Avant-Texts, Characters and Factoids: Interpreting the Genesis of La luna e i falò Through an Ontology

Abstract

This study introduces the Real-To-Fictional Ontology (RTFO), a structured framework designed to analyze the dynamic relationship between reality and fiction in literary works, with a focus on preparatory materials and their influence on narrative construction. While traditional Italian philology and genetic criticism have distinct theoretical and editorial approaches to avant-text, this ontology addresses their limitations by integrating fine-grained textual analysis with contextual biographical avant-text to enhance character interpretation. Modeled in OWL2, RTFO harmonizes established frameworks such as LRMoo and CIDOC-CRM, enabling systematic representation of narrative elements. The ontology is applied to the case study of Cesare Pavese’s La luna e i falò, with a particular focus on the biographical avant-text of Pinolo Scaglione, the real-life friend who inspired key aspects of the novel. The fragmented and unstable nature of avant-text is addressed through a factoid-based model, which captures character-related traits, states and events as interconnected entities. SWRL rules are employed to infer implicit connections, such as direct influences between real-life contexts and fictional constructs. Application of the ontology to case studies demonstrates its effectiveness in tracing the evolution of characters from preparatory drafts to final texts, revealing how biographical and contextual factors shape narrative choices.

1. Introduction

The intense debate around the term ‘avant-text’ is witnessed by the various definitions within the Lexicon of Scholarly Editing,1 which, in a time span from the 1970s to 2010, highlights the semantic shifts, the different approaches of international philology to different kinds of preparatory materials. Wanting to approximate a general definition, from reading its entries, one could refer to the term ‘avant-text’ as the preliminary stage of a text, the set of notes, outlines, drafts, manuscripts, annotations or fragments that the author produces during the creative process, before arriving at the final version of the work, which usually coincides with its publication. This definition is, however, far from shared by the scholarly community and has found different declinations, particularly in the studies of Italian authorial philology and French critique génétique.2 While a synthesis of theoretical positions is possible (Segre 1995, 1998), these positions struggle to represent the interplay between an author’s lived experiences, preparatory materials and the final published work. The challenge remains editorial—how these materials are published—and methodological—juxtaposing structurally and conceptually distinct compositional phases.

Digital technologies offer a potential solution.3 The online representation of facsimiles with transcriptions, critical editions with digital variant apparatuses, allows for indiscriminate publication of all avant-text materials. As Italia notes, it is likely that, after centuries of spending review in which philologists have been ingenious in finding the most discreet and concise way to present variants and abbreviations in their apparatuses, finding themselves faced with the boundless terrain of an available virtual environment, they have preferred to reproduce all the witnesses, choosing the textus certus over the ideal reconstruction of an uncertain archetype (Italia 2020, p. 56). In this case, however, there is the risk of producing digital editions that are anything but user-friendly. In fact, there is a prevailing trend toward documentary–diplomatic editions focused on diplomatic transcription of individual witnesses, often neglecting the author’s corrective textual evolution (Nava 2023). This trend is attributed to the inadequacy of digital tools for such approaches and the difficulty of representing textual diachrony and its linear and non-linear evolution. To overcome the bi-dimensionality of digital representation—which mimics the page format—we must embrace the open semantic network of the graph.

Literary studies have benefited from semantic technologies4 (Ciotti 2012) and modeling interpretation—particularly provenance metadata to attribute assertions—has become crucial (Tomasi 2022; Daquino et al. 2022; Massari et al. 2025). Other ontologies have focused instead on the internal structuring of manuscripts and its intertextual connections, in a descriptive rather than philological dimension (Jordanous et al. 2012; Bartalesi et al. 2022); others like GENO (Christen and Spadini 2019), on the genetic relationships between the manuscripts, diary and publications of Swiss author Gustave Roud and based on CRMtex and FRBRoo, do not explore in-depth textual development within the materials; still others focus on the narrative and narratological dimensions of the text, with an analysis on the static relationships between characters and their attributes (Ciotti 2016; Bartalesi et al. 2016; Tuominen et al. 2017). These limitations derive from a disciplinary division: philology focuses on textual teleology, genetic criticism on processual non-linearity and biographical studies on external contingencies, with no framework unifying all three.

This paper proposes an ontology, Real-To-Fictional Ontology (RTFO),5 designed to model the complex relationship between biographical sources and textual genesis in literary works, overcoming the limitations of traditional philological approaches, using Cesare Pavese’s La luna e i falò (Pavese 1950) as a case study.6 To track the genesis and development of the novel, it is necessary to consider not only preparatory materials, manuscripts and typescripts, but also letters, pages from Il mestiere di vivere and above all, the materials of that ‘biographical avant-text’7 that represent the written and oral testimony of Pavese’s friend and inspiration of the novel, Pinolo Scaglione. Through these materials, characters, places and events in the novel can be interpreted through the perspective of the author’s real biographical experience.

Integrating LRMoo (Library Reference Model—Object-Oriented),8 a new version of FRBRoo (Functional Requirements for Bibliographic Records—Object-Oriented) developed to describe bibliographical entities and aligning with CIDOC-CRM,9 RTFO uses events and factoids to interpret the novel’s genesis. The ontology is designed for scholars who are experts in the field of philology, but is also conceived as a tool with educational purposes to make clear to students the complexity of the genesis of a literary work, the inspiration that produced it, the author’s corrections to manuscript materials that are not easily accessible or understandable, up to the edition available to the public. The novelty lies in its hybrid methodology: combining factoid models (used in prosopography for biographical, partial and contingent data and not usually used in the genetic field) with philological precision, while offering a replicable framework for every work where biography, fiction and philology intersect.

2. State of the Art

The current landscape of ontologies in digital humanities reveals the difficulty of representing the complex interplay between biographical sources and literary creation, through the study of the preparatory materials of a work. This is also the result of the fact that the theoretical positions on the avant-text diverge.

2.1. The Question of the Avant-Text and Its Theoretical Approaches

In genetic criticism, Bellemin-Noël was the first to argue that the study of avant-text is fundamental because it reveals the artist’s intention—what they managed to express in a work, even unconsciously—which would otherwise remain hidden (Bellemin-Noël 1972). Yet he also highlights its inherently interpretative dimension, as scholars retrospectively classify materials as preparatory or finalized post-publication. Hence, he presents the following definition: “[…] un avant-texte est une certaine reconstruction de ce qui a précédé un texte, établie par un critique à l’aide d’une méthode spécifique, pour faire l’objet d’une lecture en continuité avec le donné définitif”10 (Bellemin-Noël 1977, p. 9). Over time, genetic criticism—whose investigative aim is the writing process and its non-teleological reconstruction, rather than the text itself (Grésillon 1998)—has distanced itself from the term avant-text, preferring dossier génétique: “un ensemble constitué par les documents écrits que l’on peut attribuer dans l’après-coup à un projet d’écriture déterminé dont il importe peu qu’il ait abouti ou non à un texte publié”11 (Grésillon 1994, p. 109). The triangular model of textual genesis—endogenesis, exogenesis, epigenesis—seeks to overcome the idea of linear textual development (pre- and post-publication), presenting genesis as a dynamic, interactive process where external materials (exogenesis) can influence both the internal phase (endogenesis) and the post-publication phase (epigenesis) (De Biasi and Wassenaar 1996; Van Hulle 2019).

In Italian philology, however, the term ‘avant-text’ is more strictly defined12 as “[…] l’insieme dei dati materiali relativi a tutto ciò che ha preceduto il testo”13 (Italia and Raboni 2010, p. 26), emphasizing the direct subordination of pre-textual elements to the compositional end point represented by the text. The text as a goal becomes the criterion for classifying preceding data, divided into materials with no direct relation to the text, such as keywords, character lists, intertextual references and materials with an immediate relation to the text, like drafts, versions, final revisions (Italia and Raboni 2010, p. 26). As Stussi argues, “[…] si deve distinguere, fin dove è possibile, tra materiali preparatori e testi, cioè, per esempio, tra una lista di proverbi siciliani e i manoscritti dove comincia a prendere forma I Malavoglia”14 (Stussi 2011, p. 149).

The temptation to include an author’s prior works and everything they have seen or heard within the avant-text risks overextending the concept, rendering it impractical for editorial purposes (Stussi 2011, p. 170; Segre 1985, p. 85). Despite these legitimate concerns, the importance of materials such as notes, diaries and correspondence—not part of the author’s defined project but showing how biographical elements link to textual references in the work—remains undeniable:

Biographie et correspondance de l’auteur, connaissance de l’œuvre dans son ensemble, témoignages de tiers, événements historiques, tout cela nous renseigne sur les conditions externes dans lesquelles se situe une genèse. Même les plus farouches adeptes du “ tous les manuscrits et rien que les manuscrits “ font, heureusement, cette incursion biographico-historique qui, cependant, ne permet qu’un constat, non une explication. Mais là où ce type d’investigation serait plus intéressant, c’est-à-dire dans l’évaluation de l’exogenèse et son éventuelle interférence avec le processus d’écriture à proprement parler, les outils nous font cruellement défaut.15.

(Grésillon 1994, pp. 25–26)

The study of avant-texte materials—especially in the case of La luna e i falò—requires a delicate balance between acknowledging biographical influence and avoiding deterministic interpretations of the final text. As Grésillon notes, the notion of exogenèse—the influence of external cultural, intertextual or personal contexts—remains underexplored.

The case of Pinolo Scaglione, whose oral testimony offers direct insight into Pavese’s creative process, highlights the need for a methodological synthesis. His account calls for a model that integrates the philological rigor of textual documentation with the openness of genetic inquiry into authorial context. The materials surrounding the published novel—fifteen pages of drafts and outlines, a clean manuscript with minor variants and typescript corrections largely absent from the published edition—underscore this interplay.16 There is a clear need for a broader concept than exogenesis, for everything that precedes and influences the creation of a work, including written and oral evidence, even non-authorial, concerning personal biographical events and external historical events: the ‘biographical avant-text’.

2.2. The Question of Incompleteness and Processuality of Avant-Text and Its Ontological Modeling

Projects like GENO (Genetic Networks Ontology)17 represent a significant step in modeling genetic relationships in an ontological model. Following French genetic criticism, GENO moves beyond the traditional avant-text model, mapping, instead, connections between texts, like diary entries, drafts and publications, to create a dynamic genetic network (Christen and Spadini 2019). The ontology distinguishes avant-texts linked via temporal properties and reuse relationships to track post-publication epigenesis. In addition, GENO attempts to align with FRBRoo’s bibliographic hierarchy (Work/Expression/Manifestation/Item), critiquing FRBRoo’s limitations in capturing provisional draft materials (Christen and Spadini 2019, pp. 90–91). But separating the writing process into multiple genetic phases, while useful as a general guide, could be problematic because genetic stages often overlap (Van Hulle 2004, p. 5). While it classifies avant-textual materials, the ontology is not concerned with going inside the text, linking it to fictional entities like characters, places and events. The explicit writing process deals only with the external materials and not with the internal evolution of the characters in the text with their attributes.

Established frameworks like LRMoo and CIDOC-CRM provide robust structures for modeling bibliographic entities and historical provenance, but they fall short in capturing the fluid, often fragmentary nature of avant-text materials.

CIDOC-CRM (Doerr 2003) is highly effective for documenting events and object provenance. However, it lacks dedicated classes for preparatory materials, forcing scholars to categorize drafts under broad classes such as E33_Linguistic_Object. This class represents linguistic artifacts—such as texts and annotations—that convey meaning but do not hold the formal status of a finalized expression. This approach, however, risks oversimplification. Avant-texts often occupy a hybrid role: they are neither generic linguistic objects nor definitive expressions.

The LRMoo model organizes bibliographic entities into a structured hierarchy, distinguishing between the abstract F1_Work, its intellectual realization as F2_Expression, the physical F3_Manifestation and the individual F5_Item. While this framework effectively captures published and finalized texts, it encounters significant challenges when applied to avant-texts, preparatory materials such as drafts, notes and sketches that document the fluid, often non-linear process of literary creation.

LRMoo defines F2_Expression as “the outcome of the intellectual or creative process of realizing a Work” (Bekiari et al. 2024), implying a stable and complete form. However, avant-text materials are inherently unstable, fragmentary and provisional. For instance, Cesare Pavese’s note—“Odor di Canelli punta di mosto, di sabbia del Belbo […]”—was later reworked into the manuscript of La luna e i falò, but in its initial state, it lacks the structural coherence expected of an Expression. Classifying such fragments as Expressions risks misrepresenting their nature, as they are not definitive outcomes but rather dynamic, evolving stages in the creative process. We need to consider that the very concept of ‘text’ emerges as the product of a complex negotiation between multiple dimensions—none of which can be disregarded without distorting our understanding. On one level, there is the author’s conscious intentionality, which, in the act of writing, delineates certain elements as the ‘true text’, distinguishing them from what remains in the realm of draft, variant or discarded material—the avant-text. Yet this authorial design does not exhaust the nature of writing, for the creative act is always permeated by unconscious forces, by an excess of meaning that escapes rational control. To reduce the text to pure authorial will is an illusion, for even the author does not possess full mastery over what they produce (Bellemin-Noël 1979, 2004).

LRMoo guidelines also state that any change in linguistic signs generates a new Expression, but this principle becomes problematic when applied to avant-text. Drafts often contain countless minor revisions—word substitutions, marginal annotations or micro variants—that may not signify a substantial intellectual shift. If every such modification were treated as a distinct Expression, the model would produce an impractical proliferation of entities, complicating archival and cataloging efforts without meaningful scholarly benefit.

A deeper issue lies in the assumption that every Expression must derive from a pre-existing and unique Work (Sanfilippo 2021). In fact, many avant-text materials—such as preliminary outlines, exploratory sketches, genealogies of characters and described fictional events—do not yet correspond to a fully conceived Work. Forcing these materials into the Work–Expression framework imposes an artificial teleology, suggesting a clear progression from draft to final text when, actually, early notes might explore multiple potential directions, some of which are later abandoned. At the same time, the critic or interpreter, in determining what constitutes the ‘text’ and what is relegated to preparatory material or marginalia, engages in an act of mental construction. Through selection, classification and attribution of meaning, they reconstruct an assumed authorial intent. This process, though inevitable, is far from neutral: it is an interpretive gesture that reveals as much about the critic as it does about the work itself (Cadioli 2017, p. 193). The issue becomes even more pronounced when considering the tension between textual materiality and abstraction. In textual philology, a central challenge is determining whether multiple texts by the same author should be considered variations of a single work or classified as distinct works. If the Work is what functions as such beyond the original intentions, for example, in reception, then its identity is never fixed but instead evolves through successive readings and reinterpretations (Sanfilippo et al. 2024). Similarly, the avant-text is not merely what the author discarded, but what criticism has chosen to exclude from the textual canon—often based on debatable criteria.

A more flexible ontological model might introduce an intermediate category of Work and Expression to account for these embryonic creative stages.

2.3. The Question of Modeling Statements in the Literary Field

The semantic representation of scholarly assertions remains a core challenge in digital humanities, particularly in philology and textual criticism, where interpretative complexity and textual fluidity demand models that are both expressive and rigorously contextualized.

Surveying existing ontologies, like CIDOC-CRM and Wikidata, and computational methods in digital humanities, Sanfilippo’s MITE project report Observation Modeling: State of Art highlights challenges to document scholarly interpretations—particularly of literary characters—focusing on three key dimensions: textual, rhetorical and inter-subjective (Sanfilippo and Ferrario 2024). In this context, nanopublications (Groth et al. 2010) or ontologies like PROV-O18 and HICO19 (Daquino and Tomasi 2015) offer foundational models for handling in a general way provenance, uncertainty and interpretive responsibility. But, especially in the literary field, recent studies emphasize precisely how conceptual modeling of argumentative structures like statements, interpretations and their relationships raises critical questions about granularity, contrasting opinions and formal precision (Sanfilippo et al. 2023). Other studies demonstrate, in a strictly practical way, the importance of scalable techniques, such as Named Entity Recognition, topic modeling and Linked Open Data for linking literary texts, secondary literature and metadata within interoperable structures (Schöch et al. 2022).

Although these lines of research provide valuable insights, neither is sufficient on its own. What is needed is an integrated, philologically grounded environment that combines formal semantic modeling with editorial awareness of sources, textual instability and interpretation. In this context, the factoid model, borrowed from prosopography, emerges as a promising bridge between structured semantic assertions and the contingent nature of genetic materials.

A factoid20 is “an assertion made by historians (in a DH context, usually a project team) that a source ‘S’ at location ‘L’ states something ‘F’ about person ‘P’” (Ciula et al. 2023, p. 56). This modeling strategy is particularly well-suited to the nature of the sources examined in this study, the written and oral testimony of Pinolo Scaglione and the avant-textual materials of La luna e i falò—which share many properties with factoids. Like avant-text materials, factoids emphasize both information extraction processes and their interpretative nature, shaped by editorial choices (Kowaleski 2021, p. 319). They share contingency, fragmentation, non-linearity and relation to the truth—in the case of the avant-text with respect to the final development, in the case of factoids with regard to historical truth.

The model includes three main entities (Pasin and Bradley 2015):

-

An agent, person or collective entity, creating the factoid.

-

A document inspiring the factoid.

-

A state of affairs, what the factoid discusses.

The factoid model,21 exemplified by the STAR Model, has the advantage of being linked to a widely used ontology like CIDOC-CRM. The STAR (Structured Assertion Record) model, proposed by Tara Andrews, represents the state of the art (Baillie et al. 2021; Andrews et al. 2024; Andrews 2024). This ensures semantic interoperability, flexibility and adaptability. It also allows for dynamic representation of information, essential in the fluid context of early creative phases, as well as the representation of incomplete or contradictory information, which is particularly useful for biographical avant-text, given that the most direct testimony close to the novel’s genesis comes from Pinolo himself or those who wrote about Pavese after speaking with him. This structure is especially appropriate for modeling assertions such as Pinolo’s claim that a real person inspired the character of Nuto or that specific individuals correspond to fictional counterparts.

This modeling approach inevitably increases semantic complexity, particularly in representing n-ary relations—statements involving more than two entities. According to W3C recommendations,22 these are best modeled not as reified RDF statements, which rely on blank nodes and metadata annotations on triples (Orlandi et al. 2021), but as instances of relationship classes that explicitly link all involved entities. This strategy is more suitable when the relation itself carries contextual information, as in the case of assigning a biographical state or literary trait to a fictional character. Nevertheless, this design choice implies a proliferation of classes and properties, potential challenges in reasoning with OWL restrictions and the need to manage inverse and auxiliary properties.

3. Real-to-Fictional Ontology

The ontological architecture of RTFO aims to model a set of functional requirements emerging from the study of the complex dynamics of literary genesis, particularly as exemplified by Pavese’s La luna e i falò. These requirements are articulated through informal Competency Questions (CQs).

A first area of focus concerns the evolution of textual fragments—from avant-text to the final published version. The ontology is meant to bridge the gap between the Italian tradition of philology, which tends to emphasize linear reconstruction, and the French critique génétique, which instead highlights the non-linear, process-oriented nature of creative writing. In this context, the ontology should be able to address questions such as the following: What materials can be identified as avant-text? How does a particular draft fragment evolve into the final version of a chapter? How did the outlines or planning notes change over time, and to what extent were they respected in the final structure? The model must also support the identification and classification of annotations found in the manuscript pages and allow researchers to query both the structural organization of the materials and their archival provenance—for instance, determining when a given passage was written or from which collection a certain folio comes.

A second set of functional requirements involves representing the provisionality of the creative process. Draft revisions often reflect shifts in interpretation, intention or biographical context. A clear example of this can be seen in the character Nuto, whose occupation changes between different versions—from waiter in early notes to carpenter in the final novel. The ontology must be able to tell what the process of creating and writing the work is and trace such transformations across different states of the text, representing traits and roles assigned to fictional characters at various stages of composition or how places and participants change in specific fictional events described.

A particularly innovative aspect of the proposed ontology is the integration of biographical material into the interpretative framework. Italian philology has often underplayed the role of personal experience in shaping fictional works, yet this dimension proves essential in Pavese’s case. The ontology must support the representation of interpretative layers produced by Pinolo or Pavese himself, allowing for the explicit modeling of how biographical influences are traced and attributed to textual elements: what are the contextual materials that accompany the genesis of the work, its writing, textual modifications and the creation of fictional entities?

Below we report a brief description of the principal classes and properties of the ontology.

3.1. Semantic Avant-Texts

In the context of a computational ontology designed to model the genetic and contextual genesis of literary works, the class rtfo:GeneticWork—a subclass of F1_Work and further constrained by the properties rtfo:isGeneticPartOf some F1_Work and rtfo:isGeneticWorkOf some GeneticExpression—is introduced to account for the status of fragments and drafts as creative yet incomplete acts. These entities emerge during the creative development of a stable work but have not yet reached its final, consolidated form. They represent potential textual entities whose provisional nature is central to understanding the dynamic formation of a literary work.

The ontological framework distinguishes between expressions that fully realize a work and those that document the writing process (see Figure 1). While F2_Expression corresponds to completed realizations of an F1_Work—including final manuscript versions, typescripts or printed texts—these are directly linked to their physical instantiations via the property lrmoo:R4_embodies and include subclasses such as rtfo:ManuscriptVersion and rtfo:TypescriptVersion. In contrast, the class rtfo:GeneticExpression is introduced to represent the intermediate, processual stages of textual becoming. These expressions are not realizations of an F1_Work, but rather document the intellectual and material progression toward it. For this reason, they are distinct from F2_Expression, which presupposes the existence of a completed textual outcome.

Within GeneticExpression, the subclass rtfo:AvantTextualMaterial captures all written materials that participate in the creative development of a text but are not structurally integrated into it in the same way as a manuscript or typescript. This class is defined as equivalent to GeneticExpression and rtfo:isAnnotatedIn some rtfo:Page, indicating their attachment to a physical support. Avant-textual materials including notes, sketches and drafts are further refined into specialized subclasses.

Among these, the class rtfo:Draft is used for texts that are more developed than annotations and marginalia. It is defined by the axiom AvantTextualMaterial and (R5i_is_component_of some rtfo:DefinitiveExpression) and (evolvesTo some rtfo:TextChunk). It is at this level that micro-variants or non-substantive textual variants—currently not separately modeled in the ontology—can be recorded. The class rtfo:LiteraryProjectMaterial, also a subclass of AvantTextualMaterial, includes authorial materials pertaining not strictly to composition but to the broader conceptual planning of the literary work. It comprises subclasses such as rtfo:ChapterPlan, rtfo:Keyword, rtfo:ListOfCharacters, rtfo:LiteraryReference and rtfo:Sketch. Another relevant subclass is rtfo:ProcessualFragment, including working annotations and reworkable fragments such as rtfo:MarginalNote and rtfo:TextualUnit, which reflect the intellectual content of the preparatory phases.

In addition to the textual materials that participate directly in the formation of the literary text, the ontology defines a further subclass of GeneticExpression named rtfo:ContextualMaterial. This class comprises materials that do not directly contribute to the text’s structure but are essential to interpreting its genesis, inspiration and underlying creative logic. The formal definition of this class includes the following constraints:GeneticExpression and (isStatedBy some rtfo:RealPerson) and (isStatedIn some rtfo:Edition). Subclasses include rtfo:BiographicalAvantText, which encompasses both rtfo:BiographicalSource and rtfo:HistoricalSource. This conceptualization broadens the traditional notion of exogenesis by including not only cultural, historical and intertextual influences, but also personal and emotional elements relevant to the author’s creative process. In the case of La luna e i falò, for example, this includes reflections on memory, identity and autobiographical resonance of Pavese’s life. Also part of the upper class are rtfo:JournalEntry and rtfo:Letter, materials traditionally incorporated into contextual avant-texts due to their interpretive relevance.

To capture these entities’ connection to physical supports, several object properties and data properties have been defined. The property rtfo:isAnnotatedIn links an AvantTextualMaterial to a rtfo:Page, which in turn may be part of a rtfo:Manuscript or rtfo:Typescript through rtfo:isPartOf. Pages are further described using data properties such as rtfo:hasSignature, for identifying archival signature, and rtfo:hasPageSide, to indicate recto or verso. The data property rtfo:hasTextualContent allows for the literal transcription of a resource’s textual content.

The ontology also includes additional object properties to articulate semantic relations: rtfo:evolvesTo, a transitive property that links the fragments together and indicates their evolution according to the text of the final edition (domain: AvantTextualMaterial, range: AvantTextualMaterial or rtfo:TextChunk), and rtfo:refersTo, which links a text, avant-text or chunk to an entity it references.

Regarding the physical realization of this process, in addition to physical media such as manuscripts, pages and typescripts, the class rtfo:Edition was introduced as a subclass of F3_Manifestation, with object properties like rtfo:hasPublicationPlace, rtfo:hasPublisher and rtfo:hasAuthor, to describe the final novel or critical studies on it.

Temporal information, instead, is modeled via rtfo:hasDate (range: E52_Time-Span). With regard to temporal information, following CIDOC-CRM, each act of writing or modification should be considered a distinct event associated with a time span entity (E52_Time-Span). A hybrid approach was chosen, linking dates directly to the fragments themselves via rtfo:hasDate (range: E52_Time-Span). This choice significantly reduces the number of additional entities and facilitates queries.

3.2. Semantic Interpretation

The ontology focuses on representing real and fictional entities, with particular attention to the interpretative relationships that link events, traits and states to both real and fictional entities.

3.2.1. Some Real or Fictional Entity

In relation to CIDOC-CRM, the class rtfo:RealEvent (a subclass of E5_Event) includes real events, divided into rtfo:BiographicalEvent, events linked to the author or to the genesis of the work, and rtfo:HistoricalEvent, historical events, broadly defined. As a subclass of E21_Person in CIDOC-CRM, a class rtfo:RealPerson has been added, representing real people, such as Pinolo Scaglione, who may inspire fictional characters.

For the fictional dimension of the novel, the class rtfo:FictionalEntity has been added as a subclass of E28_Conceptual_Object in CIDOC-CRM, representing imaginary entities. Its subclasses include: rtfo:FictionalCharacter, for fictional characters like the individual rtfo:NutoFictional, inspired by a real person; rtfo:FictionalPlace, for imaginary places such as landscapes or cities in a novel; and rtfo:FictionalEvent, for narrative events often derived from historical or biographical events.

3.2.2. Some Real or Fictional Interpretation

Interpretative relationships lie at the core of RTFO, as they formalize the attributions generated by Pinolo’s life and Pavese’s creative process. These attributions bridge real-world sources and fictional constructs and expose the internal dynamics of character and event development.

The heart of this pattern is the CIDOC-CRM class E13_Attribute_Assignment, which we reify in order to treat each attribution as a first-class object. This factoid model allows us not only to state statements like “Nuto is similar to Pinolo”, but also to annotate who makes that claim and where.

The base class is E13_Attribute_Assignment in CIDOC-CRM, with new subclasses23 (see Figure 2):

-

rtfo:InterpretativeRelation, representing assertions that link real sources to fictional elements. Its subclasses include rtfo:SymbolicRelation, which connects symbols to thematic meanings; rtfo:IntertextualRelation, which maps references to other literary works; rtfo:SimilarityRelation, which links elements based on similarities; rtfo:ThematicRelation, which connects themes to passages or between fictional entities; and rtfo:TextualRelation, which indicates attributes that relate to the text. It has as subclasses rtfo:ChapterRelation, which connects fictional events to the chapter hypothesized in the author’s outlines or later described in a chapter of the final work, and rtfo:EventRelation, to establish which places or participants relate to a specific event.

-

rtfo:BiographicalRelation, which links real and fictional entities to their characteristics. Its subclasses include rtfo:HumanRelation, which assigns a type of social interaction between two entities; rtfo:StateRelation, which assigns a state to an entity; and rtfo:TraitRelation, which assigns a specific trait to an entity. For example, rtfo:StateRelation can assign Pinolo Scaglione’s occupation ‘carpenter’ to the individual rtfo:Nuto_fictional.

We distinguish entities—instances of CIDOC-CRM’s E1_CRM_Entity, like a real person or a fictional character—from attributes, modeled as instances of E55_Type, like a trait or a state. Traits and states define the characteristics of entities, whether real or fictional. These are modeled as subclasses of E55_Type in CIDOC-CRM, such as rtfo:Trait, rtfo:State, rtfo:Symbol, rtfo:Theme. Individuals of these classes are specified through the property P127_has_broader_term. Some classes are also defined to indicate the type of property asserted in an attribution, enabling more precise modeling of information (see Figure 3). For instance, the class rtfo:StateType includes individuals such as affiliation, age, education, occupation, residence and sex, allowing detailed descriptions of a real person’s or fictional character’s life circumstance.24 The class rtfo:TraitType includes both personality and physical traits. Additional classes have been introduced to capture more complex relationships: rtfo:NarrativeModificationType includes individuals such as narrativeOrder and contains the subclass rtfo:EventRelationType, which in turn covers aspects like eventParticipant and eventPlace. Finally, the class rtfo:RelationType is divided into subclasses such as rtfo:HumanRelationType—which covers familyRelation, friendshipRelation and socialRelation—and rtfo:SimilarityRelationType, which includes biographicalSimilarity, allowing comparisons between different lives and narrative trajectories.

An E13_Attribute_Assignment instance, therefore, points to two kinds of resources:

-

Subject: the entity whose attribute is being articulated. It is indicated through the object property rtfo:hasSubject, subproperty of P140_assigned_attribute_to, from CIDOC-CRM.

-

Object: the type or value assigned to that entity, which must itself be an E55_Type (for traits, states, symbols, themes, etc.). It is indicated through the object property rtfo:hasObject, subproperty of P141_assigned, from CIDOC-CRM.

These properties are functional, and only one subject and one object are allowed per assignment instance to preserve semantic clarity; when multiple assignments must be made, multiple E13 instances are created. This avoids ambiguity and ensures each statement remains atomic, adhering to STAR’s factoid logic.

The model is completed by properties linked to individuals of the subclasses of E13_Attribute_Assignment, enabling the modeling of complex relationships:

-

rtfo:hasSource, which indicates the textual or historical source of a relationship, under P17_was_motivated_by, from CIDOC-CRM.

-

rtfo:assignsRelationPropertyOfType, which assigns a type of property to a relation, under P177_assigned_property_of_type, from CIDOC-CRM.

At the same time, it was necessary to overcome the static nature of using E13_Attribute_Assignment and highlight how the same trait linked to a character may evolve across different textual representations. To address this problem, the newly introduced class rtfo:TextModificationEvent, a subclass of F28_Expression_Creation, allows one to model such changes explicitly. This class is formally defined to include exactly one previous attribution it overrides (rtfo:overridesAttribution exactly 1 E13_Attribute_Assignment) and exactly one new attribution it introduces (rtfo:modifiesAttribution exactly 1 E13_Attribute_Assignment), enabling the tracking of diachronic narrative shifts. Within this same ontological framework, the process of creating fictional entities is captured by rtfo:FictionalEntityCreation, a subclass of E65_Creation, which is restricted to the creation of instances of rtfo:FictionalEntity (E65_Creation and (P94_has_created only FictionalEntity)). This class includes the property rtfo:includesTransformation, linking it to a rtfo:RealToFictionalTransformation, a subclass of E12_Production, which formalizes the shift from biographical to fictional through the modification of a rtfo:BiographicalRelation. This transformation is constrained by a set of axioms that ensure ontological precision: it modifies exactly one attribution whose subject is a fictional entity (modifiesAttribution only (hasSubject only FictionalEntity)) and overrides exactly one attribution concerning a real person, real event or real place. These formal constraints ensure that the ontology can represent the transformation of real-world biographical references into fictional constructs with logical consistency.

The STAR model treats each relation as an entity, not a mere statement, which better fits the functional requests of this study. As anticipated, this approach inevitably increases ontological complexity: more classes, more properties and more instances are needed to encode what would otherwise be a single triple. Nevertheless, these trade-offs are justified by the benefits: the ability to model contradictory and provisional assertions and the complexity of the creative process.

3.2.3. SWRL Rules

Using SWRL (Semantic Web Rule Language),25 which combines OWL with RuleML (Rule Markup Language) to express logical rules over OWL ontologies, it is possible to infer direct triples about the nature of relationships between people and characters based on already asserted triples.

The following rule infers that, if an author creates a fictional character based on a biographical source about a real person, the fictional character has that real person as its analog:

rtfo:FictionalEntityCreation(?fec) ^ rtfo:carriedOutBy(?fec, ?author) ^ rtfo:hasSource(?fec, ?source) ^ rtfo:isStatedBy(?source, ?realPerson) ^ cidoc-crm:P94_has_created(?fec, ?fictionalChar) -> rtfo:isFictionalAnalogOf(?fictionalChar, ?realPerson)

(Rule.1)

When a transformation modifies a real relation into a fictional one, this rule establishes the analog relationship between the entities involved.

rtfo:RealToFictionalTransformation(?t) ^ rtfo:modifiesAttribution(?t, ?fictionalRel) ^ rtfo:overridesAttribution(?t, ?realRel) ^ rtfo:hasSubject(?realRel, ?realEntity) ^ rtfo:hasSubject(?fictionalRel, ?fictionalEntity) -> rtfo:isFictionalAnalogOf(?fictionalEntity, ?realEntity)

(Rule.2)

When two entities are involved in an authoritative similarity relationship, then the fictional entity is inspired by the real one.

rtfo:SimilarityRelation(?s) ^ rtfo:hasSubject(?s, ?fictionalChar) ^ rtfo:hasObject(?s, ?realPerson) ^ rtfo:SimilarityRelationType(?type) ^ rtfo:assignsRelationPropertyOfType(?s, ?type) -> rtfo:isFictionalAnalogOf(?fictionalChar, ?realPerson)

(Rule.3)

Since the object property rtfo:isFictionalAnalogOf is subproperty of rtfo:isInspiredBy, these rules allow us to directly express the relationship of inspiration between fictional and real entities, in a context where these triples are not asserted uncritically, but are enriched with provenance and interpretative information.

4. Methodology

The ontological modeling process was carried out through a four-stage methodology (Poveda-Villalón et al. 2022). First, the preparatory materials—including manuscripts, annotations and biographical sources—were collected and classified using Excel spreadsheets. This phase involved the identification of key entities, such as characters, places and events; the mapping of genetic relationships between drafts and final versions; and the annotation of creative transformations from real-world references to fictional constructs. In the second phase, the RTFO ontology was developed in OWL2 via Protégé v.5.6.5, adopting a modular architecture to ensure interoperability with CIDOC-CRM and LRMoo. Specific properties and classes were introduced to guarantee traceability and accuracy in the representation of the domain. Formal validation was performed using the HermiT 1.4.3.456 reasoner to ensure logical consistency. The third phase focused on experimental querying, where Competency Questions were expressed as SPARQL queries structured across three levels of analysis. Below are three examples of some possible queries:

−

CQ1. What preparatory materials are related to the genetic development of the novel La luna e I falò? See Query.A1, in Appendix A.

−

CQ2. What participants and places were assigned to the corpse discovery episode during the creative process? See Query.A2, in Appendix A.

−

CQ3. What materials from the biographical avant-text of Pinolo Scaglione influenced the preparatory materials and the creation and writing of the novel? See Query.A3, in Appendix A.

The phase of valutation was implemented through the construction of two complex case studies to fully verify the potential of the ontology.

4.1. La luna e i falò and Two Case Studies

La luna e i falò was written by Cesare Pavese in just two months,26 between 18 September and 19 November 1949, and published in 1950. In this work, referred to as a novel of the “post-resistenza” era (Pavese 2020, 17 Novembre 1949), Pavese draws both on the dramas of collective history, in particular the persistent ghosts of the Resistance that weigh on the nation’s conscience and on his personal memory: his hometown and its inhabitants, past and present, become the setting and characters of the story.27 The idea for the novel originated, in fact, from Pavese’s frequent visits to his hometown, Santo Stefano Belbo, between 1948 and 1950 (Lajolo 2008, p. 252). There, during long walks in the hills with his childhood friend Giuseppe (Pinolo) Scaglione,28 a carpenter and clarinetist, Pavese began to develop the idea for a novel that would recount the lives of farm laborers and tenant farmers, describing their difficulties and dreams of escape. With the Belbo Valley as the backdrop for the story, Nuto, a character inspired by Scaglione himself, emerges as a sort of Virgil who guides a foundling, Anguilla, who has become a successful man, on a journey back to lost time and forgotten roots (Vaccaneo 1999, p. 15). Hot off the press, Pavese sent his friend a copy of the novel inscribed “A Pinolo questo libro—forse l’ultimo che avrò mai scritto—dove si parla di lui, chiedendo scusa delle invenzioni”.29

4.1.1. Case Study 1: Mentina Roggero or Nuto or Valino?

This first case study shows how a diary note about Mentina Roggero, a real person, mother of Pinolo Scaglione, known by Pavese, is transformed—via a genetic fragment—into two fictional characterizations of the characters of Nuto and Valino, each placed in different chapters of the published novel.

The fragment, influenced by rtfo:JournalEntry_3, is part of the overall genetic work for the novel (rtfo:W_LaLunaEIFalo) and evolves to rtfo:TextualUnit_2, a fragment of the avant-text, which includes a reflection of Pavese about someone who “has never left these places = has never seen history” and refers to Mentina Roggero (See Figure 4).

The journal entry that influenced the beginning of this creative process is from Il mestiere di vivere, and Pavese writes this way about Mentina, on 25 June 1949: “La vecchia Mentina, alla Cabianca, che cosa vede nella vita? Che cosa sa della massa enorme di pensieri, di fatti del mondo? Non ha mai mutato il senso, il ritmo che avevano per te i giorni remoti dell’infanzia. E adesso che la rivedi, 70enne, pronta a morire, e che non si pensa nemmeno che possa mutare questa statica immobile vita, che cosa ha meno di te? Che cos’è tutta la molteplice esperienza, davanti a questo? Per 70 anni ha vissuto come tu nell’infanzia. C’è qualcosa che dà i brividi. Questo vuol dire, ignorare la storia” (Pavese 2020, p. 371).

rtfo:TextualUnit_2 evolves into two successive drafts (See Figure 5):

−

rtfo:Draft_2, annotated on manuscript page AP I 10, 10 and part of chapter 1 of the manuscript, which refers to the character of Nuto. This draft later evolves into rtfo:TextChunk_3 in chapter 1 of the published novel, where Nuto is described as someone who “[…] never left the Salto” (See Figure 6).

−

rtfo:Draft_3, annotated on manuscript page AP I 10, 44 and part of chapter 5 of the manuscript, which refers to the character of Valino. It evolves into rtfo:TextChunk_4 in chapter 5 of the published novel, where it says that Valino “[…] from the Belbo Valley had never left” (See Figure 6).

This case study models the evolution of a genetic fragment related to La luna e i falò through different textual stages, connecting biographical inspiration, preparatory materials and fictional outcomes.

4.1.2. Case Study 2: Pinolo, Waiter or Carpenter?

This case study traces how Cesare Pavese transforms and reinterprets a biographical occupation in the fictional construction of the character Nuto in La luna e i falò.

The real person Pinolo Scaglione is described in a biographical source (rtfo:BiographicalSource_6) as a carpenter, specifically a maker of bigonce, wooden tubs, in his own words. This occupational attribution is represented in the ontology as rtfo:StateRelation_1. The biographical resource comes from a passage from an article written by Pinolo Scaglione in La voce dell’Artigiano in 1970, in which he writes: “Sono un artigiano di S. Stefano Belbo. Ho settanta anni. Mi chiamo Scaglione Giuseppe, ma qui mi chiamano Pinolo. Ho provato la città (Torino), ho suonato il clarinetto e diretto una banda composta di tutti miei allievi, per tanti anni, poi mi sono sposato e con mio fratello (sono figlio di falegname) mi sono messo a costruire bigonce, lavoro che faccio tutt’ora. Ho trascorso una vita piuttosto tormentata; senza stenti, ma anche senza agi” (Scaglione 1970, p. 3).

However, in a preparatory character list (rtfo:ListOfCharacters_1), the fictional character Nuto is described as a waiter, resulting in a new attribution (rtfo:StateRelation_2), which overrides the original carpentry-related state. The avant-textual fragment is particularly interesting and says: “Teresa con Nuto tresca. a Genova ratto. Il padre li va a prendere e li sposa. Vanno in America Sud con soldi di Vincenzo Nuto fa il cameriere sul treno Teresa viveva in capanna Il padre va a prenderla Intanto Nuto fa fortuna e la richiama” (AP I 10, 289). This shift is captured in the ontology by the event rtfo:RealToFictionalTransformation_1, indicating that Pavese actively modified the occupation in the early fictional planning (see Figure 7).

Later in the process, Pavese reverts this change in the actual published novel, where Nuto is again described as someone who provides bigonce and presses to the local area. In fact, in the novel we read: “C’è Nuto, il mio amico del Salto, che provvede di bigonce e di torchi tutta la valle fino a Camo” (Pavese 2021, p. 6). This is encoded as rtfo:StateRelation_3, which overrides the previous attribution (rtfo:StateRelation_2) through a rtfo:TextModificationEvent_1 (see Figure 8).

In summary, this case study documents how Pavese re-aligned the fictional occupation of Nuto with the real-life model of Pinolo after an intermediate fictional deviation.

5. Conclusions

Although RTFO does not address epigenesis (post-publication revisions), it provides a tool to bring Italian and French philology closer, demonstrating how semantic technologies can enhance genetic criticism—both editorially and interpretively. By treating biographical data not as background but as active avant-textual layers, RTFO challenges the false dichotomy between ‘internal’ and ‘external’ genesis. We lack models that embrace creative ambiguity as a core feature, and this ontology addresses these gaps through an integrated event-driven model that reconciles the granularity of genetic criticism, the precision of factoid modeling and a redefined bibliographic classification. Its hybrid approach—where a factoid about Pinolo’s testimony dynamically interacts with Pavese’s manuscript revisions—offers a blueprint for modeling how lived experience crystallizes into literature. While the complexity of n-ary relations poses implementation challenges, the ontology’s alignment with CIDOC-CRM and LRMoo ensures interoperability without sacrificing the interpretative nuance demanded by genetic scholarship. The next steps involve the creation of an XML encoding as a structured database, following relevant examples in the field of prosopography (Schwartz et al. 2022) with the aim of semi-automatic extraction of the remaining data and linking entities to an open repository like Wikidata. Future works will expand case studies and refine temporal modeling of creative processes, with the aim of integrating the ontology into a digital scholarly edition.