undergraduate classes (12 per class) and 4 from the master’s class. This group allocation enabled an exploratory comparison of how different participant profiles approached the same visual reading and narrative reconstruction task. Specifically, the study examined whether students’ reading habits and their familiarity with traditional folktales were associated with differences in their interpretation of a coded picturebook. The inclusion of these participant profiles provided a means of characterizing the sample without altering the collaborative nature of the instructional activity, which was designed as a group-based task centred on discussion and the collective construction of meaning.

3.3. Student Profiling Instrument: Reading Habits and Familiarity with Traditional Folktales

As described in the previous section, prior to the classroom activity participants completed a questionnaire designed to characterize their reading profile and their familiarity with traditional folktales. The questionnaire comprised 10 Likert-type items grouped into two dimensions and scored on an 11-point scale ranging from 0 (strongly disagree) to 10 (strongly agree) (see

Table A1. Initial questionnaire on reading habits, knowledge of traditional folktales, and their educational value.

Dimension	Item
Reading Habits	I regularly devote time to reading literary works in my daily life.
Reading Habits	Reading books gives me pleasure and personal satisfaction.
Reading Habits	During the past year, I have read books on my own initiative on a regular basis (e.g., at least one book every two months).
Reading Habits	I believe that reading literature enhances my critical and creative thinking.
Reading Habits	I prefer to read a literary work before watching its film or audiovisual adaptation.
Traditional Folktales	I can easily recognize and remember traditional folktales such as Cinderella, Little Red Riding Hood, Sleeping Beauty, Hansel and Gretel, Rapunzel, Snow White, and Beauty and the Beast.
Traditional Folktales	I feel confident telling a traditional folktale to a group of children.
Traditional Folktales	In recent years, I have read or listened to traditional folktales beyond my childhood.
Traditional Folktales	I consider the transmission of traditional folktales to be important in early childhood and primary education.
Traditional Folktales	I believe that traditional folktales are a valuable resource for teaching cultural values and heritage in the classroom.

). The first dimension assessed students’ reading habits and their attitudes towards literary reading, whereas the second focused on their familiarity with, recollection of, and perceived educational value of traditional folktales. The questionnaire was developed with reference to the work of Yubero and Larrañaga (2015), particularly the items related to reading habits, which examine aspects such as reading frequency, voluntary reading, reading enjoyment, and personal appreciation of literature.

Based on participants’ responses, two scores were calculated for each student: one reflecting reading habits and the other assessing familiarity with and perceptions of traditional folktales. The median score for each dimension was then used as the cut-off point to classify participants into high or low categories for both variables. This procedure resulted in four participant profiles: LX, comprising students with high reading habits and high familiarity with traditional folktales; FL, with high reading habits and low familiarity; RD, with low reading habits and high familiarity; and NV, with low reading habits and low familiarity. After this individual classification, students within each class were grouped according to their assigned profile. In the undergraduate classes, three working groups were created for each profile, with group sizes adjusted to between four and six members to preserve profile homogeneity. Consequently, each working group was characterized by the shared individual profile of its members.

The reliability of the two questionnaire dimensions was examined using Cronbach’s alpha. The Reading Habits dimension showed good internal consistency (α = 0.89), while the Familiarity with Traditional Folktales dimension showed moderate internal consistency (α = 0.66), which was considered acceptable for the exploratory profiling purposes of the study. These profiles were used to explore whether students’ reading habits and familiarity with the hypotext were associated with differences in the sequencing of Cendrillon’s visual narrative and in the verbal elaboration of their descriptions. The profiles were therefore treated as pedagogical and analytical groupings derived from the questionnaire scores, rather than as discrete categories inherent to the variables themselves.

3.4. Materials

The instructional materials comprised both physical and digital resources. The physical materials consisted of 12 sets of laminated cards, each containing the 20 visual sequences that comprise Warja Lavater’s Cendrillon leporello. The use of laminated cards depicting the visual sequences was directly aligned with the objectives of the study, as it enabled participants to manipulate the scenes, experiment with alternative sequence arrangements, and collaboratively negotiate the narrative progression. This adaptation transformed the original leporello format, sacrificing the possibility of perceiving the narrative as a continuous visual whole while creating opportunities for participants to construct their own interpretations of the individual scenes. Providing each working group with a complete set of cards also facilitated the organization of participants according to the four profiles identified through the initial questionnaire and ensured the active involvement of all group members throughout the activity.

During the implementation of the instructional activity, the working groups were not provided with the visual key accompanying Cendrillon. This methodological decision was intended to examine how participants interpreted the system of visual codes and identified characters, settings, and objects relying solely on their prior knowledge of the hypotext. In the absence of the visual key, participants were required to pay closer attention to recurring visual patterns, spatial relationships, variations in size and colour, and other visual cues. Rather than establishing direct correspondences between signs and their referents, they were encouraged to formulate and refine interpretive hypotheses through collaborative discussion. Before the group work began, the instructor provided a brief introduction to the activity, including general notions of visual literacy, the functioning of coded picturebooks, and the main characteristics of Warja Lavater’s work. The instructor’s role as mediator was then limited to presenting the task, clarifying the use of the cards and the digital platform, and encouraging groups to justify their interpretations through visual evidence.

The second resource employed in the study, which played a central role in both data collection and AI-assisted narrative generation, was a custom-developed web application. The platform was designed to enable each working group to record the sequence of the visual scenes, assign the corresponding narrative phase, and provide a verbal description of each sequence. In addition, it automatically stored all participant responses and generated the input required for AI-assisted story generation. The application was integrated with the OpenAI API, allowing participants’ descriptions to be incorporated into a predefined prompt that served as the input for generating a complete version of the story. Beyond supporting the instructional activity, the platform functioned as the principal data collection tool by systematically recording the sequence arrangements, narrative phase classifications, scene descriptions, and AI-generated narratives produced by each working group, thereby facilitating the subsequent analysis of the data. Figure 1 presents a screenshot of the web application, while Figure 2 shows students working collaboratively with the laminated cards during the classroom activity.

3.5. Procedure and Data Collection

Data collection was organized around the records generated throughout the instructional activity. All data were obtained through the custom-developed web application used during the intervention. Each working group entered the interpreted sequence of the 20 visual scenes, assigned a narrative phase to each scene, and provided a verbal description of every sequence. These data were captured through a script developed in Google Apps Script, which automatically recorded each submission as a new row in a spreadsheet. In addition to participants’ responses, the system also stored the AI-generated narrative produced from the groups’ descriptions. The stories were generated through the OpenAI Chat Completions API using the GPT-4o model. For each working group, the prompt combined a fixed instruction with the sequence of scenes provided by the students, including the narrative phase and verbal description. The maximum output length was set to 1200 tokens, and no additional generation parameters were manually specified in the script. A single output was generated for each working group, with no alternative versions produced. The complete prompt structure is provided in Appendix B. This procedure ensured that all records were collected in a standardized format and linked to the corresponding working group and participant profile.

Data analysis was structured around four analytical dimensions. First, each group’s reconstructed sequence was compared with the original sequence of Cendrillon. Five indicators were calculated to assess sequencing accuracy. Exact Position captured local accuracy, that is, whether each scene was placed in its original position within Lavater’s sequence. Adjacent Pair measured the recognition of immediate narrative continuity between consecutive scenes. The Longest Increasing Subsequence (LIS) assessed the extent to which participants preserved the global progression of the visual narrative, even when some scenes were displaced. Inversions identified disruptions in sequential order, indicating the degree to which scenes were arranged against the original narrative progression. Finally, the ±1 indicator measured near-position accuracy by identifying scenes placed either in their exact position or within one position of it. Second, the distribution of the narrative phases assigned by each group to the visual scenes was examined, considering the five categories of initial situation, conflict, development, resolution, and final situation. Third, participants’ verbal descriptions of the visual sequences were analyzed using several linguistic indicators, including total word count, relative lexical diversity, number of unique words, and mean sentence length. Finally, the AI-generated stories were compared with the descriptions produced by the participants. The analysis examined the vocabulary shared between both texts, the lexical items introduced by the AI, and the use of discourse connectives in the generated narratives. This comparison was complemented by a brief qualitative examination of selected group productions, aimed at identifying how the AI-generated stories reorganized, connected, or expanded the students’ visual descriptions. Data processing, statistical analyses, and figure production were carried out using RStudio (2026.05.0). GenAI (OpenAI GPT-4o, accessed via the OpenAI API) was used solely as a supplementary tool to review selected patterns identified during the analysis. All findings were subsequently verified by the research team.

Based on this research design and the procedures adopted for data collection and analysis, the following section presents the results obtained for each of the analytical dimensions.

4. Results

The corpus analyzed in the following sections comprised the productions of five class groups: two from course 1DLCLEI26, two from course 4DLE25, and one from course INVLIJUA26. A total of 52 working groups were included in the analysis, together with their corresponding sequence reconstructions, narrative phase classifications, verbal descriptions of the visual sequences, and AI-generated narratives. The distribution across the four participant profiles, defined by the combination of reading habits and familiarity with traditional folktales, was balanced, with 13 working groups assigned to each profile. As described previously, the four profiles were LX (high reading habits/high familiarity with traditional folktales), FL (high reading habits/low familiarity), RD (low reading habits/high familiarity), and NV (low reading habits/low familiarity). This distribution enabled comparisons across the participant profiles established in the study design.

The results are presented according to four analytical dimensions. First, the sequential ordering of the twenty visual scenes is examined to assess the extent to which participants reconstructed the original visual progression of the coded picturebook. Second, the assignment of narrative phases to the visual sequences is analyzed. Third, the verbal descriptions produced by the participants are examined using indicators of text length, lexical diversity, and sentence segmentation. Finally, the AI-generated narratives are analyzed as a form of narrative reformulation based on the students’ descriptions, with attention to lexical preservation, newly generated vocabulary, and discourse connectives.

4.1. Sequential Reconstruction of the Visual Narrative

The first analysis examined the sequence reconstructions produced by the working groups. The aim was to assess the extent to which participants’ interpretations reproduced the original narrative progression of Lavater’s Cendrillon.

Each working group’s reconstructed sequence was compared with the reference ordering of the 20 visual scenes. This comparison yielded five indicators that captured different aspects of sequencing accuracy. First, the number of inversions was calculated as a measure of the overall disruption of the sequence, with higher values indicating greater departures from the original ordering. Second, the Longest Increasing Subsequence (LIS%) measured the length of the longest correctly ordered subsequence, providing an estimate of the extent to which the original sequence was preserved despite isolated ordering errors. Third, the percentage of adjacent pairs assessed the proportion of consecutive scenes that remained in the correct relative order. Fourth, the percentage of exact positions measured the proportion of scenes placed in their original positions. Finally, the ±1% indicator represented the percentage of scenes placed either in their exact position or within one position of the reference sequence.

Table 1. Overall sequencing accuracy indicators by participant profile.

Participant Profile	Inversions	LIS(%)	Adjacent Pairs (%)	Exact Positions (%)	±1 (%)
LX	35.9	66.1	33.6	28.5	53.8
FL	31.1	71.5	39.7	35.4	65
RD	45.1	62.3	31.2	25	48.8
NV	39.5	65.4	31.6	22.3	45.8

presents the overall values of these five indicators for each participant profile.

As shown in Table 1, the FL profile achieved the highest mean values across the principal positive indicators of sequencing accuracy, including Exact Positions (35.4%), Adjacent Pairs (39.7%), LIS (71.5%), and ±1 (65.0%). It also recorded the lowest mean number of inversions (31.1), indicating the closest reconstruction of the original sequence among the four participant profiles. The LX profile occupied an intermediate position, with values generally higher than those of RD and NV, although consistently below those of FL. Considering LIS as a reference indicator, LX reached 66.1%, a value close to that of NV (65.4%) and higher than that of RD (62.3%). This finding suggests that a substantial portion of the original narrative progression was preserved across all participant profiles, despite differences in overall sequencing accuracy. The RD and NV profiles obtained the lowest values on several indicators. NV showed the lowest percentage of exact positions (22.3%) and the lowest ±1 score (45.8%), whereas RD recorded the lowest LIS value (62.3%) together with the highest mean number of inversions (45.1).

To complement this overall comparison,

Table 2. Sequencing accuracy indicators by class group and participant profile.

Class Group	Participant Profile	Inversions	LIS (%)	Adjacent Pairs (%)	Exact Positions (%)	±1 (%)
1DLCLEI26 GR3	LX	9.7	80	50.9	56.7	81.7
1DLCLEI26 GR3	FL	17	78.3	47.4	51.7	85
1DLCLEI26 GR3	RD	51	58.3	17.5	20	50
1DLCLEI26 GR3	NV	54.7	53.3	12.3	6.7	26.7
1DLCLEI26 GR4	LX	32	71.7	36.8	36.7	65
1DLCLEI26 GR4	FL	29.7	71.7	31.6	26.7	63.3
1DLCLEI26 GR4	RD	27.3	71.7	50.9	41.7	61.7
1DLCLEI26 GR4	NV	28.3	75	49.1	43.3	68.3
4DLE25 GR2	LX	56	53.3	15.8	13.3	41.7
4DLE25 GR2	FL	46.7	66.7	43.9	30	53.3
4DLE25 GR2	RD	49	63.3	42.1	23.3	45
4DLE25 GR2	NV	32	70	38.6	23.3	50
4DLE25 GR3	LX	47.7	61.7	31.6	8.3	31.7
4DLE25 GR3	FL	26.3	75	42.1	36.7	66.7
4DLE25 GR3	RD	56.7	56.7	21	13.3	36.7
4DLE25 GR3	NV	49	58.3	19.3	15	36.7
INVLIJUA26	LX	31	60	31.6	25	40
INVLIJUA26	FL	46	55	21	25	40
INVLIJUA26	RD	35	60	10.5	30	55
INVLIJUA26	NV	21	80	52.6	25	50

presents the sequencing accuracy indicators disaggregated by class group and participant profile.

The results indicate that the distribution across participant profiles was not consistent across the different class groups. In 1DLCLEI26_GR3, the LX and FL profiles achieved the highest values for Exact Positions, Adjacent Pairs, LIS, and ±1, whereas the RD and NV profiles obtained substantially lower scores. By contrast, the distribution was more balanced in 1DLCLEI26_GR4, where the NV and RD profiles outperformed the other profiles on some indicators. Among the two 4DLE25 class groups, the FL profile consistently achieved the strongest performance, particularly in 4DLE25_GR3, where it recorded the highest values across the principal positive indicators of sequencing accuracy. The results for INVLIJUA26 should be interpreted with caution, as only one working group represented each participant profile. Within this class group, the NV profile obtained the highest values for Adjacent Pairs and LIS.

Figure 3 and Figure 4 provide complementary visualizations of the sequencing results. Figure 3 presents the distribution of two main sequencing indicators, LIS and Exact Positions, across working groups and participant profiles. Figure 4 displays the mean positional error for each visual scene, making it possible to identify the points in the sequence where deviations from the original ordering of Lavater’s Cendrillon were greatest.

The patterns observed in Figure 3 and Figure 4 are consistent with the trends reported in the preceding tables, while also showing considerable variability within participant profiles. In Figure 3, the FL profile achieved higher descriptive values in both LIS and Exact Positions, whereas the RD and NV profiles obtained lower percentages of Exact Positions. These differences, however, should be interpreted descriptively, in line with the exploratory inferential results.

Figure 4 provides a more detailed view of the distribution of positional errors, showing that deviations were not evenly distributed across the sequence. The final scenes, particularly Scenes 19 and 20, tended to be placed earlier than their original positions across all four participant profiles, whereas the initial and middle scenes exhibited greater variability in their positional deviations.

4.2. Distribution of Narrative Phases

The second analytical dimension focused on the distribution of the narrative phases assigned to the twenty visual scenes of Lavater’s Cendrillon. For each working group, the number of scenes classified into each of the five narrative phases was calculated: initial situation, conflict, development, resolution, and final situation.

Table 3. Mean distribution of narrative phases by participant profile.

Participant Profile	Initial Situation	Conflict	Development	Resolution	Final Situation
LX	3	2.3	10.4	2.6	1.7
FL	2.5	1.9	10.4	3	2.1
RD	2.8	2.5	9.6	2.9	2.2
NV	2.5	3.1	9.5	3.1	1.8

presents the mean values for each participant profile.

A clear concentration of scene classifications was observed in the Development phase, which accounted for approximately half of the scenes across all participant profiles. The LX and FL profiles recorded the highest mean values for this phase, with 10.4 scenes each, whereas the RD and NV profiles showed slightly lower averages. Differences across the remaining narrative phases were relatively modest. The NV profile assigned a higher mean number of scenes to the Conflict phase, while the FL, RD, and NV profiles recorded slightly higher values than LX for both Resolution and Final Situation.

The predominance of the Development phase emerged as the most consistent pattern across the four participant profiles, although slight variations were observed in the distribution of the Initial Situation, Conflict, Resolution, and Final Situation phases. Overall, the results indicate a general tendency to concentrate the narrative interpretation on the central portion of the visual sequence.

4.3. Verbal Description of the Visual Scenes

This stage of the analysis examines the verbal descriptions produced by the working groups in response to the visual scenes of Lavater’s Cendrillon. Four indicators were calculated: total word count, number of unique words, the type-token ratio (TTR), which measures relative lexical diversity, and mean sentence length. Together, these indicators provide an approximate account of the length of the descriptions, their lexical diversity, and some basic features of their linguistic formulation.

Table 4. Verbal description indicators by participant profile.

Participant Profile	Word Count	Unique Words	TTR	Mean Sentence Length
LX	364	138.1	0.39	23.6
FL	351.2	136.1	0.39	15.8
RD	424.2	167.8	0.4	21.7
NV	366.3	137.5	0.38	14.6

shows that the RD profile recorded the highest mean values for both word count (424.2) and unique words (167.8), followed by NV and LX in terms of total word count. The FL profile obtained the lowest mean values for both word count (351.2) and unique words (136.1), although differences among the LX, FL, and NV profiles were relatively small for the latter indicator. With regard to the type-token ratio (TTR), values were highly similar across the four participant profiles, with the RD profile showing a slight advantage (0.40). Mean sentence length was greatest in the LX (23.6) and RD (21.7) profiles, whereas the FL and NV profiles produced shorter sentences on average.

Figure 5 presents the distribution of the two most directly comparable indicators, word count and unique words, across working groups and participant profiles. The figure clearly shows the greater verbal production of the RD profile, both in terms of text length and lexical diversity. By contrast, the remaining three participant profiles display more similar values, particularly with regard to the number of unique words. Overall, the results indicate that greater verbal elaboration did not necessarily correspond to the higher levels of sequencing accuracy observed in the first analytical dimension.

4.4. AI-Mediated Narrative Construction

The aim of this analysis was not to evaluate the literary quality of the narratives generated by GenAI, but rather to examine how the tool transformed the participants’ descriptions into coherent narratives. To this end, the complete set of descriptions produced by each working group was compared with the corresponding AI-generated story. This comparison made it possible to examine the extent to which GenAI preserved participants’ vocabulary while integrating the individual scene descriptions into a cohesive narrative.

Table 5. AI-mediated narrative indicators by participant profile.

Participant Profile	Story/Description Ratio	Shared Vocabulary	New Vocabulary	Connectives per 100 Words
LX	0.69	26.2	72.5	2.02
FL	0.65	25.3	71.9	2.39
RD	0.57	28.3	70.2	2.27
NV	0.72	26.5	72.7	2.40

presents the indicators used to evaluate the mediation between the participants’ descriptions and the AI-generated stories. The story-to-description ratio provides a comparative measure of the length of the generated story relative to the combined length of the original descriptions. This is followed by two lexical indicators: shared vocabulary and new vocabulary introduced by the AI-generated story. Finally, the number of discourse connectives per 100 words was calculated to examine the incorporation of temporal and causal markers that contribute to narrative cohesion in the generated stories.

As shown in Table 5, the Story/Description Ratio remained below 1 across all four participant profiles. The RD profile recorded the lowest value despite producing the longest descriptions. These results indicate that the AI-generated stories did not increase in length proportionally to the amount of input provided by the working groups.

Regarding the lexical indicators, shared vocabulary ranged from 25.3 to 28.3 content words. The RD profile achieved the highest value for shared vocabulary, suggesting a greater retention of participants’ lexical choices in the generated stories. Nevertheless, all participant profiles showed substantially higher values for new vocabulary introduced by the AI, with averages close to 70 words. Finally, the number of connectives per 100 words showed only moderate variation across participant profiles, with the FL and NV profiles recording the highest values.

A consistent pattern can be observed across the four participant profiles with regard to the lexical composition of the generated stories. In all cases, the vocabulary shared with the participants’ descriptions accounted for approximately one quarter of the total lexical content, ranging from 26% to 29%. These results are consistent with the findings reported in Table 5, indicating that the generated stories retained part of the original vocabulary while incorporating it into a broader lexical reformulation.

A brief qualitative comparison of selected group productions further illustrates this mediating function. In the case of one group with a highly accurate sequence reconstruction (1DLCLEI26_GR3/LX-3), the AI-generated story mainly condensed and connected the students’ detailed descriptions, preserving the main narrative progression while giving it a more cohesive verbal form. By contrast, in a group whose proposed sequence departed substantially from the original order (4DLE25_GR3/RD-2), the generated story reorganized the fragmentary visual interpretation into a more conventional retelling of Cendrillon, adding temporal transitions and causal links that were not always explicit in the input. These examples suggest that GenAI operated as a mechanism of narrative normalization, capable of producing coherence while potentially smoothing over ambiguities or uncertainties in the students’ visual reading.

4.5. Exploratory Inferential Analyses

In order to complement the descriptive comparison among participant profiles, exploratory inferential analyses were conducted on the main outcomes directly produced by the working groups. The models focused on two indicators of sequential reconstruction, LIS (%) and Exact Positions (%), and on one indicator of verbal elaboration, Unique Words in Descriptions. These outcomes were selected because they represent the two central dimensions of the students’ activity: reconstructing the coded visual sequence and verbally elaborating their interpretation of the scenes. The models included reading habits, familiarity with traditional folktales, and their interaction, with class group entered as a control variable. These analyses are presented in

Table 6. Exploratory inferential analyses of the main group-produced outcomes.

Outcome	Reading Habits F	p	η2p	Folktale Familiarity F	p	η2p	Interaction F	p	η2p
LIS (%)	1.70	0.200	0.037	1.21	0.277	0.027	0.09	0.765	0.002
Exact Positions (%)	1.49	0.229	0.033	0.10	0.756	0.002	0.50	0.482	0.011
Unique Words in Descriptions	1.53	0.222	0.034	1.72	0.196	0.038	1.27	0.266	0.028

All models were fitted at the working-group level and included class group as a control variable. η²p = partial eta squared.

.

The inferential analyses did not identify statistically significant main effects of reading habits or familiarity with traditional folktales, nor significant interaction effects between the two factors, for the selected outcomes. Effect sizes were small across the models. These results indicate that the descriptive differences observed among participant profiles should be interpreted cautiously. In particular, although some sequencing indicators showed higher descriptive values for profiles characterized by stronger reading habits, these differences were not statistically significant when class group was included as a control variable.

5. Discussion

The discussion follows the three research questions formulated at the end of the Introduction. The first subsection considers the relationship between participants’ reading profiles and their reconstruction of the visual sequence. The following two subsections address the organization of narrative meaning, first through the assignment of narrative phases and then through the verbal descriptions produced from the coded images. The final analytical subsection examines the AI-generated stories as a further stage of narrative mediation. This structure makes it possible to consider visual decoding, sequential reconstruction, verbal transduction, and GenAI-mediated reformulation within a unified framework of visual literacy, as successive dimensions of the same interpretive process.

5.1. Visual Literacy as Sequential Reconstruction

The first research question concerned the extent to which students’ reading habits and familiarity with traditional folktales were associated with the reconstruction of Lavater’s visual sequence. With regard to this question, the findings support one of the central ideas presented in the theoretical framework. Visual reading cannot be understood as an immediate or transparent process; rather, it involves an act of reconstruction shaped by the inferences made by the reader. In the present study, this interpretative demand was further intensified by the use of a coded picturebook, in which the narrative is not supported by recognizable figurative representation.

In this respect, the sequencing indicators require a more nuanced interpretation than a simple distinction between correct and incorrect reconstructions of the original sequence. Although the percentages of Exact Positions were relatively low, the LIS values exceeded 60% across all participant profiles. This finding suggests that, even when the working groups did not place the scenes in their exact positions, a substantial portion of the original narrative sequence was preserved. These results indicate that visual literacy developed in a relatively balanced and progressive manner: participants were able to maintain continuity between scenes, group related actions, and partially reconstruct the narrative progression through the activation of the hypotext.

When comparing the participant profiles, it is important to avoid concluding that reading habits and familiarity with traditional folktales exert a direct influence on the accuracy of sequence reconstruction. Although the LX and FL profiles showed higher descriptive values on some sequencing indicators, these differences did not reach statistical significance in the exploratory inferential models. The FL profile, despite being characterized by lower familiarity with traditional folktales, achieved the highest descriptive values on the sequencing indicators, whereas the RD profile did not show a corresponding advantage despite its higher familiarity with the hypotext. These findings suggest that familiarity with the hypotext may support interpretation, but should be understood as descriptive tendencies rather than as evidence of a statistically supported relationship between participant profile and sequence reconstruction. More broadly, the results support previous research highlighting the active reading demanded by the coded picturebook (Colistra, 2023), in which intertextual knowledge contributes to meaning-making. They also underline the pedagogical value of making processes of inference and negotiated meaning explicit in the development of visual literacy in teacher education.

5.2. Narrative Structure and the Dominance of the Central Action

The second research question addressed how participant profiles differed in the assignment of narrative phases and in the verbal elaboration of the visual scenes. The distribution of the narrative phases provides an opportunity to discuss an important aspect of reading Lavater’s Cendrillon: how readers recognize the internal structure of the narrative. Given that Lavater’s imageries not only encode characters, settings, and objects but also offer a reinterpretation of the tale through a spatial sequence, the connection with formalist models of narrative structure requires readers to actively reconstruct that underlying organization.

The results reveal a clear pattern, particularly when the four participant profiles are considered together. Across all profiles, the Development phase encompassed nearly half of the visual scenes. The main difference lies in the two profiles characterized by high reading habits (LX and FL), which assigned a slightly greater number of scenes to this phase. The mean of 10.4 scenes in these profiles not only suggests the recognition of a larger portion of the narrative but also reflects a tendency to assign less importance to the opening and closing phases, consistent with the subordinate role traditionally attributed to these stages in formalist models of narrative structure. It is within the Development phase that the actions and episodes most readily associated with the hypotext are concentrated. The predominance of this phase lends itself to two complementary interpretations. On the one hand, it confirms participants’ identification of the narrative’s principal zone of action; on the other, it points to greater difficulty in delimiting the transitions between narrative phases, particularly the moments at which the conflict emerges and is resolved.

Differences across the remaining narrative phases were more modest and less consistent, although they suggest an interesting pattern. The greater representation of both the Conflict and Resolution phases in the NV profile, characterized by low reading habits and low familiarity with traditional folktales, may reflect a broader interpretation of the scenes associated with the narrative conflict, but should also be interpreted cautiously. This pattern may reflect a broader interpretation of the scenes associated with the narrative conflict or greater difficulty in distinguishing the typically brief Conflict and Resolution phases from the more extended Development phase.

5.3. From Visual Reading to Verbal Transduction

The verbal descriptions produced by the working groups offer a complementary perspective on the second research question. As discussed in the previous sections, wordless picturebooks, and even more so coded picturebooks, require readers to engage in a process of transduction. Given the characteristics of the picturebook used in this instructional activity, this process was particularly demanding, as participants were required to transform a coded visual system into verbal language. Describing the visual scenes therefore involved not simply stating what was depicted, but formulating hypotheses about what was taking place at the narrative level.

According to the results obtained, verbal production did not follow the same patterns as sequential reconstruction. Although the FL profile achieved the highest overall scores on the sequencing indicators, the RD profile recorded the highest values for both word count and unique words. This suggests that greater accuracy in reconstructing the visual sequence was not necessarily associated with longer descriptions or greater lexical diversity.

Considering the verbal description indicators in relation to the participant profiles, one further observation can be made. The descriptive patterns for verbal elaboration did not coincide with those observed in the sequencing task. Although the RD profile recorded higher values for word count and unique words, the inferential analyses did not reveal statistically significant effects for the selected indicator of verbal elaboration. Accordingly, the observed differences should be interpreted as descriptive variation across profiles rather than as evidence that reading habits or familiarity with traditional folktales promoted more extensive verbal expression.

Taken together, these findings indicate that the visual literacy fostered through the instructional activity cannot be reduced to the development of a single competence. Sequential reconstruction, narrative phase classification, and verbal description are closely related dimensions, yet they do not develop in the same way across the different participant profiles.

5.4. Generative AI as Narrative Mediation

The third research question focused on the role of GenAI in transforming students’ visual descriptions into coherent verbal narratives. As discussed in the theoretical framework, the incorporation of GenAI into the instructional activity was not conceived as a tool for the immediate production of text, but rather as part of a mediated process through which prior visual reading was transformed into a verbal narrative. In addition to the participants’ descriptions, the prompt submitted to the AI instructed it to integrate the scenes according to the proposed sequence, construct a coherent narrative, and adopt a style consistent with the oral storytelling tradition. From this perspective, the generated stories did not replace the interpretations developed by the working groups; rather, they built upon them, organizing and giving narrative cohesion to the material already produced.

The analysis of the AI-generated stories illustrates this mediating function. The vocabulary shared with the participants’ descriptions accounted for between 26% and 29% of the lexical content of the generated stories. This indicates that GenAI retained part of the vocabulary contained in the prompt while incorporating it into a broader lexical reformulation. The slightly higher proportion of shared vocabulary observed in the RD profile may be related to the greater length and lexical diversity of its descriptions, although the differences across participant profiles were relatively small. Consequently, the findings do not support a strong relationship between participant profile and the lexical reformulation of the descriptions. Rather, they suggest that the AI consistently generated narratives characterized by a relatively stable proportion of newly introduced vocabulary.

In this context, the educational value of GenAI lies in the opportunity to critically examine this transformation by analyzing which elements of the participants’ descriptions are preserved, which are reformulated, and what the tool adds to produce narrative cohesion and continuity. Within the instructional activity presented here, GenAI can therefore be understood as a second layer of mediation, making visible both the potential and the limitations of entrusting narrative construction to artificial intelligence.

Taken together, these findings suggest that the visual literacy explored in this study should be understood as a multi-layered process. Participants first decoded visual signs and inferred relationships among scenes; they then reconstructed a narrative sequence, assigned narrative functions, and transformed visual interpretations into verbal descriptions. GenAI introduced a further stage in this process by reformulating those descriptions into cohesive stories. From this perspective, visual literacy, narrative reconstruction, verbal transduction, and AI-mediated writing are not separate components of the study, but successive and interrelated dimensions of the same interpretive activity.

5.5. Educational Implications and Limitations

Finally, it is important to highlight several educational implications for teacher education that emerge from this experience. The instructional activity confirms the value of the coded picturebook as an appropriate format for foregrounding visual literacy. Throughout the classroom activity, participants did not merely identify visual signs; they were required to formulate hypotheses, establish relationships between scenes, and collectively justify a narrative progression grounded in the relationship between the visual system and the hypotext.

In this respect, the use of physical cards facilitated a process of collaborative reconstruction. Although separating the visual scenes altered the material experience originally afforded by the leporello format, it enabled the working groups to manipulate the scenes, experiment with alternative sequence arrangements, and discuss the underlying narrative logic. From a pedagogical perspective, this approach shifted the activity from an individual reading experience towards a collaborative process of interpretation (Sidra & Mason, 2026).

Among the limitations of this study, it should be noted that it is an exploratory investigation conducted within a specific university context; consequently, the findings cannot be generalized. Although class group was included as a control variable in the exploratory inferential analyses, other contextual and participant-related factors, such as course type or educational level were not modelled separately and may have influenced the results. Future research should not only replicate the instructional activity with a larger sample drawn from different educational contexts but also extend it to other imageries by Warja Lavater based on folktales that are less familiar to our participants within their cultural context, such as Kaguyahime. Lavater’s reinterpretation of this Japanese folktale would make it possible to examine the extent to which prior knowledge of the hypotext, acquired through reading the verbal version of the tale during the classroom activity, influences the subsequent reconstruction of the visual sequence and the production of verbal descriptions.

A further limitation concerns the treatment of reading habits and familiarity with traditional folktales. Although the original questionnaire produced continuous scores, the classroom activity was organized around profiles derived from median splits. Future studies should retain the direct correspondence between individual questionnaire scores and working-group products in order to model these variables continuously and preserve greater statistical information.