Concepts of Typicality in Literary Studies

1 Introduction

Literary scholars occasionally use the term ›typical‹. Many literary histories, for instance, contain claims suggesting that a feature is typical of an author’s work, a text is typical of a genre, a genre is typical of a period, and so on. Here is an example:

As is typical of the comedy genre, the plot [of Gustav Freytag’s play Die Journalisten, M.K.] is dominated by intrigue, some of the characters have speaking names, in the final scenes all the entanglements are resolved (planned long in advance) and there is a wedding. (Stockinger 2010, 201)^[1]

The quote says that a work (Gustav Freytag’s Die Journalisten) exhibits certain features (a plot dominated by intrigue, etc.) and that these features are typical of a genre (the comedy). Certainly, the statement is informative and intuitively comprehensible. On closer inspection, however, one might wonder what exactly the term ›typical‹ means in this instance. Does it mean that most comedies share the features mentioned? Or perhaps that these features occur more frequently in comedies than in other genres? Or is ›typical‹ supposed to mean that (at least some of) these features contribute to a text belonging to the category ›comedy‹ in the first place? It is not obvious which meaning was intended, or at least it is not explicit. It may also be the case that the term ›typical‹ is intended to encompass all of the meanings mentioned above, or a combination thereof, or an entirely different meaning.

The aim of this contribution is to clarify the concept of typicality in literary studies. I want to show that the term ›typical‹ can have (and is used in) different meanings, and that very different phenomena may appear typical, depending on which meaning is employed. The distinctions proposed in this paper should help to speak more precisely about typicality, both in one’s own research and in reconstructing what others might have meant when they used the term.^[2]

There is hardly any previous research on the use of the term ›typical‹ in literary studies. While some studies examine how literary texts represent ›the typical‹ (for an overview, see Winkler 2007), this research does not focus on clarifying the term ›typical‹ as an analytical category in sentences such as ›Text A is typical of author B‹. More relevant are reflections on typicality in prototype theory and its application in literary studies. However, as I will discuss, there are references to typicality in literary studies that are not captured by prototype theory, and even when literary scholars explicitly refer to prototype theory, they may mean different things when using the term ›typical‹. Finally, there are some remarks on typicality in relation to other disciplines (see, e.g., Trap-Jensen 1996; Wilhelm 2022), but without further investigation it would remain unclear whether the meanings explicated in these contributions play any role in literary studies.

I confine my remarks to literary studies, but include examples and methods from computational literary studies, respectively. This is not only useful in order to cover the spectrum of literary studies as comprehensively as possible, but also because computational approaches are sometimes helpful in providing particular explicitness. All explanations will apply equally to both computational and non-computational approaches.

In Section 2, I provide some preliminary clarification of what I call typicality propositions. Section 3, the main focus of the paper, distinguishes five concepts of typicality. The fact that exactly these five concepts are discussed, and that they are explicated the way they are, is due to two main criteria. On the one hand, the concepts should be close enough to real language use, i.e., not introduce completely idiosyncratic notions of typicality, but cover (roughly) what scholars mean when they use the term. At the same time, the concepts should be fruitful from a theoretical perspective, i.e., they should be clearly delimited and internally consistent.^[3] In Section 4, I apply the concepts from Section 3 to some use cases that require a little more explanation: modeling typicality via machine learning, and prototype theory.

2 Typicality Propositions

Sentences such as ›This feature is typical of parodies‹, ›This author is not very typical of modernism‹, ›This text is more typical of Kafka than that text‹, or ›Is this text a typical ballad?‹ contain what I call typicality propositions. Typicality propositions are propositions about something being typical (or having a certain degree of typicality) of something. When analyzing the various meanings of the term ›typical‹ in Section 3, I will focus on its usage in the context of such propositions. Let us consider some basic components and properties of typicality propositions, and the way the term ›typical‹ functions in these contexts.

Target Base: When something is said to be typical, it is necessary to specify what it is typical of. Elements are not typical per se, but only in relation to, for example, an author’s work, a genre, or a period. A text can be very typical of an author’s work, and at the same time not very typical of a genre, and so on. I will call the ›of what‹ part of a typicality proposition the target base. Usually, the target base is represented (explicitly or implicitly) by a text corpus. For example, if one wanted to study what was typical of literature from the second half of the 19^th century, the target base would be represented by a corpus of texts from that period.

Typical Elements: Typicality propositions can (and, for the sake of clarity, should) specify which elements are or could be typical. I call this part of a typicality proposition the typical element(s).^[4] Typicality propositions can deal with very different kinds of elements. For example, one might talk about the typicality of features, or of texts, or of authors, or of genres. It is possible to say ›This feature is typical of period X‹, but also ›This text is typical of period X‹, or ›This author is typical of period X‹, or ›This genre is typical of period X‹. All these kinds of statements appear (in these or similar forms) in literary studies.^[5] It may be worth clarifying what exactly the relationship is between different kinds of typical elements, e.g., between features and texts that are typical of the same target base. Obviously, there should be a close connection, but the precise relationship may be a matter of operationalization. From a theoretical perspective, I will not pursue this question any further.

Gradedness: In typicality propositions, the term ›typical‹ can be used in a graded or non-graded way. For example, one can make claims about which features or texts are ›typical‹ and which are not (non-graded), or claims about the ›most typical‹ features or texts, or about one feature or text being ›more typical‹ than another (graded). Both options are used in practice.^[6] One possible interpretation is that the non-graded way of speaking is a (sometimes useful) simplification in that it divides an originally continuous phenomenon (the degree of typicality) into discrete categories. The term ›typical‹ used in a non-graded way might mean the same as ›having a degree of typicality that is higher than X‹, much like the word ›intelligent‹ used in a non-graded way might mean ›having a degree of intelligence that is higher than X‹.^[7] In the next chapter, I will adopt not only the graded usage, but also, especially when the additional differentiation is not essential, the non-graded usage for the sake of simplicity.

Feature Base: Typicality propositions are usually based (explicitly or implicitly) on some features that are used to assess the typicality of the typical elements. If researchers make claims such as ›This text is typical of Kafka‹ or ›This author is typical of modernism‹, they must base their claim on an analysis of some criteria or features, such as the occurrence of a particular set of motifs and themes (›Considering motifs and themes, text A, but not text B, appears to be typical of Kafka‹), or stylistic and formal features (›Considering stylistic and formal features, text B, but not text A, appears to be typical of Kafka‹). I refer to the set of features used as criteria for assessing the degree of typicality as the feature base.^[8] As the examples indicate, different feature bases can lead to different results. Depending on whether researchers look at motifs and themes or at stylistic and formal features, different texts may appear to be typical.

I take the term ›feature‹ very broadly. In principle, not only textual aspects can be used as features to assess typicality, but also contextual aspects such as ›year of publication‹ or ›being published in an anthology‹, and even specific perceptions, judgments, or attitudes such as ›is praised by critics‹. Also, depending on the nature of the potentially typical elements, one might consider features of texts (this is what the previous examples referred to), or of authors (›has written many ballads‹, ›was born in the 19^th century‹, etc.), or of genres, and so on.

Specifying the feature base seems intuitive in the case of, say, texts or authors. A text or an author may appear more or less typical depending on which features are considered. But what about cases where the (potentially) typical elements are features themselves? At first glance, it may seem odd to ask what kind of feature base was considered when faced with statements such as ›Containing direct character speech is typical of ballads‹. However, even when the potentially typical elements are features themselves, it is sometimes useful to specify the feature base, i.e., the set of features considered. For example, if you want to identify all typical features, and the results do not include feature X, it makes a relevant difference whether this is because X was not even considered, or because X was considered but is not typical.

So far, I have made some basic clarifications about typicality propositions, but I have not yet tackled the question of what the term ›typical‹ means within these propositions. This, of course, is the crux of the matter and the main focus of this paper, as well as the subject of the next section.

3 Concepts

The meaning of typicality propositions depends very much on what exactly is meant by the term ›typical‹. I begin by outlining two criteria that are useful for drawing distinctions between concepts of typicality and then distinguish between five such concepts.

Contrastivity: It is possible to conceive of the term ›typical‹ in a contrastive or non-contrastive way. I call concepts of typicality contrastive when the identification of typical elements requires the consideration of phenomena that are not part of the target base but of some other base. For example, the term ›typical‹ in the question ›What features are typical of literature from around 1900?‹ is used in a contrastive way when it is necessary to contrast texts from around 1900 with other texts in order to answer the question. Let us call these phenomena beyond the target base the contrast base. What appears to be typical may depend a lot on the composition of the contrast base. For example, if the contrast base consists of texts from before 1900, different elements may appear typical than if the contrast base consists of texts from after 1900.

Concepts of typicality are non-contrastive when it is not necessary to consider phenomena beyond the target base in order to identify what elements are typical. For example, if you answer questions such as ›What features are typical of literature from around 1900?‹ by looking solely at texts from that period, you are using a non-contrastive concept of typicality. In the case of non-contrastive typicality concepts, there is no contrast base.^[9]

Feature openness: Some concepts of typicality are what I call feature-open, while others are not. This criterion has something to do with the feature base, i.e., the set of features used as criteria for assessing the degree of typicality (see Section 2). In the case of feature-open concepts, you can use any feature base for any target base. Any conceivable feature could be considered to have an impact on the typicality of elements. For example, one might focus on a particular set of words and ask which texts are typical of, say, ballads in terms of these words, while acknowledging that other texts might appear typical in the same sense of the term when considering other features (e.g., some other words, or formal features, or contextual features).

However, some concepts of typicality do not appear to be feature-open. In their case, the feature base is restricted or fixed. The application of these concepts leads to a situation where some features are seen as incapable of having a positive (or negative, or any) impact on how typical elements are for a given target base. The features are disregarded not just for pragmatic reasons, or because one happens to be uninterested in them (as may be the case with feature-open concepts), but because one assumes that these features are inherently irrelevant to the typicality of elements for a given target base. For example, one might adopt a concept of typicality which leads to treating features such as the frequency of a single word as incapable of affecting how typical texts are of the genre ›ballad‹, no matter how frequent or infrequent that word is.

The two proposed criteria do not claim to be sufficient for distinguishing all possible typicality concepts, as they clearly are not. They are, however, useful for delineating at least some of the relevant differences.

The remainder of this section is devoted to distinguishing five concepts of typicality. As indicated in the introduction, these concepts should be both fruitful from a theoretical point of view and used in practice. Most concepts are rather broad and encompass many subvariants. For each concept, I first explain its general idea and relate it to the criteria of contrastivity and feature openness, then provide some exemplary operationalizations, and finally give some real-world examples that use the term ›typical‹ in the sense of the concept at hand. Table 1 provides an initial overview.

Typicality as Centrality: This concept of typicality captures the intuition that an element is all the more typical the more it is ›common‹, ›widespread‹, or ›average‹ within the target base. These formulations are not identical since, for example, the average (mean) does not have to be identical with the most common (mode). Typicality as centrality therefore allows for many subvariants. However, there are also key commonalities. One important commonality is that typicality as centrality is non-contrastive. To identify what is typical in the sense of being central, one does not need to consider phenomena outside the target base, but only the target base itself. Moreover, typicality as centrality is feature-open, i.e., one can ask what is ›common‹, ›widespread‹, or ›average‹ in terms of any kind of feature base – be it form, content, contextual features or whatever else one happens to be interested in.

To make the concept of typicality more concrete, I will sketch some potential operationalizations for identifying typical features and texts,^[10] using the genre ›ballad‹ as an exemplary (and interchangeable) target base. These operationalizations, like those for other concepts of typicality discussed later, are not without alternatives, nor do they claim to be optimal. Their primary purpose is illustrative: to clarify the core idea behind the concept in question.

In terms of typical features, an extremely simple operationalization could be to assume that a (binary) feature is all the more typical the more texts share it. In this view, a good candidate for a feature typical of ballads would be something like ›is rhymed‹. It is likely that not all, but most ballads are rhymed, so being rhymed should be very common for ballads, therefore the feature ›is rhymed‹ is typical for ballads – so the reasoning goes. On the contrary, a feature like »contains the word ›king‹« would probably have a much lower degree of typicality. Perhaps only a minority of ballads contain the word ›king‹, so containing the word ›king‹ is rather an exception, so the feature »contains the word ›king‹« is untypical for ballads.

A possible point of contention is whether features that apply to 100 % of texts should also be called ›typical‹. This may concern features such as ›is a poem‹ or ›narrates a story‹, which potentially apply to 100 % of ballads (depending on one’s understanding of the term and the composition of the corpus representing the target base). If one wants to reserve the term ›typical‹ for features that do not apply to all texts, then one might avoid labeling such features as typical.^[11] This consideration also applies to the other concepts of typicality.

In terms of typical texts, typicality as centrality requires finding out which texts are ›average‹ for the target base or, metaphorically speaking, positioned at its ›middle‹ or ›center‹. One possible operationalization is to represent each text of the target base as a feature vector, measure the distance between each vector and all others (e.g., using Euclidean or cosine distance), and conclude that the smaller the mean (or median, etc.) of these distances, the more typical a text is.^[12] In less technical terms, a ballad would be considered more typical the more similar it is to other ballads, and less typical the more it differs from them.

There is evidence that literary scholars sometimes indeed understand typicality as centrality. A clear example is a computational study by Mark Algee-Hewitt and Erik Fredner (Algee-Hewitt/Fredner 2020) that investigates which texts are typical of the U.S. novel. Algee-Hewitt and Fredner conceive of typicality in a way that closely aligns with the concept of centrality, using an approach that resembles the operationalization just sketched, i.e., focusing on minimal distances between texts. They explicitly state: »we seek typicality in the […] form of the average« (ibid.). There are also typicality propositions in non-computational literary studies suggesting typicality as centrality. Consider this example: »As is typical of the majority of decadent literature (starting with Huysmans’ novel Làbas, the overpowering role model that can also be felt here), the fascination with the decadent is also articulated as a criticism of its consequences« (Sprengel 2004, 287). The typical is associated with the »majority«, which signals typicality as centrality. Here is another example: »While Lord Jim is considered aesthetically typical of its historical moment, Wuthering Heights is considered an aberration« (Shires 2000, 61). The contrast with »aberration« suggests that »typical« is intended to mean something like non-aberrant, normal, or average, which again corresponds to typicality as centrality. However, without explicit clarification, there remains some doubt as to what the last two quoted authors had in mind.

Typicality as Distinctiveness: From the perspective of typicality as distinctiveness, elements are all the more typical of the target base the more different or delimited they are in opposition to a contrast base. In this view, what is typical of, say, ballads is what distinguishes them from non-ballads. Importantly, typicality as distinctiveness – at least as I construct the concept here – refers to empirical differences rather than differences resulting from terminological explication. For example, whether a feature is ›distinctive‹ for ballads is not determined by whether that feature is important for explicating the term ›ballad‹ and for distinguishing it from ›non-ballads‹ on a terminological level, but by whether the feature is empirically more frequent (or more widely dispersed, or something similar) in ballads than in non-ballads (for further discussion, see typicality as category membership). Typicality as distinctiveness is obviously contrastive, since the concept necessitates comparing the target base with a contrast base. Furthermore, typicality as distinctiveness is feature-open, since distinctiveness can be assessed considering any feature base for any target base.

Here is a simple operationalization for binary features: the greater the value of ›proportion of feature X among ballads‹ minus ›proportion of feature X among non-ballads‹, the more typical the feature is of ballads.^[13] Now, in contrast to typicality as centrality, a feature like ›is rhymed‹ may no longer be typical, or at least a lot less typical. While most ballads are rhymed, perhaps an equal proportion of texts in the contrast base – maybe non-balladic poems from a similar period as the ballads – is also rhymed. In this case, being rhymed does not distinguish ballads from non-ballads, so it would not be typical of ballads. On the other hand, a feature like »contains the word ›king‹«, which was not typical before, may appear typical now. Even though only a minority of ballads contain this word, it may still be more common in ballads than in non-ballads. In this case, the feature distinguishes ballads from non-ballads, so it would be somewhat typical of ballads. Of course, the results just mentioned are only hypothetical, but they still illustrate that, in principle, it is possible for different concepts of typicality to lead to different features appearing typical.

In terms of texts, a possible operationalization of typicality as distinctiveness could be to calculate the centroid of the ballads and the non-ballads and subtract them in order to determine the direction in which the ballads ›move away‹ from the non-ballads, and then identify which texts move furthest in this ›ballad direction‹. According to this operationalization, the more a text has of what distinguishes ballads from non-ballads, the more typical it is of the genre ›ballad‹. Again, the corresponding results may differ from typicality as centrality. For instance, containing the word ›king‹, to stick with the example above, would tend to make a text less typical in terms of centrality, but more typical in terms of distinctiveness.

In literary studies, typicality does indeed sometimes seem to be understood as distinctiveness. Consider this quote: »Ideas of décadence and degeneration are typical of European literature around 1890« (Ajouri 2009, 181). It is of course possible that the author means to say that most of European literature around 1890 expresses ideas of décadence and degeneration (typicality as centrality). But perhaps even more likely is the interpretation that ideas of décadence and degeneration are typical of European literature around 1890 insofar as they play a more prominent role in this literature than in other literatures, e.g., European literature from earlier periods (typicality as distinctiveness). Here is a second example: »In literary modernism it is the novel, almost alone among genres, which is typically linked to […] innovations in the representation of temporality. Despite T. S. Eliot’s Four Quartets (1944) being about time, time is not a structural concept for poetry« (Banfield 2007, 48). One possible interpretation is that the author does not (or at least not exclusively) mean that most novels during modernism feature innovations in the representation of temporality, but that such innovations are more common in novels than in texts from other genres. This reading is suggested by the parenthetical »almost alone among genres« and the contrast with poetry.

Identifying cases of typicality as distinctiveness is easier when it comes to computational approaches. Examples include studies using so-called keyness measures. Keyness measures are sometimes explicitly linked to the identification of ›typical‹ elements. For example, a recent overview begins like this: »The purpose of a keyness analysis is to identify (lexical) items that are typical for a certain text variety« (Sönning 2024, 263).^[14] Most keyness measures work by contrasting the frequency, distribution or dispersion of items (often words) in different corpora. Accordingly, the cited paper continues as follows: The »text variety is represented by what is referred to as the target corpus, and typical items are then detected by way of comparison to a reference corpus, which serves as a baseline to foreground relevant structures« (ibid.). This approach is similar to the operationalization of typical features outlined above. Using the term ›typical‹ in the context of contrastive keyness measures can usually be interpreted as referring to typicality as distinctiveness.^[15]

Typicality as Category Membership: Another concept of typicality is related to category membership. Elements may be seen as typical if they belong to the target base to a high degree, or, in the case of features, if they contribute to membership in the target base. In this view, a text is a typical ballad if it belongs to the category ballad to a high degree and a feature is typical of ballads if it contributes to a text being a ballad. As the phrase »to a high degree« suggests, most uses of this concept of typicality assume that category membership is gradable, an assumption which may be debatable depending on the category.^[16]

Typicality as category membership is a contrastive concept. For example, to judge whether a text is a ballad, one must consider at least two classes, ballads and non-ballads. In terms of contrastivity, typicality as category membership is similar to typicality as distinctiveness. An important difference lies in the fact that typicality as distinctiveness (as well as typicality as centrality) operates at an empirical level, focusing on what can be empirically observed (e.g., in terms of differences between ballads and non-ballads), whereas typicality as category membership operates – at least in the way that I am conceptualizing it here – at a terminological level, focusing on what is linked to the very meaning of terms such as ›ballad‹. Furthermore, typicality as category membership is not feature-open. The concept does not allow to use any conceivable feature base for any target base (i.e., to consider any set of features as criteria for assessing the typicality of elements for any category). Instead, the concept I am constructing prescribes to use exactly those features that are constitutive for membership in the respective category. Which features are constitutive of category membership depends on the category and is, of course, controversial in many cases. The most diverse set of features can be proposed as constitutive for a particular genre, period, etc. However, no matter what the set of constitutive features ends up being, once one assumes that such a set exists for a given category, this set must be the feature base, while all other features cannot be part of the feature base, since they are irrelevant to category membership. As an example, consider the feature »contains the word ›king‹«. If typicality is understood as distinctiveness, one can use any feature base imaginable for any target base, so it is entirely reasonable to include this feature (or any other feature) in the feature base, and it may turn out that containing the word ›king‹ is typical in the sense of distinctive. However, it is plausible to assume that the feature »contains the word ›king‹« cannot be part of the feature base for typicality as category membership (regarding the category ›ballad‹) and cannot be typical in the sense of contributing to category membership. The reason is that the word ›king‹ does not seem to have a positive (or negative, or any) impact on a text belonging to the category ›ballad‹, no matter how frequent or infrequent the word is. It may still serve for identifying ballads (e.g., in a machine learning context), but it does not really contribute to a text being a ballad. Instead, constitutive features such as ›narrating a story‹ are responsible for a text being a ballad, and it is on these kinds of features the concept of typicality described here focuses exclusively.^[17]

How to determine which features are constitutive of category membership is a separate issue and cannot be treated in detail here. A researcher could, for example, perform a terminological explication and argue for the (debatable) position that the features ›is a poem‹, ›narrates a story‹, ›has multiple stanzas‹, and ›contains direct character speech‹ contribute to membership in the category ›ballad‹.^[18] In this case, these features form the feature base and are at the same time typical. They are not identical to the central and distinctive features. For instance, while ›is rhymed‹ is probably central and »contains the word ›king‹« may be distinctive, none of these contribute to category membership, at least in this example. One may also argue that the features have different weights for category membership, e.g., ›is a poem‹ and ›narrates a story‹ may be deemed even more important than ›has multiple stanzas‹ and ›contains direct character speech‹. Since this is not a paper about the ballad, I will not go into more detail.

Moving to the text level, a simple way to operationalize the typicality concept at hand is to consider a text as more typical the more features it has that contribute to its category membership. Again, the operationalization can be made more nuanced by allowing for the possibility of different feature weights. Since the feature base will (in most cases) be different from the feature base used to assess centrality or distinctiveness, different texts may appear typical.

As usual, I will give some examples of typicality propositions from the real world to show that the term ›typical‹ is sometimes actually understood in a way that refers to category membership. Here is an example from a literary history: The writer Georg Heym »either uses the sonnet form or transgresses the typical signature of the ballad in other ways (as in the poem Pilatus). Werfel wrote a touching Ballade vom Tode der Kinderfrau (1913) and in the comprehensive volume Der Gerichtstag (1919), five – quite untypical – poems are labelled as ballads« (Sprengel 2004, 602). The reference to the sonnet, as well as to the »labeling« as ballads, may indicate that these texts are not very typical in the sense that their very belonging to the category ›ballad‹ is at issue – an interpretation that makes sense when reading Heym’s and Werfel’s poems. But of course, without clarification, it is not certain that the researcher had this concept of typicality in mind. A more prominent example can be found in Hempfer 2014 (see also Hempfer 2010). The author deals with genre theory and discusses, among other things, prototype approaches. As an example, he provides an explication of ›the narrative‹ as a ›mode of writing‹ using »the Platonic-Aristotelian speech criterion« as »a prototypical conceptualization« (Hempfer 2014, 414 sq.). Here, the »prototypical« feature is meant to contribute to category membership. I will provide additional remarks on prototype theory in Section 4.

Understanding typicality as category membership is not equally intuitive for all kinds of target bases. For example, while one can say ›This text is typical of genre X in the sense that it belongs to X to a high degree‹, and perhaps also ›This text is typical of literary movement Y in the sense that it belongs to Y to a high degree‹, it is less intuitive (and perhaps not even reasonable) to say ›This text is typical of author Z in the sense that it belongs to the works written by Z to a high degree‹, or ›This text is typical of 19^th century literature in the sense that it belongs to the works written in the 19^th century to a high degree‹. The reason is that membership in the categories of ›works written by Z‹ or ›works written in the 19^th century‹ is (in most cases) decided by a single binary feature: whether a text is written by Z or in the 19^th century or not.

Typicality as Historical Relevance: Another option is to link typicality to an element’s relevance for the history of a genre, a period, an author’s work, and so on. Of course, ›relevance‹ can mean many different things and must be specified in each use case. The clarification of this term might lead to assumptions such as ›Typical is what sets a standard for later authors‹, or ›Typical is what was very influential‹, or ›Typical is what was formative for the genre, period, etc. during its emergence‹. Typicality as historical relevance is a non-contrastive concept because it focuses exclusively on the target base, without considering the historical relevance that elements may have had for other genres, periods, etc. The concept is not feature-open, since the features used to assess the typicality of elements consist exclusively of features indicating the historical relevance of those elements.

Regarding the identification of typical texts, one could specify features that should serve as indicators of historical relevance and compare how different texts score on these indicators. The latter might include something like the number of times a text was referenced by later authors, or whether the text was published during the emergence of its respective genre, literary movement, etc. When it comes to typical features, things are a bit tricky. Features like ›number of references‹ or ›date of publication‹ may be used to judge the typicality of texts, but it does not follow that they are themselves typical. Indeed, it seems odd to say that being published during the emergence of the genre ›ballad‹ is typical of ballads. In contrast to the other types of typicality discussed so far, the typical features (in the sense of the historically relevant features) are not necessarily identical with the features that increase the typicality of the typical texts. Instead, typical features may be textual properties, such as having a particular theme or rhyme scheme, that have influenced many other writers. To find out whether a feature is typical in this sense, one could, for example, investigate how quickly and how often writers copied the feature after it first appeared. Obviously, the typical (historically relevant) features are likely to be exhibited by the typical (historically relevant) texts. It is possible, but not necessary, that these features are similar to the central or distinctive features or the features that contribute to category membership.

Again, there is evidence that the term ›typical‹ – or, more specifically, ›prototypical‹ – is sometimes used in literary studies to refer to historical relevance. Consider this statement: »The historical findings of literary genres show that their history is, on the one hand, decisively determined by norm-forming works (prototypes) and, on the other hand, characterized by the mutual complementarity of genre expectations and work responses.« (Voßkamp 1977, 30) Here, prototypes are identified as historically relevant works, insofar as the prototypes influence the history of a genre by creating norms (cf. Gymnich 2010, 152; Remele 2021, 59 sq.).

Typicality as Mental State: The term ›typical‹ can be used to designate elements towards which certain people adopt certain mental states. This is an extremely broad notion and both what is meant by ›certain people‹ and especially ›certain mental states‹ has to be clarified in each use case. For example, one can use the term ›typical‹ to mean things such as ›Typical is what first comes to mind when (professional, non-professional …) readers think of the target base‹, or ›Typical is what readers consider a good example of the target base‹, or ›Typical is what readers expect when they plan to read texts from the target base‹. An important commonality of these options is that the term ›typical‹ is used in a way that cannot be verified or falsified by pointing to externally observable features (rhyme scheme, publication in an anthology, etc.), but only to (indicators of) mental states. Relatedly, a feature that people think of first, or find expected, would indeed be typical (from the perspective of typicality as mental state), even if it was de facto neither central, distinctive, significant for category membership, nor historically relevant. When typicality is understood as a mental state, mental states are not treated as opinions about typicality to be judged right or wrong, but as what constitutes typicality in the first place.

The criteria of contrastivity and feature openness are difficult to apply to typicality as mental state. First, it is necessary to distinguish between the researchers who study people’s mental states and the people who have those mental states. I will start with the researchers. In terms of contrastivity, perhaps one could say that the concept is non-contrastive insofar as researchers probably do not have to consider a contrast base (e.g., what people expect of non-ballads), but only the target base (e.g., what people expect of ballads). However, they would not focus on the target base directly but on mental states related to the target base. In terms of feature openness, typicality as mental state is perhaps to be understood as not feature-open from the researchers’ perspective. They have to focus exclusively on mental states, while other aspects (especially features of the actual texts) do not have a direct impact on what counts as typical. When focusing on the people that adopt the mental states, the question shifts to whether the mental processes that shape these states are contrastive and feature-open. This, however, is difficult to say. The answer seems to depend highly on which exact mental state is at issue and is a question for empirical research.

To find out what features or texts readers think of first, find expected, etc., it would probably be necessary to carry out psychological experiments or surveys. As I do not want to speculate on their outcomes, I will not give any exemplary results. It may turn out that the results are similar to those from the perspective of other concepts of typicality, but this is not certain.^[19]

An example of understanding typicality as mental state can be found in Wolf 2005. The author wants to provide an explication of the term ›lyric‹. He aims to achieve his goal by using a »cognitive and prototypical approach« (ibid., 33) and by listing »prototypical components of the lyric« (ibid., 38). Certainly, his aim is closely tied to typicality as category membership, but he also incorporates typicality as mental state. For example, Wolf seems to argue that features are typical at least in part because they »seem to be present in the minds of informed readers […] as the outcome of a small empirical survey has shown« (ibid.). Furthermore, he thinks that in order to find out what the prototype of the lyric is today, »it would be rewarding for empirical literary studies to devise and carry out extensive test series in order to find out about what the public today feels to be lyric and due to what features« (ibid., 44).^[20]

4 Applications

I add some brief remarks on how the concepts of typicality distinguished so far might be applied to two specific (and very different) contexts of typicality propositions: modeling typicality as the prediction probability of a machine learning classifier, and prototype theory. In both cases, it is not obvious which concepts of typicality are at issue.

There have been some attempts to relate the probability with which a machine learning classifier assigns a class to a text to the text’s typicality for that class (especially Le Mens et al. 2023b, compare also Schröter 2023; Schröter 2024; Henny-Krahmer 2023, 64–66).^[21] At first glance, these approaches might appear to model typicality as category membership. However, I would like to argue that the machine learning approaches are better understood as a particular variant of typicality as distinctiveness. This is at least the case when the classification is based on features that are not constitutive for category membership.

Classification in machine learning usually requires not only elements that belong to the target base (e.g., ballads), but also other elements (e.g., non-ballads) to facilitate discrimination. This corresponds to the contrastive nature of both typicality as distinctiveness and typicality as category membership. An important difference, however, is that machine learning classification appears to be feature-open, just like typicality as distinctiveness, but unlike typicality as category membership. In principle, any type of feature can be considered in the classification process, including, for example, most frequent words, i.e., not just features that are constitutive of category membership from a terminological point of view. While category membership as conceived in this paper focuses on the degree to which (or whether) a text is a member of a category in terms of constitutive features, the machine learning approach seems to measure how easy it is to identify a text as a member of a category using any features suitable for that task.^[22]

The interpretation of machine learning presented here hinges on the way typicality as category membership is understood. If one also includes the degree of being able to identify a text as a member of the target base, rather than just the degree of being a member in terms of constitutive features, then the machine learning approach could indeed be said to model typicality as category membership. This perspective is entirely valid, provided one accepts that, in this case, the distinction between typicality as category membership and typicality as distinctiveness becomes somewhat blurred.

Regarding prototype theory, my main point is that what exactly is meant by ›(proto)typical‹ or what is the basis for something being ›(proto)typical‹ can be answered in many ways even within prototype theory. This is not a new insight (see e.g., Geeraerts 1989, also Taylor 1989, 52 sq.), but I want to highlight that it also applies to the use of prototype theory in literary studies. The examples above have shown that the term ›prototypical‹ can be used in literary studies to refer to category membership (Hempfer 2014), but also to historical relevance (Voßkamp 1977) and to specific mental states (Wolf 2005). In doing so, at least Hempfer 2014 and Wolf 2005 explicitly refer to prototype theory.

Nevertheless, there seems to be a special connection between prototype theory in literary studies and typicality as category membership. Although not all references to prototypes or prototype theory in literary studies refer exclusively to this notion of typicality, conceiving of typicality as category membership in most cases suggests a prototypical approach, or at least a characteristic assumption of prototype theory: that category membership is gradable. This assumption is probably part of most uses of the term ›(proto)typical‹ with respect to category membership, and it seems to be shared by all flavors of prototype theory, at least in linguistics (cf. Geeraerts 1989, 592–602). Note that the other concepts of typicality do not require or even suggest a prototypical approach. In fact, they are compatible with various alternative approaches, such as modeling category membership in terms of necessary and sufficient conditions or in terms of family resemblances. For example, you could determine the category membership of ›ballads‹ using necessary and sufficient conditions and still ask, without any contradiction, what ballads are typical in the sense of most common regarding word usage (typicality as centrality), different to non-ballads regarding rhyme scheme (typicality as distinctiveness), important to the genre’s history (typicality as historical relevance), or thought of first by readers (typicality as mental state).

5 Conclusion

The aim of this paper was to show that in literary studies, the term ›typical‹ can have (and is used in) different meanings, and that very different phenomena can appear typical, depending on which meaning is employed. Therefore, it seems useful to clarify what exactly is meant when using the term ›typical‹. The paper has offered several resources to assist in such clarification, including terminological proposals for typicality propositions in general and, most importantly, the distinction between five concepts of typicality. Instead of saying, for example, ›having multiple stanzas is typical of ballads‹, one could be more precise by saying, ›having multiple stanzas is typical of ballads insofar as most ballads have multiple stanzas‹ (typicality as centrality), or ›insofar as this feature is more common in ballads than in non-ballads‹ (typicality as distinctiveness), or ›insofar as this feature contributes to a text belonging to the genre‹ (typicality as category membership), or ›insofar as this feature was formative in the emergence of the genre‹ (typicality as historical relevance), or ›insofar as ordinary readers tend to expect this feature in ballads‹ (typicality as mental state).^[23] Whether it is truly helpful to continue using the term ›typical‹ for all these diverse concepts, or whether one should strive to narrow down the range of possible meanings, is another question and is open to debate.

The distinction between different concepts of typicality opens up intriguing avenues for research. For instance, one might explore which of these concepts is most commonly applied in literary studies practice. Another promising direction involves comparing the outcomes of different conceptualizations – such as examining whether the elements readers expect (typicality as a mental state) align more closely with the most central elements (typicality as centrality) or the most distinctive ones (typicality as distinctiveness). Additionally, the concepts of typicality could prove valuable in analyzing literary change. For example, it would be interesting to investigate how the central and distinctive features of a genre evolve over time. Finally, it is highly likely that other concepts of typicality beyond those identified here are used in literary studies which could be explored and described.