Interpreting Literary Characters Through Diagnostic Properties

Abstract

This paper investigates an approach to studying analytic relations (identity, similarity, borrowing, etc.) between literary characters using properties and, in particular, properties that are interpretively considered as diagnostic. In our proposal, properties serve as interpretative tools rather than strict ontological features. Unlike most ontological theories of literary characters developed in analytic philosophy, our study focuses on how real-world interpreters construct textual meaning while remaining agnostic about the ontological status of literary entities (ficta, in a more general sense). By integrating perspectives from literary criticism, philosophy, and formal methods, we explore how scholars infer relations between characters through textual evidence, common knowledge, and interpretive frameworks. This research aims at refining methodological approaches to character analysis and at contributing to broader discussions on literary interpretation and fictionality.

1. Introduction

For literary scholars interested in the interpretation of characters, including the analysis of so-called transfictional characters “migrating” from text to text (Saint-Gelais 2011), it is sometimes common to focus on the study of their properties to “trace” characters across the multiple texts where they can be found (Latham 2015; Parey and Roblin 2015). Although the terminology can change, some refer to diagnostic properties as those that are particularly relevant for identifying characters across texts.

To give an example by Eco (2009): “a fluctuating character exhibits a core of properties that seem to be identified by everybody: for instance Little Red Riding Hood is a girl, she wears a red cap, she met a wolf […], even though different people can have different ideas about the age of the girl, the kind of food she had in her basket, and so on” (p. 87). In the same paper, Eco adds: “we can thus assume that a fictional character remains the same even if it is set in a different context, provided diagnostic properties (to be defined for each case) are preserved” (p. 89). In a similar vein, while discussing the character of Sherlock Holmes, Richardson (2011) asks: “is this the same individual as the sanitized version in expurgated versions for teenagers or the still more simplified movie characterizations? Insofar as an essential aspect of a character is altered or omitted, I believe we are justified in referring to the new figure as a different being” (p. 536). In more general terms, according to Richardson, “most characters have a few essential attributes as well as a cluster of traits that may or may not be included in subsequent presentations. […] In the case of Holmes, […] among other attributes, he must be highly knowledgeable, brilliant in deductive reasoning, he probably has to be English, and I suspect he must be male. Otherwise, he would not be Holmes” (pp. 536–37).

In this paper, we aim to present a theoretical framework—at the intersection of literary studies, philosophy, and formal methods (in logic and linguistics)—that seeks, first, to investigate how a characterization of characters can be constructed from properties and, in particular, from properties that interpreters consider as diagnostic; second, to use this characterization to study the analytic relations between characters, especially between characters of different texts (and authors). Throughout this paper, we will consider identity as a prominent example of a relation that has been particularly debated in philosophy (Kroon and Voltolini 2023), but we will also look at (weaker) relations of similarity. As with other research efforts (Gius and Jacke 2017; Pierazzo and Ferrara 2021), we believe that this conceptual analysis can support literary scholars in their studies, as it can allow them to analyze why certain relations between characters exist through the comparison of the properties ascribed to them. Our framework can also help digital scholars in developing tools grounded on theoretical work reflecting actual debates in literary theory and philosophy. Finally, we hope that our work can contribute to the philosophical debate on literary characters by emphasizing the connection between characters and interpretation.

Prior to delving into the sections below, a few further clarifications are in order to contextualize our work. First, our focus is on how real-world agents, such as literary critics and scholars, interpret literary texts.1 Accordingly, we depart from a purely ontological/metaphysical perspective on characters and focus, instead, on experts’ linguistic interactions, realized through natural language sentences, and on their agreements on such sentences. As we will show, we adopt a stipulative stance that avoids ontological commitments on characters, thereby remaining agnostic regarding their existence. It is crucial to stress that this interpretational approach marks an important difference and departure point with respect to the current philosophical debate, especially in the analytic tradition, considering that mainstream theories on fictional entities (ficta) do not seem to ascribe any specific role to interpreters (Kroon and Voltolini 2023).2 Nonetheless, we believe that a philosophically grounded method, which aims at being meaningful for literary scholars, is necessary to systematically ascertain how two or more characters, possibly occurring in multiple texts, are related (according to interpreters).

Second, concerning interpretation, we assume that a literary text does not have a single prescribed meaning (content), but that it can be interpreted in various (possibly incompatible) manners. Following Hempfer (2024):

if explicit meanings are assigned to a literary text on the basis of textual evidence and a process of inference based on criteria of rationality, then they are not in any way ‘given’, but created in the interplay between ‘textual data’ and inferential relations, and that means constructed. As a result, there is no ultimately valid, ‘true’ interpretation, because both the data and the inferential processes can be challenged.

(pp. 22–23; emphasis is ours)

Importantly—and this is our third point—we also follow Hempfer, among others (Gius and Jacke 2017; Jacke 2014; Margolin 1990), in understanding interpretation as a form of (not necessarily deductive) inference. In particular, as we will see, one can infer different information on literary texts on the basis of lexical or common sense knowledge that may be shared among interpreters, individual knowledge on literary theories or other texts, historical knowledge, etc.

The paper is organized as follows. In Section 2, we sketch the debate on the conceptualization of diagnostic properties, drawing on various philosophical theories. Our approach is outlined in Section 3 and Section 4, where we discuss our perspective on interpretation and the attribution of diagnostic properties to literary characters, respectively. Additionally, in Section 4, we explore how to characterize relations between characters based on the properties ascribed to them by interpreters. Section 5 concludes this paper.

2. Diagnostic Properties: An Overview of the Debate

In the following sections, we present some state-of-the-art considerations of diagnostic properties. In particular, Section 2.1 presents the distinction between internal and external properties, whereas Section 2.2 discusses properties attribution. It is important to stress that we talk of properties only in a general manner as “features”, “traits”, or “attributes”; we will provide a more precise account throughout the paper.

2.1. Internal and External Properties

A first question when examining diagnostic properties is what sort of attributes they cover. For instance, considering the examples by Eco (2009) or Richardson (2011), diagnostic properties seem to be, primarily, features that are attributed from the internal perspective of the narrative where characters appear. Examples could be being a detective or living in Baker Street as properties for Sherlock Holmes in the context of The Adventures of Sherlock Holmes. One may wonder whether diagnostic properties also cover features from a perspective which is external to the narrative. Examples are those relative to the creative history of characters (being created, etc.), as well as properties reflecting scholars’ interpretations (being an engaging character, being an influential character, etc.).

If we include external properties in diagnostic characterizations, we might face some problems. Consider, for instance, being created by Doyle as an example of external property applying to Holmes. Should it be part of its diagnostic characterization, it could prevent the identification of Holmes across texts by different authors, based on the assumption that a single character cannot have multiple (unrelated) creators. This consequence is unfortunate, as it runs against the idea of introducing diagnostic properties to identify characters across texts hypothetically produced by different authors at different times. In the case of external properties relative to scholarly interpretation, the overall matter is even more complex, considering that scholars may disagree up to the point where their diagnostic characterizations are not compatible. This is one of the reasons why external properties are commonly excluded from characters’ diagnostic characterizations, which are therefore limited to internal properties—a position sometimes called internalism (Fine 1982).

Besides Eco (2009) and Richardson (2011), philosophers in the Meinongian tradition have extensively debated on the attribution of properties to fictional entities (see (Berto 2013, pp. 116–51); (Voltolini 2006, pp. 3–33); Kroon and Voltolini (2023); Reicher (2022) for an overview). The terminology varies, with nuclear and extra-nuclear properties likely being the most common notions. The core idea is that nuclear properties define an object’s nature, while extra-nuclear properties further qualify it. However, even in its most established formulations, the nuclear/extranuclear distinction is given only in primitive terms. As Parsons (1980) puts it,

if everyone agrees that the predicate stands for an ordinary property of individuals, then it’s a nuclear predicate, and it stands for a nuclear property. On the other hand, if everyone agrees that it doesn’t stand for an ordinary property of individuals (for whatever reason), or if there’s a history of controversy about whether it stands for a property of individuals, then it’s an extranuclear predicate, and it does not stand for a nuclear property. Of course, this ‘decision procedure’ is a very imperfect one.

(p. 24)

Also, within nuclear properties, some distinguish between constitutive and consecutive properties, with the latter entailed by the former on the basis of a certain form of entailment (Reicher 2022). The distinction between nuclear and extra-nuclear plays a fundamental role for identity; in this view, indeed, entities are the same if and only if they have the same nuclear properties (Parsons 1980, p. 19).

For philosophers supporting the nuclear/extranuclear distinction, a position called nuclear Meinongianism by Berto (2013), it is important to emphasize that this framework is not limited to fictional entities, but must be understood within a broader metaphysical context. In the specific case of literary characters, nuclear properties are those ascribed to them within their narratives (Parsons 1980, chap. 3), whereas extra-nuclear properties pertain to them from an external perspective, echoing the internal vs. external distinction discussed above. For example, according to Parson, Holmes is the “the object which has exactly those nuclear properties which he is understood to have in the Conan Doyle novels” (Parsons 1980, p. 54). However, this parallelism should be taken with a grain of salt. While the distinction between nuclear and extra-nuclear applies to physically existing entities, one could question whether the internal/external distinction applies to them, considering the lack of a narrative from which such properties could be ascribed.

Nuclear Meinongianism has to face a serious criticism on how exactly a systematic distinction between the two property kinds can be drawn. A notorious counterexample comes from narratives such as Pirandello’s Six Characters in Search of an Author, where even being a literary character, considered as a paradigmatic case of extra-nuclear property, qualifies as both nuclear and extra-nuclear (Voltolini 2006, p. 26). In the Meinongian debate, this difficulty, among others, pushed some philosophers to adopt a strategy based on different modes of predicating properties rather than on different property kinds, leading to the so-called dual copula Meinongianism—the terminology is always by Berto (2013). Zalta (1983), for instance, distinguishes between two senses of predication, i.e., encoding and exemplification. Accordingly, real-world entities exemplify properties; literary characters, which Zalta considers as abstracta, encode the properties attributed to them in their stories, but they can also exemplify other properties. For example, Holmes encodes the property of being a detective and exemplifies that of being a literary character or being created by Doyle. In this view, two characters are the same if and only if they encode the same properties according to their stories.3

Despite the theoretical distinctions between nuclear and dual-copula Meinongianism, some have noted analogies between the two views and have consequently raised similar criticisms.4 For instance, since it is difficult to draw a clear-cut distinction between nuclear and extra-nuclear properties, it remains equally challenging to determine when entities like characters either exemplify or encode properties. Zalta himself considers this distinction as primitive and appeals to intuition when distinguishing between the two (Zalta 1983, p. 38). Furthermore, strategies have been developed to “translate” one approach into the other. In the case of characters, nuclear properties correspond to the properties that they encode, whereas extra-nuclear properties are those that they can exemplify (Berto 2013, pp. 132–33).

An approach that at first glance combines the two presented above is proposed by Reicher (2010), who argues that her position can be formulated in terms of either predicates or properties providing a sort of “translation strategy.”5 In her words:

If a character c is determined as F, then c exemplifies (i.e., has) the property of being F as an internal property (i.e., c exemplifies the property of being F according to a story).

(p. 118; emphasis is ours)

If a character c is F, then c exemplifies the property of being F as an external property.

(ibid.; emphasis is ours)

For characters’ identity, Reicher plainly relies on (what she calls) internal properties: “a character $c_1$ of a work $w_1$ (a story $s_1$) is identical with a character $c_2$ of a work $w_2$ (a story $s_2$) if, and only if: For every [internal property] F: If $c_1$ is determined as being F in $w_1$ ($s_1$), then $c_2$ is determined as being F in $w_2$ ($s_2$), and vice versa” (p. 126). Reicher herself recognizes that this condition is strict, since even marginal changes in a given work (or story) compromise the character’s identity.

To deal with this issue, she distinguishes between maximal and submaximal characters. A maximal character is “a character that has all those properties as internal properties that it exemplifies according to a given story” (p. 129), whereas a “sub-maximal character is a character that has a subset of these properties as internal properties” (ibid.). In this perspective, only sub-maximal characters can be identical. For instance, consider the character of Faust that is maximal with respect to Marlowe’s play. If the character of Faust in another play shares some similarity with Marlowe’s character, in Reicher’s analysis this means that a part of Marlowe’s maximal character is similar (up to identity) to a part of the maximal character in the other play. In the case of serial products, the single books or episodes are to be considered as parts of one large story. For instance, the maximal character of Holmes is the character with respect to the whole series of Doyle, whereas the character of Holmes featuring in A Study in Scarlet is a submaximal part of it.

Besides the distinction between internal and external properties/predication, Reicher’s proposal strongly relies on the notions of story and work to determine what properties a character satisfies. In her view, a story is a type of (a series of) events or states of affairs that can be instantiated in concrete events/state of affairs, where—in Reicher’s words—“fictional stories usually have never been instantiated” (p. 115). The concept of work is more complex; each work comprises a story, a mode of presenting it (e.g., through words, sounds, pictures, gestures), and elements of the intended experiences, namely, “experiences […] which the author [of the work] intends his audience […] to have as a result of the reception of the mode of presentation element” (ibid.). Works are distinguished from stories, and the same story can be included in different works.

Let us focus on the concept of story, which Reicher considers more clear-cut than that of work.6 Consider a fan-fiction about Holmes: should its story be regarded as part of Doyle’s Holmes series? More generally, what criteria determine when two stories are the same or when one is included in a work? Reicher does not delve into these aspects of her theory. Based on (Reicher 2010), one could speculate that her response would involve considering the underlying entities, so that two stories are identical if (and only if) they can be instantiated by the same events or states of affairs. If this were Reicher’s stance, two further observations would arise: First, the stories told in literary works (however one defines the latter) are typically incomplete. This raises the risk of collapsing multiple stories into one, especially when they are highly underspecified. Second, one might question whether authorship plays a role in distinguishing stories and the characters that feature in them. More broadly, if the identity of maximal characters depends on the story in which they are characterized, a criterion for story identity is required. A final remark concerns the maximality of characters in relation to stories and, in particular, to series, which Reicher treats as forming a single, overarching story. It remains unclear how Reicher accounts for cases in which different episodes within a series ascribe contradictory properties to a single character, e.g., due to a possible oversight by the author.

2.2. Properties Attribution

We now turn to the question of how characters come to possess properties. In this regard, at least two perspectives can be identified. The first holds that a character’s properties—including diagnostic ones—are determined by the text in which the character appears, particularly with reference to traits internal to the narrative. This view can be extended by appealing to the intentions of the author, a position endorsed by Reicher (2010). In her words: “what determines a character’s internal properties? […] I strongly tend to an author-centred view on this matter, because, first of all, I’d like to take serious[ly] the familiar idea that authors are the creators of their characters (as well as of stories and works), and, second, I’d like to take serious[ly] the view (which is at the heart of what has been said so far) that characters are objective entities (just like stories and works), i.e., something that can be grasped, in principle, by distinct subjects at distinct occasions” (pp. 121–22).

There are at least three possible lines of reaction to this view. First and foremost, literary texts can be ambiguous. For example, it may not be clear whether a text attributes a property to a character. For example, in Dante’s Commedia, Ugolino is imprisoned in a tower with some of his younger relatives and starved to death (Inferno, Canto XXXIII). Due to the text’s ambiguity, readers across the centuries have debated whether Ugolino resorted to eating his sons and nephews to satisfy his hunger.

Second, some properties may not be explicit in a text, although they could be inferred from it. To give a simple example, from textual evidence in Doyle’s texts about Holmes living in Baker Street and from knowing this street is located in London, which is well-known to be in England, one ultimately derives that Holmes lives in England. Assuming that the latter information is not mentioned in the texts, for it to be inferred one must rely on commonsense and lexical knowledge, as well as on basic inference capabilities.

Third, that a property is diagnostic for a character is not something “written in bold” in a novel, but rather something that requires a selection by interpreters. To quote Reicher (2010) again: “the tricky question is, of course, which of a character’s internal properties are supposed to be essential and which are not. […] I am afraid that it is impossible to draw a sharp line between a character’s essential properties on the one hand and its inessential properties on the other without arbitrariness” (p. 127; emphasis is ours). It is precisely because of such arbitrariness in the selection of essential (diagnostic) properties that Reicher rejects characters’ characterizations in these terms, calling for an “objective” way of describing them with reference to authors. Richardson (2011), on the other hand, is much more cautious when saying that Holmes “probably has to be English” (emphasis is ours). To quote another example, in discussing Flaubert’s character of Emma Bovary, Eco (2009) claims that the fact that she commits suicide is not relevant for her identification across texts, so that Emma remains the same character in Woody Allen’s short story The Kugelmass Episode, where suicide is not mentioned at all. One could entirely disagree with Eco, ascribing relevance to Emma’s suicide, as this latter may suggest fundamental traits of her psychology—if characters are the sort of things allowed to be ascribed with psychological traits (see Phelan (1989) on this latter point).

These three considerations suggest that relying only on texts to characterize characters is not sufficient, inasmuch as referring to interpreters and the way in which they select diagnostic properties is needed. In our view, appealing to the author’s intention à la Reicher (2010) does not solve the matter. Without entering into the debate on authors’ intentions (Krausz 2010), it suffices to notice that, from the pragmatic perspective of interpreting characters, it is hardly possible to trace the will of their authors, especially if they did not leave any trace of what they meant with their texts. In addition, authors themselves often leave to readers the interpretation of their creative production. It is worth quoting Tucker (2015) on an interesting episode relative to Doyle and his conception of Holmes: “When William Gillette, the American actor, once asked Doyle if he could introduce a love interest in his Holmes drama, the author briskly cabled back You may marry him, murder him or do what you like with him” (p. 4; emphasis is ours).

Such claims suggest that, as interpreters, we can ascribe diagnostic properties to characters to engage in debates about them, including a debate on how characters in multiple texts relate. Appealing to authors’ intentions does not seem to be a promising strategy, as authors themselves may defer the question of what counts as diagnostic to interpreters.

The second view to ground the selection of (diagnostic) properties relies, therefore, on interpreters. The three objections mentioned for the first view do not stand in this case. For instance, if a text is ambiguous, interpreters must decide what to do, e.g., by selecting a specific property. In principle, being a matter of selection by the interpreters, characterizing characters through diagnostic properties may not always be possible, e.g., because the text at hand does not provide enough elements for a selection that is meaningful from a literary perspective. Also, this interpretive stance makes sense of the fact that some of the properties ascribed to characters are inferred, with such inferences being subjective to interpreters, so that different interpreters may provide different inferences on the same texts. Notice that it remains often unclear the sort of entailment that some philosophers have in mind when discussing characters’ properties, as there is no reference to real-world interpreters, their knowledge, capabilities, etc. We will further dig into this aspect in the sections below.

Finally, recalling the previous section, it is important to emphasize that the interpretive nature of characters’ characterization applies not only to the external discourse, but also to the internal perspective. It is therefore possible that, for a single text, interpreters provide alternative and even incompatible internal diagnostic characterizations for a character, in which case one may actually wonder whether they refer to the same character—an issue that we shall address in Section 4.

3. The Interpretive Game

So far, we have considered diagnostic properties and their selection in an intuitive manner, whereas we now attempt to provide a more precise approach for describing characters and their interrelations (see Section 4). In particular, since we conceive the selection of properties as interpretive, we need to spend some words beforehand on interpretation so as to further develop what we said in the Introduction. The comprehensive approach we adopt is presented by Sanfilippo et al. (2024) and borrows methods from logic and linguistics;7 here, we only present the main concepts without going into the detailed formalism. Specifically, we discuss in Section 3.1 a stipulative approach to predicate over literary characters (and ficta in a more general sense); the notion of commitment, which is central to the presented research, is discussed in Section 3.2. Finally, interpretation is discussed in Section 3.3.

3.1. Pragmatic Stipulations

As mentioned in the Introduction, we depart from a purely ontological/metaphysical account on fictional characters. This move seems legitimate with respect to the practice of literary interpretation, where experts rarely engage in debates to establish the ontological nature of Holmes, Emma Bovary, count Ugolino, or any other character. What scholars seem to do, instead, is to collect information about characters as they appear in texts and use this information for their analyses (examples are Barolini (2000); Latham (2015); Saint-Gelais (2011), among others). Hence, one does not need to delve into ontological considerations to talk to others about novels, as it is enough to embrace a stipulative attitude towards fiction. From a semantic perspective, a name has a referent, which is only stipulatively considered and which is characterized by the interpretations of the text in which the name appears. As we will see, this approach can allow for different interpreters to share the same characterization of a character without having neither a common ontological theory, nor common ontological commitments. Following Hirsch (2005), this is because each discussant can make sense of what others say based on their own theories, commitments, and shared principles of conversation. Consequently, names are not considered rigid designators because they can refer to different entities in different contexts.8

In our work, this stipulative attitude has to be understood as a technical move to predicate over characters, rather than as the expression of a philosophical stance dismissing characters’ ontological existence as, e.g., in the case of Sainsbury (2009).9 A similar approach is common in research in the digital humanities (Pierazzo and Ferrara 2021; Schöch et al. 2022) and linguistics (Maier 2017), but it is also found in literary studies and philosophy. For example, according to Genette, quoted by Galván (2017), there is something akin to mathematics in the way in which one posits the existence of fictional entities in literary contexts: “Let there be a triangle ABC; Let there be a little girl living with her mother, and so on” (p. 101). Buzzoni (2025) arrives at similar conclusions when arguing that artistic fictions like literary narratives “transport us into a purely hypothetical-counterfactual dimension”, whose entities can therefore be said to exist only in a hypothetical sense. Thomasson (2003) claims that it is common in contexts like literary interpretation practices to stipulatively introduce terms like ‘literary character’ with “rules of use embodying at least tacit existence, survival, and persistence conditions” (p. 146) for what are meant to be their referents.

Interpretation can be likened to a game, in the sense articulated by Wittgenstein in his Philosophical Investigations (Wittgenstein 2009). In such an interpretive game, interpreters are the players, their interpretive statements are the moves, and the texts they interpret function as the rules—that is, the constraints that must be respected for an interpretation to count as an interpretation of a text. This analogy highlights, as emphasized by Hempfer (2024) quoted in the Introduction, that interpretation must follow criteria of rationality: it cannot be arbitrary, since the text itself provides the constraints and evidentiary basis—what Hempfer calls “evidences”—against which interpretations must be justified. Moreover, interpreters, and especially scholars, are expected to offer reasoned arguments to support their interpretations and to be familiar with prior critical work, particularly when analyzing texts that have been the subject of extensive commentary. Finally, as with any game such as chess, players operate within a shared structure of rules and permissible moves, without necessarily committing to a shared ontology regarding the nature of the pieces. Similarly, in interpretive practices, all we have are the text and its (partial) interpretations. Agreement among interpreters emerges as a linguistic and pragmatic fact, not as a consequence of shared metaphysical assumptions about fictional entities.

As a consequence of this language-oriented approach, we primarily focus on how interpreters characterize characters by exchanging linguistic sentences involving proper names10 occurring in the texts they interpret. This view seems coherent with interpretation practices, where experts primarily confront with texts and their linguistic dimensions.

As a simplifying assumption, we take a proper name to refer to a single character within the context of a given text. For instance, the name ‘Holmes’ in the first part of the The Hound of the Baskervilles is understood to refer to the same character as ‘Holmes’ in the second part of the text. In our experience, proper names are considered by literary experts as “warnings” on the possible relations between characters across texts. For instance, that ‘Holmes’ appears in a fan-fiction may suggest that its author wishes to establish a link with Doyle’s work, even if a weak one. However, we hold that the use of the same name may be a pure contingency and, in particular, that it is possible that different characters share the very same name in different texts, even when these texts are written by the same author. For instance, we leave open whether what is called ‘Holmes’ in The Hound of the Baskervilles is different or identical to what is called ‘Holmes’ in The Sign of the Four. To avoid confusion, we contextualize proper names to texts and talk, e.g., of ‘Holmes_HB’ for the Holmes in The Hound of the Baskervilles, and ‘Holmes_4S’ for the Holmes in The Sign of the Four. Furthermore, we assume that interpreters first select (diagnostic) traits in the context of a specific text, e.g., they separately characterize ‘Holmes_HB’ and ‘Holmes_4S’, and only then they establish the relations between the underlying characters on the ground of their characterizations. Hence, according to this approach, interpreters first distinguish between different characters and their respective properties, and then establish how they relate.

We can now elaborate further on interpretation.

3.2. Public Commitments

An important notion that we need to introduce is that of commitment: we say that an interpreter a is committed towards sentence s with respect to text T, when according to the reading of T by a and according to a’s (common sense, lexical, etc.) knowledge, interpretive skills, and reference to other texts, among other aspects, the information provided by s follows from the one provided by T, where s and T are in the same natural language and a is a competent speaker of this language. For example, using a simple formal notation, one could have that $\mathtt{COMMIT}(a,$ “Holmes lives in England”, The Sign of the Four), i.e., the interpreter a commits that the (natural language) sentence “Holmes lives in England” follows from Doyle’s text The Sign of the Four. This may be possible because a has knowledge about Baker Street being in London (information from the text) and about London being in England (information from a’s knowledge). A child may not be able to draw this inference because of a lack of relevant knowledge or a poor understanding of the language.11 Commitments are therefore grounded on an inference process that, starting from the syntactic disambiguation of a text T performed by the interpreter a12 and considering the knowledge of a (revised to take into account the information in T), allows us to infer s.

There are analogies with respect to the work of Orilia (2025) that are worth considering. In Orilia’s view, “[a] story is a proposition, typically a complex conjunctive proposition made up of many conjuncts. At least in the case of literary texts, these conjuncts correspond to the sentences that sequentially compose the text” (p. 3). Also, “[t]here is a paratextual relation that links a proposition to a story and that we express with locutions such as “according to the story” […]. A proposition P is linked by this relation to a story S, when either (i) S explicitly asserts P, […] or (ii) P can be inferred from S plus all the background assumptions […]” (ibid.). Orilia argues that this “paratextual relation is an entailment relation: it is the case in story S that P […] if, and only if, $(B\&S)$ entails P, where B is the conjunction of all the appropriate background assumptions” (ibid.).

With respect to our approach, note that, first, B, S and P are propositions intended as “possible sentence meanings” (Orilia 2025, p. 3), whereas we consider inferences between sentences s and text T once “translated” into formulas in a logical language—following Discourse Representation Theory (Kamp et al. 2010). Second, while B, S, and P capture an objective and unique view on texts, our commitments are subjective because different interpreters can disambiguate T (and s) in different ways—i.e., different interpreters can translate the same text into different formulas, can have different knowledge, or can revise it differently to take into account T (see below for more details).13 Third, commitments are partial and public because, even though they are grounded on inference, interpreters may be unaware of some implicit knowledge or they may simply choose not to make (part of) this knowledge public, i.e., accessible to all the other interpreters. Four, commitments concern T, i.e., T plays a fundamental role in inferring s. In particular, either (the translation of) s can only be inferred by supplementing a’s knowledge with the information in T, or (the translation of) s follows directly from a’s knowledge, i.e., a already knows s, but this only occurs when (the translation of) s also follows directly from (the translation of) T, i.e., the text must also state s. In the example, (the translation of) “Holmes lives in Baker Street” in the The Sign of the Four (let us suppose) must be used to infer (the translation of) “Holmes lives in England.”

To clarify how the knowledge of interpreter a and the content of the interpreted text can be integrated without generating inconsistencies, a few further details are necessary.

First, following Berto (2013), we assume that existence and non-existence are properties like any other property. In particular, they are considered contingent properties, which can therefore change across different worlds. Second, interpretations are made by agents who live in a world and have knowledge of that world, i.e., interpretations are given from the perspective of a given world, called the actual world. Third, interpreters can have knowledge about entities that exist in the actual world, also called real entities, but also about entities that non-exist in the actual world, called fictional entities (and about relations that hold between real and fictional entities).

Assume, e.g, that the interpreter a knows that in the actual world people have lungs but do not fly, while the text T tells the adventures of Superman, who is a flying man. How can the information in T be integrated with the knowledge of a maintaining consistency?14 Interpreter a could accept what is stated in T, i.e., that there are people who fly, while assuming that flying people are fictional entities. Superman is then classified as a fictional person, and the knowledge base of a must be revised to claim that only real people do not fly. Superman is still a person and, since T provides no information against the fact that people (real and fictional) have lungs, the knowledge of a does not require any additional revision. Hence, one can infer that Superman has lungs even though he is fictional. Accordingly, this means that, unless explicitly contradicted by the information in T, the rules governing the actual world apply to all its inhabitants, being these real or fictional.15 Conversely, if a assumes that T is talking about a real person, e.g., T is a scientific essay or a newspaper claiming that Superman exists in the actual world, then a must drop the axiom assuring that people do not fly or refine it to take into account different kinds of real people. Suppose now that a only knows that people have lungs. In this case, that Superman is a flying man is consistent with a’s knowledge about the actual world, which requires no revision. In this case, a does not have to decide whether Superman is real or fictional, because both choices are consistent with their knowledge. In more general terms, the interpretation of T may be influenced by a’s attitude towards T, their knowledge of the author of T, some stylistic similarities of T with other texts, etc.

An intriguing case of reasoning between real and fictional entities involves the “importing” of knowledge across texts. Suppose that after reading the adventures of Superman, a reads the adventures of Holmes. Also, suppose that a classifies Holmes as a fictional person. If we consider a’s knowledge before reading the adventures of Superman, a can infer that Holmes does not fly. Conversely, if we consider a’s knowledge after reading the adventures of Superman, the inference may not be made, since there are some fictional people that do fly. That is, the revision of a’s knowledge about fictional people due to the reading of the adventures of Superman prevents a from inferring that Holmes does not fly even though the text says nothing about Holmes’ ability with respect to it. A different strategy is to assume that Superman is the only exception to the rule, i.e., that the rule applies by default to all people except Superman. By acquiring information about other entities that do not fly, additional exceptions can be added. An intermediate level where exceptions concern fictional entities of certain kinds can also be considered.

3.3. From Commitments to Interpretations

With these considerations in mind, the (public) interpretation of text T by agent a can be understood as the largest set of sentences S on which a commits with respect to T. For the sake of precision, we can write Formula (1) based on set-theory, where $\mathtt{INT}(a,S,T)$ reads: “sentences S are the interpretation of text T by interpreter a”, and $\left\{s\right|\mathtt{COMMIT}(a,s,T)\}$ is the set collecting all the sentences to which a commits with respect to T.

$$\mathtt{INT}(a,S,T)\mathrm{if}\mathrm{and}\mathrm{only}\mathrm{if}S=\left\{s\right|\mathtt{COMMIT}(a,s,T)\}\mathrm{and}S\mathrm{is}\text{non-empty}$$

(1)

It is important to stress that, as previously said, interpretation is partial in two relevant senses. Given $\mathtt{INT}(a,S,T)$: $\left(i\right)$ S may contain only a subclass of the inferences performed by a, namely those sentences that a decides to publicly commit to; $\left(ii\right)$ even when S contains all the inferences performed by a, S does not necessarily contain all the sentences that, given the knowledge of a, follow from T. The incompleteness of interpretation is an inevitable feature of critical discourse, where interpreters focus only on specific aspects of texts.

Interpretations of texts can be generalized to groups of interpreters who share their individual interpretations, either completely or partially.

4. Characterization of Literary Characters and Their Relations

We will focus on the attribution of properties (or relations) to literary characters, since these are at the core of our analysis. However, as we will see, our approach is more general in that it can be applied to real people as well.

4.1. Diagnostic Properties

We understand the attribution of properties to characters as a form of commitment to sentences involving characters’ names. That is, an interpreter commits to some of the properties attributed to the character to which a name stipulatively refers. Among these properties, interpreters can select properties that they consider diagnostic, given their peculiar role in identifying characters, and therefore in studying how multiple characters relate. Consequently, by making this selection, interpreters also isolate what might be called marginal properties, which, as we will see, can also be useful for analyzing characters.

We write $\mathtt{dTRAIT}(a,R,n,T)$ for the set of sentences R relative to name n in text T selected by interpreter a ($\mathtt{dTRAIT}$ reads as “diagnostic traits”). Sentences R therefore represent the diagnostic properties that a attributes to the character named n.16 Since diagnostic traits are selected among interpretations (and, as commitments, they have a public dimension), they are always included in them, i.e., if we have $\mathtt{dTRAIT}(a,R,n,T)$ and $\mathtt{INT}(a,S,T)$, then R is always a subset of S. Because literary texts do not make explicit that certain properties are diagnostic for their characters, the selection of diagnostic properties likely involves cognitive processes beyond the ones considered for interpretation. We leave this topic open for further work, including research on literature and cognitive sciences (Caracciolo 2016), as well as on the role of authority, social groups, and institutions in interpretation practices (Olsen and Pettersson 2005).

Given the fundamental role played by texts for interpreters’ commitments,17 sentences R in $\mathtt{dTRAIT}(a,R,n,T)$ provide a characterization of characters in the context of the text T, i.e., these sentences focus on the information present in T in either an explicit or implicit way (made explicit via inference possibly grounded on additional knowledge that interpreters may have). According to what we said in the previous section, this means that sentences R could cover different sorts of properties. With respect to the debate on internal vs. external properties (Section 2), it remains difficult to draw a clear-cut distinction. Consider the example of Superman. If the text does not specify his nature, an interpreter a may classify him as either fictional or real, based on the available information. This choice is relevant, because it affects what can be inferred on Superman, since a’s knowledge may contain rules applying only to either real or fictional entities. If a classifies Superman as fictional, then it can be inferred that he was created by an author. In principle, in our approach, properties commonly considered as external like being created can be part of a diagnostic characterization; hence, from this perspective, our approach is not strictly committed to a internalist view on diagnostic traits, and we leave it up to interpreters to decide which properties to use to characterize individuals.

4.2. Relations Between Characters

By comparing the properties identified by scholars for the stipulated characters, we can understand what they reveal about the relations between them. To achieve this goal, we can establish a number of comparison criteria. We will focus only on two main cases—one for plain agreement and one for plain disagreement—to demonstrate the general strategy. Various intermediate positions, which consider different property types (whether diagnostic or not), could also be established.

Recall that in the example based on Eco (2009), Emma Bovary in Woody Allen’s short story is the same character as the one in Flaubert’s novel, since—in his interpretation—she keeps the same diagnostic properties in both texts. In our framework, this means that, given the diagnostic traits that Eco individuates—through sentences D—for (the character named) ‘Emma Bovary’ in Flaubert’s novel, and the diagnostic traits that he individuates—through sentences E—for ‘Emma Bovary’ in Allen’s short story, a connection between D and E must be established.

The following criterion (c1) is used for the equivalence (for a given interpreter) between the diagnostic properties of names in different texts:18

(c1)

Suppose that R are the sentences individuating the diagnostic traits for the character named n in T according to a (i.e., $\mathtt{dTRAIT}(a,R,n,T)$), and S are the sentences for the traits of m in U according to b (i.e., $\mathtt{dTRAIT}(b,S,m,U)$). Sentences R and S are equivalent for interpreter c if and only if: $\left(i\right)$ according to c, in the context of the text T the sentences R are equivalent to the sentences S once m has been substituted with n; and $\left(ii\right)$ according to c, in the context of the text U the sentences S are equivalent to the sentences R once n has been substituted with m.

In other terms, according to criterion (c1), interpreter c is able to infer the equivalence of the diagnostic properties (attributed to the characters named n and m) as they are selected by interpreters a and b. Notice that c may disagree on the choices made by a and b, but c can evaluate whether, from their perspective, the properties attributed to the characters named n and m are equivalent.19

It is important to stress that the equivalence of diagnostic properties has an epistemological and subjective dimension: diagnostic properties are not equivalent per se, but are considered as such by c. In the example of Emma Bovary, the criterion (c1) says that when Eco maps the name ${\mathrm{EmmaBovary}}_{\mathrm{GF}}$ (i.e., ‘Emma Bovary’ in the context of the text by Flaubert) onto Allen’s text, he judges that the diagnostic properties attributed to the character in this latter text are equivalent to those attributed to the name placed in its original text; the same goes for ${\mathrm{EmmaBovary}}_{\mathrm{WA}}$ (‘Emma Bovary’ in Allen’s text) in the different context of Flaubert’s Madame Bovary.

The equivalence of the diagnostic properties associated with names in texts provides important information about the relation between the stipulated characters. This resembles the case of identity in the philosophical literature, although equivalence does not necessarily imply character identity for (at least) the following reasons.

First, in literary practices, diagnostic properties are not always conceived as both necessary and sufficient conditions. Consider a character who is only loosely sketched in a text. For such a character, it may be possible to identify some necessary properties, but not a definitive characterization. Nevertheless, literary scholars may wish to identify some of its unique properties and compare them with the properties of characters in other texts. Second, scholars’ interpretations are usually incomplete because they focus only on aspects of characters that are relevant to their research interests and objectives. No matter how fine the granularity of the diagnostic characterizations, they remain only partial representations of the corresponding characters. This consideration reinforces the idea that diagnostic characterizations can seldom be considered as sufficient conditions. Third, as a consequence of the above, scholars involved in literary debates may not share the same ontological commitments. It is therefore possible that they agree on the selection of diagnostic properties, whereas each interpreter makes sense of what others claim with reference to their own domain entities. Given these three considerations, one could argue that when different names are ascribed equivalent diagnostic properties by interpreters, there is a good chance of identifying them, as no evidence has emerged to distinguish between different characters.

Let us now examine the case where interpreters are able to infer information about the context in which texts and their characters were created. In this case, even if authorship is not part of the diagnostic characterization, the equivalence of diagnostic properties associated with names in texts by different authors might be interpreted as a case of “borrowing”, rather than of identity. For instance, one could claim that Allen’s character of Emma Bovary borrows from Flaubert’s Emma Bovary; hence, although these two characters have equivalent diagnostic properties, they are not identical because of the difference in their authorship. To avoid instances of coincidental overlap in diagnostic properties, stating borrowing relations requires interpreters to hypothesize about how texts are connected. For example, one must assume that Allen intended to borrow Flaubert’s character. It is important, however, to stress that, in our approach, it is always up to interpreters to decide on how texts and their authors are possibly related.

The case of borrowing is based on the preservation of diagnostic properties. Another case is when diagnostic properties are either expanded or reduced. Following criterion (c1), this could be done by introducing a relation of (partial) overlap (as opposed to equivalence) between properties.

Analyses of these sorts can be applied independently to property types, whether diagnostic or marginal, to identify similarities between named characters. In addition, one could apply different formal methods, including (quantitative) similarity metrics, to compute how characters relate. This may be achieved by reusing strategies adopted in cognitive sciences, as well as knowledge representation and reasoning (Porello and Endriss 2014; Porello et al. 2018; Smith et al. 1988). This line of research is explored by Porello et al. (2025), although it requires further work to integrate explicit reference to diagnostic properties. Also, by clustering characters sharing common properties, one may individuate recurrent “literary patterns.” To outline an example, Don Juan (Don Giovanni) is a character sometimes understood as a great universal of worldwide literature (Macchia 1991). Clearly, one has to distinguish the individual character in Tirso de Molina’s El burlador de Sevilla (1630) from the character in Byron’s Don Juan (1819), or in Da Ponte’s and Mozart’s Don Giovanni (1787), among others. By analyzing the similarities in their properties, one may abstract from them the “core” that the individual characters share. In a sense, this core is a character which is not found in any text, but is formed a posteriori by clustering the shared properties identified by interpreters across individual characters. For instance, the characters of de Molina’s Don Juan and Da Ponte/Mozart’s Don Giovanni share this common core, which is then “expanded” by means of the specific properties through which each individual character is characterized.20 Whereas throughout this paper we have focused on individual characters, our analysis could be extended and refined to capture character-types—like the hero, the femme fatale, the detective in 20th century novels, etc.—possibly analyzing how such types are created and characterized.

Finally, the analysis of disagreement between interpreters would deserve an extended treatment on its own. For the sake of the present discussion, an interesting and borderline case is when the disagreement is due to the incompatibility between all the diagnostic properties ascribed to a named character. In a similar manner to criterion (c1), we can introduce the following criterion:

(c2)

Suppose that R are the sentences individuating the diagnostic traits for the character named n in text T according to interpreter a (i.e., $\mathtt{dTRAIT}(a,R,n,T)$), and S are the sentences for the traits of m in U according to b (i.e., $\mathtt{dTRAIT}(b,S,m,U)$). Sentences R and S are incompatible for interpreter c if and only if: $\left(i\right)$ according to c, in the context of the text T the sentences R are incompatible with the sentences S once m has been substituted with n; and $\left(ii\right)$ according to c, in the context of the text U the sentences S are incompatible with the sentences R once n has been substituted with m.21

Incompatibility here means that, according to c’s conceptual system, the properties ascribed through the sentences to the named characters n and m generate logical inconsistencies. For example, if the (diagnostic) properties being an historical character and being a fictional character are declared in c’s conceptual system as being incompatible, they cannot be ascribed to the same character. This suggests that interpreters might not refer to the same character if the diagnostic properties they select are incompatible.22 Nevertheless, as mentioned above, even for these cases, one could compute similarities between (incompatible) characters to study how they relate. Once again, this analysis can be generalized by taking marginal properties into account.

4.3. Fictional Characters and Real People

A further interesting case concerns possible relations between fictional characters and real people, e.g., Napoleon in Tolstoj’s War and peace, and Napoleon as the historical emperor of France. In principle, the selection of (diagnostic) properties can also be applied to real people, e.g., to compare them with literary characters. This is common in literary criticism, where scholars sometimes wonder if certain characters are inspired by real people (an example is Barolini (2000) on Dante’s character of Francesca).

As mentioned, whether a figure is interpreted as real or fictional can depend on various factors. First and foremost, it depends on the interpreter’s attitude towards the texts. For example, if a name appears in a text recognized as having historical value, this could serve as a cue to attribute real existence to the corresponding figure. A similar reasoning can be applied to names found in texts regarded as fictional. Clearly, there may be cases where interpreters cannot determine whether a text is real or fictional. In such cases—this being a second aspect to consider—a suggestion of fictionality could emerge from the information interpreters glean from the text, especially when this information conflicts with their knowledge on real entities (as in the case of Superman). In the case of highly realistic texts, when there is no clear indication of their status, interpreters may struggle to definitively distinguish between their real or fictional nature. In such situations, they may need to employ alternative strategies, such as comparing the text to others in order to identify commonalities and, consequently, gain insight into its content.

A brief note on the relation between real people and fictional characters: what emerges from literary contexts is that fictional and real entities are rarely regarded as the same. Characters, as fictional constructs, are typically shaped by their authors to fulfill specific narrative roles within literary works. Therefore, even if a fictional character is, for instance, inspired by a real person, the character remains a distinct entity, tailored to meet the needs of the story rather than serving as a precise reflection of reality.

5. Conclusions

In this paper, we present an approach to studying the relations between literary characters that combines analytic philosophy and literary studies. The central idea is to examine how interpreters understand and characterize characters, particularly through the selection of properties, among which diagnostic properties play a key role. To achieve this, we proposed to analyze (public) statements through which interpreters express their interpretation of a text and its characters, without delving into the interpreters’ underlying ontological commitments. This methodological choice stems from the observation that scholarly discussions rarely make explicit reference to specific ontological frameworks—if any are assumed at all. In line with this perspective, we have also adopted a stipulative stance regarding the existence of literary characters, setting aside the traditional realism vs. anti-realism debate in analytic philosophy, as this approach seems to better reflect the way scholars in literary criticism engage with characters.

It is important to emphasize that diagnostic properties, as we have conceived them, are not attributes that characters inherently possess—whether because of their textual origins or authorial intentions. On the contrary, they are selected by interpreters in relation to texts, which means that characterization is inevitably subjective, shaped by the knowledge, skills, interests, and interpretive frameworks of scholars, among other criteria.

Admittedly, the conclusions about character relations that emerge from our study are weaker than those derived from other theories (Kroon and Voltolini 2023). For instance, it is difficult to establish identity criteria, because relations between characters depend on the interpretations provided by specific (groups of) interpreters. Hence, even if identity criteria were available, identity relations would remain relative to particular interpretations. Moreover, as discussed throughout this paper, scholars cannot often specify necessary and sufficient criteria for identity, as they may have access only to a limited set of textual evidences from which to derive conclusions. At best, when different interpreters converge in attributing equivalent diagnostic properties to characters appearing in different texts, we can conclude that there is no evidence to treat them as distinct. The absence of evidence is not, however, equivalent to a proof of identity, and such conclusions remain contingent on the scope and depth of interpretation.

Despite this apparent weakness—rooted in our reliance on subjectivity, situatedness, public discourse, linguistic skills and practices, and contextual knowledge—our approach seems to align well with real-world interpretative practices. In scholarly discourse, experts tend to rely on interpretive hints and patterns, rather than on strict definitional criteria, when establishing relations between characters. Our contribution should be thus understood in this light, as an attempt to offer a systematic way to elaborate and analyze characters based on interpreters’ claims. This does not mean that our current work aims for a mechanical procedure that, given multiple named characters, identifies their diagnostic properties. On the contrary, we have claimed that it is up to the interpreters to provide the properties they deem important with respect to the characters being interpreted. Once the characterizations are set, our framework can identify and explore character relations based on specific criteria. Further research, especially on our case studies, is necessary to refine and expand upon this proposal, including the identification of more specific types of relations (e.g., parody, adaptation) whose analysis can be useful for interpretation practices.