In this section we will frame DH modeling activities by engaging with recent work on model building [Kralemann and Lattmann 2013]. Our focus is on what Kralemann and Lattmann call the “Act of the Interpreter” or “End” of the model or “Pragmatics”. We aim to highlight how that aspect of pragmatic modelling might need further attention in a DH context with respect to the two following elements: flexibility of language and fuzzy polarities (in particular object vs. model). By using the term pragmatic modelling we intend to emphasise an understanding of the act of modelling as anchored to theory and language. Kralemann and Lattmann explain “the semiotic structure of the model relation” as following:

[...] in the context of an intentional act (i.e. the pragmatics), a subject chooses a set of objects Oi=1,...,n (i.e. the extension of the model) and a theory or language (i.e. the intension of the model), which together determine the semantics of the model, and an object Omod the attributes of which act as the syntax of the model on the basis of a representational relation between themselves and the semantic model attributes: in the context of a theory and with respect to the respective end, Omod becomes a model of the objects Oi=1,...,n  [Kralemann and Lattmann 2013, fig. 5]

Their theory of models as icons in the Peircean sense acts as a generic framework to contextualise the creation and use of models in the sciences, humanities and, we could even say, life. To exemplify how this applies to DH (in line with established understandings of modelling), one could think of a researcher who is modelling a set of historical documents (her objects). In doing so she must engage with a theory of text and hence with an understanding of textual features relevant to her historical research. In her modelling activity, the researcher maps the theory of text and scholarly language she uses to describe and understand her documents to a digital model of her texts. She may use, for instance, a formalisation of the widely adopted Ordered Hierarchy of Content Objects (OHCO)[6] model combined with a Text Encoding Initiative (TEI) document model. Our researcher chooses a hierarchical model of text (the OHCO model) which she presumably thinks will allow her to map some of her understandings of the historical documents under study to a pre-defined structure (e.g. a TEI schema) and to process those data in some automatic form (e.g. to create indexes of certain selected elements, to perform combined searches or specific content analysis). However, in the modelling process itself she might question deeply the OHCO model as well as the TEI categorisations, reflecting on her own theory of the historical documents at hand and realising the constraints of both theory and model with respect to her objects of interest. She will most likely go back and forth in her modelling efforts to match – based on her knowledge and scholarly language – what she would like to elicit in the document (e.g. the organisation of the diplomatic formulas, the occurrence of names of witnesses in certain locations in the document, the abbreviations occurring in the date clauses) with the formal hierarchical structure proposed by one or more of the TEI guidelines chapters on the encoding of primary sources. Once this mapping reaches a certain stability and she is able to actually process that model, her interpretative activity will likely have to zoom in and out of these manifestations (e.g. between single cases and emerging patterns). She will often puzzle over exceptions that will allow her to iteratively refine the original mapping or require a total rethinking of categories and of the OCHO model being adopted. Single components or parts of the model itself (e.g. the specific schema adopted and its syntax, the technology, the versions of the model at x moments in time) cannot account for the complexity of “model relation” as described above by Kralemann and Lattmann. Within such a reductive analysis of modelling practices, the contextual and relational construction of the semantics of the model as well as its pragmatic end are lost.

A pragmatic understanding of modelling, as we understand it here, and as practiced in DH research, can facilitate the recognition that modelling operates within relational and dynamic cycles which are elicited via negotiations over the use of modelling languages (e.g. by renaming categories of analysis or adopting neologisms). The interplay (usually referred to as mapping) between the object of analysis (texts in this case) and the model, as well as across different levels of the interpretative process (e.g. close and distant reading, symbolic/syntagmatic and semantic/paradigmatic levels of text analysis), exemplify some of these dynamic aspects.

Texts are or at least have been so far the privileged objects of modelling activities in DH.[7] Here we chose to focus on some interrelated aspects of the study of texts, which seem to us particularly relevant to exemplify their complexity and openness with respect to modelling in DH: “dynamicity”, “multidimensionality”, historicity, and processuality.

First, we highlight “dynamicity”[8] as an aspect in the study of texts which emerges particularly in the relationship between models and objects as well as models and interpretations. DH researchers have engaged with linear texts all the way towards discontinuous narratives, for instance as structured in relational databases. In addition, texts as objects being modelled in DH research are not only or mainly verbal and encompass hybrid modalities such as maps [Eide 2013]. Indeed, dynamic aspects of texts as modelled in DH research and connected to the nature of cultural artefacts as such could be exemplified in a variety of ways. Researchers engaged in scholarly digital editing, for instance, have attempted at modelling the whole range of material incarnations of texts and the associated interpretative processes from manuscript witnesses to printed editions, from diplomatic transcriptions to critical editing. Hence, often within one and the same digital editing project, textual models do not sit still; they shift along multiple axes with respect to their objects and interpretations. The models themselves change shape. So, the axis of textual sources vs. models of analysis and interpretation of those same sources is mobile; for example, during the modelling process the sample of sources being selected might reveal new important features to model – such as a late pagination within a codex under study originally not included in the encoding model - which in some cases might also challenge the previous knowledge of those sources such as the biography of that codex. The axis between the textual models and the reading interfaces being generated out of them is also mobile; while developing an interface to a digital edition, the DH modeler might realise that some features of the original encoding model are not refined enough to allow for certain visualisations or specific manipulation of the data.[9]

Secondly, we mention “multidimensionality” as an aspect of the study of texts that emerges from the relation between objects and interpretations. With this respect, we refer to Sahle’s pluralistic model of text [Sahle 2006] [Sahle 2012] which states that the definition of text (e.g. text as semantic and intentional vs. text as version containing specific graphemes; text as complex visual sign vs. text as linguistic utterance; text as material document vs. text as work or rhetorical structure) depends on how we look at it, on the aspects we are most interested in making explicit in our modelling efforts and the tacit knowledge invested in those efforts. This is why the TEI standard aims to account for many models of texts and is a mobile guide in itself.

Next we turn to historicity, which is the aspect via which we put models in relation to their historical interpretations. The concept of historicity is useful to argue that, usually, in DH, modelling acts are also and necessarily meta-modelling activities because they explicitly engage with non-neutral objects, that is with objects mediated by prior interpretative activities even if not always passing via computational formalisations.[10] Therefore, our models of texts also embed historical understandings, categorisations and definitions of those texts.[11]

The last aspect we mention concerns the processuality of the study of texts located at the interface between objects and subject’s interpretations. This concept aims at highlighting a further connection between the reading (and hence also modelling) of texts and our being human. How we read, interpret, engage with texts is intimately situated in the performative and eventful process of creating and confirming our own identity [Meister 2007].

In conclusion, what this brief encounter with texts aimed to highlight is a reiteration of how entangled any DH modelling process is with the cultural, historical and personal engagement we have and had with these complex and open objects.

In philosophy of science there has been a lively debate around theorisations of models and their position in the middle between theory and data or objects (e.g. [Frigg and Hartmann 2012, 24–25]). Models as used in experimental settings, including computer simulations, are often conceptualised as “mediators between theory and material systems”  [Morrison 2009, 47, note 22]. Hence, also computer simulations, as a special type of model, can be perceived as links “between theory and experiment”  [Winsberg 2003]. In a simplified view of experimental settings theorised within such frameworks, a rather stable physical system exists; any new knowledge acquired during an experiment is mediated by models of that system (models of materiality) via a causal connection.

Without aiming at defining experimental practices in the techno-sciences, this paper dwells on how they are commonly perceived and how DH practice relates to them.[12] These perceptions and relations explain the borrowing from the techno-sciences we aim to unpack and also suggest directions towards a model of modelling integral to DH research. In the case of computational models, for example, a physical system is represented by a simulation model, which is the result of a “discretizing method”  [Morrison 2009, 45] passing via various mathematical models of that system.[13] Now even if this is a simplified view which does not take into account that a pure empiricism where objects and theories are entirely separate entities has been questioned by scientists, philosophers[14] and of course by the framework of quantum physics,[15] what is interesting for our analysis is that computer simulations in a techno-scientific context seem to be perceived as adding a level of “immateriality” (or better “fictionality”) to experimental settings, while in DH they tend to be perceived as adding an element of materiality to humanistic interpretation.[16] The latter resonates in particular with respect to the popular claim of DH as evidence-based humanities - or what we could call a Cartesian approach to the humanities - as well as to the criticism towards DH as a new positivistic approach to scholarship [Eyers 2013].

History of culture and science can be read as progressively distancing and detaching human life from the world of nature via the meditation of tools and technologies.[17] This is a process that can be summarised as a dialectic tension between two polarities, that of physis and of techne. Philosophers tried to explain and mediate this process, but the dichotomy between objects and thoughts, models and objects, theories and practices, explanans and explanandum, observer and observetur has become pervasive and intrinsic to the development of Western thought. So we ask: does the fuzziness of the two polarities of objects and models emerge in a new vest in a DH context?[18] Above we saw this with respect to objects (texts); here we also see it contextualised with respect to the language we use, whether intended as modelling language, language of representation or language as vehicle of theories and methods. That modelling practices and language are tightly connected is not new. This is also the case for modelling in DH (our italics):

You may find you have to name things; you may have to reify them, not in the social science sense of making them … beneath attention but in the AI [Artificial Intelligence] sense of making them into things that you can address and talk about and think about. That’s one of the reasons for modelling in general. One reason why do [sic] we care about making our assumptions and fundamental beliefs explicit? So that we can look at them and say, “Well, actually I don’t like that one! Could we do without that one? Can I build a system that doesn’t rely on that assumption?” If you don’t ever surface your assumptions, you’re never going to be able to think about building that system, let alone build it. And you might want to build it, because our assumptions turn out to have teeth.  [Sperberg-McQueen 2012]

What a pragmatic understanding of modelling enables us to acknowledge though is that the opportunities enabled by modelling (e.g. emergence of patterns of relation, behaviour, and shape) are rooted not only in a “demonstrative” and “literal language” but also in a metaphorical one [Marras 2013] [McCarty 2006]. In the context of modelling, language operates at different levels; for our purposes we could identify at least two: (1) functional and (2) conceptual. In the former case, a language is used to describe a model in descriptive yet operational terms; an example is, for instance, how effectively we apply genealogical terminology (e.g. parent, ancestors, siblings, children) to navigate the hierarchical structure of an XML document or to describe any arboreal organisation of knowledge. On the other hand, in the case of metaphorical language, the terminology being chosen and the relevant metaphorical models being applied organise the modelling processes themselves. Below we provide some examples of what this means in practice.

DH scholarship has been reflecting on the limits of adopting uncritically one language of computer sciences. Starting from semantic data modelling and the widely adopted yet controversial term of “ontology”, Pasin and Ciula, stated that DH approaches seem to privilege a pragmatic rather than a realist understanding of the term in light of the fact that “ontologies for the humanities must support diversity and variety of viewpoints; thus they cannot adhere to an underlying model which neglects multiplicity in favor of a monolithic vision of the world.” The authors promoted the pragmatic

concept of an ontology as the agreement reached by multiple parties (e.g., programmers, scientists, collaborators, librarians) with the aim of accomplishing some objectives (e.g., data exchange between applications, communication between people, integration of disparate representations). Using a metaphor, ontologies are contracts, they are the currency used to perform some valuable operations. Thus, their importance is ultimately related not to their truth or beauty, but to the ease they bring to the collaboration among people. To use a less “commercial” metaphor an ontology is a compromise or a point of contact between specific and possibly divergent models. The issue is therefore not only to identify commonalities between projects, for instance, but also to agree that the compromises so found wonʼt diminish the value of the underlying idiosyncratic models, the specificity of any single project or interpretation. We believe that in the humanities this agreement is not necessarily reachable once for all or hoped for, because it may imply the negation of the interpretative efforts that make a work or a project unique and the negation of the evolutionary nature of scholarship. However, we also think that the possibility to make two incommensurable categorical systems communicate could be a challenge worth pursuing.  [Pasin and Ciula 2009]

Some modelling attempts in DH have embarked on dedicated efforts to problematise terminology. For example [Crofts et al. 2011] developed refined formalisations of concepts for cultural heritage information systems; [Eide et al. 2013] tackle the complexity of spatio-temporal concepts in humanities and arts; [Brown and Simpson 2013] discuss how concepts of difference and “personhood” are flattened and smoothened out while cherry-picking extant semantic web ontological models for research in the humanities; [Renear et al. 2010] call for a rigorous analysis of what “datasets” in multidisciplinary contexts are for libraries, publishing, and data curation.

Other DH scholarship has also ventured in creative attempts at establishing neologisms. Short and Bradley [Bradley and Short 2005], for example, discuss the choice of name for the prosopographical model they developed and used in various DH projects – “factoid” - as possibly ironic but mirroring the “historian’s worry” [Bradley and Short 2005, 8] that what they recorded as assertions in the historical sources under study are not the same as facts.[19] The name is therefore the face of a very much context-aware approach to history, but, as explained by Pasin and Bradley, it becomes also a metaphor structuring the corresponding digital resources in very practical terms:

During the “data acquisition” phase, factoids provided historians with both a guiding metaphor (helping them conceptualize the broader approach being used) and a usable structure for the data entry work. Secondly, within the “data storage” context, factoids proved to be a practical, flexible, and sustainable schema for designing databases. Thirdly, during the “data presentation” phase, the factoid notion has been used with success to the purpose of building user interfaces that are simple yet rich in the way they combine and organize information about people and make it available to the historians using our online resources.  [Pasin and Bradley 2013, 4]

These cases cast light on the trade-offs of projecting historical lexicons in new contexts. They are also evidence that, in DH modelling practices, language is not purely observational, but correctable and ambiguous. A model is not an addition to a theory; the non-univocal character of scholarly language has to be recognised. In this perspective, the relationship across borders of disciplines is also to be considered as a translation, which establishes a sort of metaphorical relationship. As the result of an “intercultural marriage” between digital and humanities, DH is a realm of grafting across disciplines. Its scholarship requires adaptable models that are able to grasp such specificity of language as well as models that emerge from the specificity of its theories and objects of analysis, models which emerge out of a modelling process not directly linked to its observational contexts. Without diminishing the value of formal and functional languages at the interface between human and machine communication, what is therefore of interest to us is the strong intrinsically metaphorical component of the DH modeler’s use of language (discussed below as modeler’s talk).

In the continuous effort[20] DH researchers undertake in embracing and fighting a pragmatic model of model building, a constant resistant dialectic between conceptual expansion and formalisation takes place. When examined from a diachronic and comparative perspective, it is also interesting to note that this effort finds parallel approaches in the history of use and functions of metaphors around the organisation of knowledge.[21]

The pragmatic take on modelling, as outlined above, implies some equilibrium between the two polarities (observation/objects/literal language, on one hand, and theory/model/metaphorical language, on the other) and overcomes the misleading dichotomy between observation of the objects and theorisation (hence modelling) of those objects. A nomologic-deductive process in building models which distinguishes sharply between theory and observation has been surpassed in DH theory. With respect to this but mainly focusing on visual languages in DH, Drucker argues that:

all data have to be understood as capta and the conventions created to express observer-independent models of knowledge need to be radically reworked to express humanistic interpretation.  [Drucker 2011]

Linguistic tools and metaphors[22] enable us to move forward the theoretical reflection as framed so far. In this respect, metaphorical reasoning exemplifies a specific strategy that can guide modelling in DH to move away from the dichotomy between bottom-up (models emerging from particulars) and top-down (models imposed on particulars) approaches. We argue that pragmatic modelling brings into focus a middle-out method which acts at the crossing point between data and models adapting itself to specific “textual contexts”. Three interrelated properties belonging to pragmatics (adapted from [Verschueren 2012]) can be associated with this method and explain how metaphorical reasoning functions:

• Variability - the range of choices in the use of language cannot be seen as static in any respect;
• Negotiability - such choices are not made mechanically or according to strict rules or fixed form-function relationships, but on the basis of highly flexible principles and strategies, thus also implying the indeterminacy and unexclusiveness of the choices being made;
• Adaptability - such negotiable choices can be adapted based on specific needs and contexts according to a variable range of possibilities.

These properties characterise pragmatics not merely in terms of its interpretative function in reading a specific communicative domain, but as a cognitive, social and cultural study of language (and communication) as anticipated at the beginning of this paper. The awareness of the value of pragmatics in modelling acts contributes to making rigorous practices open to a creative and imaginative dimension. In this context, metaphors often function as models to integrate the interpretation of theories, especially when there are not terms or concepts to be used that are directly related to the observed facts/objects, in other words when there is an indirect or remote relation between observer and observatur, explanans and explanandum.[23] In these cases, metaphors compensate or fulfill a “linguistic gap”, the “inadequacy” of the ordinary language for scholarly purposes.

Metaphors have been largely used by scientists in building experimental and theoretical sciences, in representing and describing a domain, and in promoting understanding in scientific investigation. Scientific metaphors are pervasive in science from antiquity to our time, from Aristotle’s De Partibus Animalium, to the early modern scientific attitude to modelling as a partnership with craftsmanship and engineering; from the notion of the world machine, of God as artisan or architect and the world as his handicraft, to Darwin’s “tree of life”, from the mind-computer to the protein-protein interaction as network in more recent science. Metaphors often act through different “pragmatic” mechanisms directly related to the properties listed above: from transposition - moving concepts from a context to another (variability), to corrections and interpretations due to new contexts (negotiability), to its reformulation (adaptability to new contexts). Metaphors describe novelties, help in understanding and interpreting theories but should not be reduced to be simply the language of a model [Hoffman 1985]. In this sense, the use of metaphors is a strategy to guarantee a non-rigid, descriptive approach to modelling. This allows the preservation of the “openness” of a middle-out pragmatic approach to modelling as a DH context of study requires.

In DH research, the use of structuring and modelling metaphors is more and more evident. The metaphorical lexicon certainly pervades the DH discourse, where terminology and metaphors are intertwined (e.g. surfing, navigating, artificial intelligence, computer machinery, etc.). Moreover, metaphors are not only linguistic entities; an interesting range of metaphors in DH is used for structuring concepts at a macro-level, for example, in the use of “ecosystem”  [Marras 2015] to stand for research environment [TextGrid] or the cartography/landscape metaphor where the “aim of the research is to extend the cartographic metaphor beyond visual analogy, and to expose it as a narrative model and tool to intervene in complex, heterogeneous, dynamic realities, just like those of human geography”  [Knowledge Cartography]. Metaphors are also used to organize content and visualise knowledge and data [Hestia] or to conceptualise the main features and aims of a research project, for instance in the use of “discovery” in the Steve project [Steve], where the social tagging is conceived to lead to new ways to describe and access cultural heritage collections. A metaphor is the “agreement” in the framework of scholarship online promoted by [NINES]; metaphors design the process-orientation and cross border approach of the review platform for European history (Recensio.net). Furthermore, metaphors are used to structure the knowledge embedded in some research projects, highlighting for example their innovative aspects, as in the Discovery project (“You can travel through the territory at will but can also design new maps and author guidebooks for others to read and respond to”  [Discovery Project]) or making emerge the interrelated work space as formal arena as in [AGORA].

In a pragmatic understanding of model building theory and object are complementary; the model takes a middle position operating within a “middle-out approach”. The model preserves its representational characteristics vis à vis the object – for instance its intentionality and extensionality as explained in section 1.1 - but the reality such a model organises at a conceptual level is not based on a purely descriptive language nor in a projection or interpretation of properties.[24] The model has its syntax, grammar, and semantics but within a processual consideration of the use of language (a non-static perspective according to the pragmatic properties listed above).

Bringing to the fore the metaphorical language element allows DH modellers to also give prominence to the openness and complexity of their objects of analysis. So within a pragmatic understanding of modeling any model is primarily pragmatic because it privileges the specificity of the objects and the contexts of use (of the humanities), but also because its pivot lies in the manipulability of the models themselves, in terms of their negotiability and flexibility.