Critical Modelling and Measurement

As both Anita Traninger and Bart Soethaert note in their contributions to the work on ‘Critical Modelling’ in Research Area 5 of EXC 2020 Temporal Communities, literary studies has always involved—and been conducted as a practice of producing—models: including of what constitutes literature in general, as well as literary periods and traditions, canons and genres, and authors and readers. My contribution focusses on modelling as it has come to prominence in the past twenty years as a specific feature of computational literary studies, or CLS. Computational models are often taken to epitomise the problems that make literary scholars reluctant to accept models in general. As Soethaert elaborates, these problems arise from the representational status of models, including their corresponding ‘factual and/or universalist aspirations’ and limited capacity to ‘respond to a dynamic reality’. While Soethaert examines the other critical practices that literary models can offer alongside representation, I argue that this representationalist paradigm undermines and is incompatible with critical modelling. It is incompatible because it confines modelling to an atomistic and dualist ontology that literary and other humanities scholars have long rejected. Often called ‘Cartesian dualism’,¹On Cartesian dualism as a theological misappropriation of Descartes’ theory of matter see Full reference in Zotero Library. this ontology divides the world into subjects or objects, thoughts or things, meaning or matter, including by placing models and other representations between these supposed opposites. In place of representation, I propose measurement as an enabling framework for critical modelling in literary studies, including for exploring how existing disciplinary practices might work productively with emerging computational ones.

For those who have followed the direction of arguments in CLS in the past two decades, my focus on measurement might seem anachronistic. One of the key developments in this field during this time has been a shift in conceptions of modelling away from positivist measurement to perspectival constructions. Where the field once measured literature in the past to gain a direct and supposedly—or at least relatively—transparent picture of what happened (let us call this the positivist approach) (e.g. Full reference in Zotero Library, Full reference in Zotero Library), now it engages in an ongoing and subjective process of ‘coming-to-know’ Full reference in Zotero Library, constructing an understanding by gradually and iteratively building up and refining our representations to achieve more complex understandings (the perspectival approach) (e.g. Full reference in Zotero Library, Full reference in Zotero Library). This latter approach, which recognises, among other things, the entanglement of the present in representations of the past, and of subjectivity in representations of objects, seems infinitely preferable to the absolutism of the former. It certainly seems closer to something that might be called critical modelling for the humanities. Yet both positivist and perspectival modelling rely on the same underlying assumption, which Karen Barad calls representationalism: namely, the belief that ‘representations and the objects (subjects, events, or states of affairs) they purport to represent are independent of one another’ Full reference in Zotero Library.

While the positivist approach claims that models re-present objects-in-the-world fully and transparently, and the perspectival one that models refract those objects through the lens of a particular historical moment or cultural context, in both cases, objects (including subjects) remain separate from and unaffected by representations (of or by them). Whether the subject knows/sees everything (positivism), or knows/sees only from one place or time (perspectivalism), what is in the mind and the world remains separate. Though the model allows the subject to know the world, and to express their knowledge in material form, the necessity of such mediation simultaneously embodies the different nature of these realms. Two approaches that are routinely presented as opposites in CLS can thus be understood as two sides of the same Cartesian coin. In particular, the perspectival approach, intended to challenge and dethrone the positivism claimed by the god-subject, leaves us with many little gods—now situated, individual, specific—but still at the centre of epistemology, now relativised. Perspectivalism thus multiplies the individual, independent, enlightened subject rather than challenging the terms in which that individuality, independence, and enlightenment are claimed.

Many paradigms across the humanities and social sciences challenge such dualism, including poststructuralist and posthumanist paradigms in literary studies, and what has been called the ontological turn in social science disciplines such as anthropology, economics, politics, and sociology.²For a discussion of the “ontological turn” in anthropology, for instance, see Full reference in Zotero Library; for indicative work in this vein in politics see Full reference in Zotero Library, in economics see Full reference in Zotero Library, or in sociology see Full reference in Zotero Library.Such paradigms have had substantial influence on discussions of computational modelling, as explored in a recent article by Guilherme Sanches de Oliveira Full reference in Zotero Library, which argues that representation fails as a framework for modelling on both epistemological and ontological grounds. On the one hand, what he calls ‘mind-independent’ conceptions of modelling (the idea that models relate to their targets by virtue of some qualities of the model or target; what I have termed the positivist approach in CLS) fail on ontological grounds, in that philosophers have not been able to resolve the asymmetry of modelling (why a model represents a target system but a target system does not represent a model). They have also been unable to supply an overarching framework to explain the relationship between model and target (neither isomorphism nor similarity work under all circumstances, even as either might pertain to particular fields and modelling practices). On the other hand, ‘mind-dependent’ modelling (in which the relationship between model and target is an ‘agential accomplishment’; what I have called the perspectival approach in CLS) offers a more solid ontological basis for modelling: a model relates to a target because it is intended for that purpose. Yet, it fails on epistemological grounds, in that intentionality does not establish the quality or accuracy of the model, and thus, cannot specify or explain the different capacities of different models to investigate a target. A substantialist account of modelling, one that adds mind-dependent criteria to a mind-independent account of modelling, or vice versa, simply poses the same conundrum at a step removed. Although ‘representationalism is a dead end’, as Sanches de Oliveira puts it, it is also unnecessary, because what makes modelling “work” in the sciences is not the nature of or the relationship between model and target, a dualistic and atonomistic formulation, but the practice of modelling in specific material and normative conditions:

Models are “about” target phenomena as much as they are “about” the discipline in which they are used, the theoretical context they are meant to fit and advance, the methodological and technological background they are built upon, the intended users, and the intended goals they are meant to help accomplish. Rather than thinking that models tell us something useful about the world, it is more accurate to think that scientists are the ones who tell us something about the world, something that they have learned by harnessing their skills in particular ways to build and manipulate objects of various sorts. Full reference in Zotero Library

Theories of performativity provide an established basis for developing these ideas for critical modelling as an alternative to representationalism. A central claim of these theories is the indivisibility of subjects, objects, practices, and situations of knowledge. To give a well-known example, gender is not an attribute adopted by pre-gendered subjects, but part of the condition of subjectivity that materialises bodies according to a binary model of sex-gender-sexuality Full reference in Zotero Library. Yet, it can be difficult to move from acknowledging this inseparability to enacting it within modelling as a critical practice. Sanches de Oliveira’s suggestion that models are about the disciplines or other knowledge systems they come from explains why models in literary studies, for example, are different—created for different reasons and from different materials—than models in chemistry. But it does not really offer any guidance for how to proceed in a discipline in which models are not an explicit part of the disciplinary tradition and computational models, in particular, are often understood to run counter to it. It does not help, in particular, in moving us to a situation in which practices of modelling can effectively function, in Barad’s words, as ‘productive evocations, provocations, and generative material articulations or reconfigurings of what is and what is possible’ Full reference in Zotero Library.

Some science studies scholars offer nonrepresentationalist accounts of modelling, including the constitutive role of infrastructure in the truth of climate models, and of bodies in the creation of models of proteins and computer graphics (e.g. Full reference in Zotero Library, Full reference in Zotero Library, Full reference in Zotero Library). As a small contribution to this effort, for literary studies, I offer measurement as a potential framework for understanding how situated practices can operate in ways that are legible to as well as constitutive of a community of scholars. To work in this way, however, measurement must be understood not in the positivist sense, of offering an objective description of the world, nor in the operationalist sense, that we cannot know a concept until we have a measure for it, but in the constitutive manner of knowledge, in Donna Haraway’s words, as ‘always an engaged material practice and never a disembodied set of ideas’ Full reference in Zotero Library, and conversely, I would add, of matter as always a way of knowing and doing (including by non-human agents) and never a blank slate or mute substance.

To explain this notion of measurement, let me turn to a paper by Fotis Jannidis Full reference in Zotero Library, and use the same analogy of measurement he employs, but to make a contrasting argument. Noting that computational literary studies lacks an accepted account of its research process, Jannidis argues that, in devising one, the complexity of measurement – and what measures are ‘good enough for […] literary studies’—should receive substantial attention Full reference in Zotero Library. Jannidis uses blood pressure to illustrate this complexity. He notes that it can be measured

[…] reliably by using either the auscultatory or the oscillometric method. Blood pressure is a rather complex phenomenon and the oscillometric method doesn’t represent all of its aspects, but uses a very specific aspect, the oscillations of a cuff pressure. Full reference in Zotero Library

Analogously, Jannidis argues that two models of the same literary phenomena can be very different—and even incommensurable—because they do not ‘represent the [literary] phenomena in their full complexity. They are just one single aspect’ Full reference in Zotero Library. I take a different lesson from this example, drawing on histories of measurement in science studies. In this field, scholars such as Emanuel Lugli Full reference in Zotero Library, Dylan Mulvin Full reference in Zotero Library, and Jaqueline Wernimont Full reference in Zotero Library argue that measurement does not find or define something that exists, but is part of the world’s production. Measuring technologies, including standards carved into the stone of medieval buildings, test images in digital processing, and disease and mortality tables, are not re-presenting something that exists somewhere else; they are fundamentally performative in the sense of creating that which they purport to describe (commerce, technology, public health). Likewise, these blood pressure measurements in Jannidis’s example are not two different ways of finding out about the same thing: a thing that already exists in bodies. They are rather different ways of enacting or creating bodies, and populations, that have blood pressure, and maintaining those entities through ongoing measurement.

As a constitutive knowledge practice, measures are not empirical and instrumental exercises as opposed to theoretical or semiotic propositions: they are formed of and manifest the inseparability of concepts and physical arrangements. Creating them means both bringing a community along and forming one as part of the specific arrangements involved in performing measurements. Understood as a collective practice of interfering with reality—rather than a way of standing back and pronouncing on it—measurement, as a practice of critical modelling, might offer a way of exploring and extending, rather than contrasting or dismantling, the humanities tradition. In Anita Traninger’s discussion of ‘Modelling “Literature”’, this concept of “literature” with which we work is presented as a performative effect of modelling: one that ‘create[s] the field that is supposedly denoted’. Could measurement, in these terms I have outlined it, offer something like what Lindsey Drury describes as ‘The Not-Model’ in her insight to the discussion on ‘Critical Modelling’? Understood, or rather done, like the anteater’s snout, not to represent the world (‘modelled upon the entrance to an anthill’) but to think with/in – with and inside – it (in the
anteater’s case, with the anthill’s tunnels), measuring could enable a mode of critical modelling that contributes to creating the literary phenomena our discipline studies.