In the digital humanities, the creation of a data model usually represents a pivotal stage in the research process. Data modeling not only serves to describe the domain in question and to guarantee interoperability, but it also helps to structure the available information and to gain new insights and perspectives into the subject at hand. At the same time, the concomitant processes of classification may have a considerable impact on subsequent studies. In the present paper, this tension is explored through the analysis of two exemplary use cases, each of which employs a diefrent data model, whose impact on the research process is in turn examined: the modeling of the entity of a 'burial,' and the modeling of the spatial relations between objects in a grave. The conclusion will examine questions arising from this inquiry with a special focus on scholars working with research-centered databases.
As the digital humanities continue to grow and develop, it has become increasingly clear that the research process starts long before the actual analysis whenever digital methods or digital data are involved. Any engagement with the data at hand must be preceded by a thorough consideration of the characteristics of the phenomenon to be studied, and the structuring of its information domain – or in other words, by the creation of a data model. This holds true not only for research groups working on large-scale projects such as digital editions or corpora, but also for individual scholars focusing on specific data-driven questions. In terms of data structure and data models, these two scenarios are characterized by very distinct requirements and limitations: while the objectives and architecture of larger projects are often conducive to the use of standardized formats, schemata, and ontologies, the application of the same principles to studies conducted by individual researchers or small research groups can prove to be quite a challenge.
For research projects rooted in the humanities, data modeling can serve
a variety of purposes which often – but not necessarily – result in the
functional implementation of the respective models
In addition, modeling one’s data also creates a basis for human
communication in that it describes the domain in question, structures the available
information, and oefrs new insights and perspectives regarding the subject
at hand. Ciula and Eide go as far as to prioritize this function of modeling as
“a creative process of thinking and reasoning where meaning is made and
negotiated” over its more prosaic uses in database implementation
However, the processes of classification that accompany such eofrts can also shape subsequent studies to a considerable extent. For example, choices of ontologies and decisions on how to apply them can open up new paths of inquiry, but, at the same time, can also close of others well before the collection, analysis, and interpretation of data has even begun. The assignment of classes and properties, especially in the early stages of research, when some implications or theoretical perspectives might not yet have been thoroughly explored, can canonize certain points of view with regards to the data, when a more flexible understanding would have been beneficial (cf.
Star, 2008).
Bowker and
Obviously, such challenges are nothing new in humanities research
The humanities already possess the tools required to meet this challenge –
most notably in the form of the critical reflection of research practices that is
part and parcel of postmodern and post-processual theory, and that can be
readily transferred into the domain of the digital humanities. Indeed, there
have been many instances of importation and adaptation of critical
perspectives over the last decades. For example, Michael Shanks proposes a
“symmetrical archaeology”
As far as the practice of data modeling is concerned, these examples remind us that researchers must be aware – in creating as well as in working with data models – that there cannot be one true data model representing the entirety of a certain real-world phenomenon. Instead, modeling practices reflect conventions, traditions, and individual habits of researchers, as well as the particular goal of the respective modeling project.
These considerations seem especially relevant as the way that knowledge is
represented and structured has a profound and immediate relevance for
society. As Hui states, digital objects can be understood as “externalized
memories that condition our retrieval of the past and our anticipation of the future”
The challenge of the subjectivity of researchers and research perspectives
applies especially to those types of databases that Flanders and Jannidis call
“research-driven”
Consequently, interoperability might be of secondary concern. Instead, diefrent ways of modeling the same ‘real-world’ phenomenon reflect different research questions, or theoretical frameworks employed to approach these questions. In such cases, data models have a profound eefct on research processes as a formal, structured way to think about the phenomena studied and the information from which these data models are built. Furthermore, data modeling shapes the scholarly process by focusing on specific aspects which are analyzed in greater detail, while others potentially remain unexplored.
This tendency can also be observed in the context of graph databases, particularly in combination with network research: on the one hand, modeling data in a graph promotes ‘thinking in relations’ even at the very beginning of the scholarly process and long before these relations are realized in the database itself, because the task of transforming unstructured data into nodes and edges can shift a researcher’s focus towards connections that were hitherto unnoticed or disregarded; on the other hand, once defined, these modeling decisions can complicate certain avenues of analysis, or prevent them from being pursued altogether.
In this paper, two use cases are presented to demonstrate how the scholarly
process starts with and impacts data modeling, and how diefrently
constructed models facilitate diefrent perspectives on the same subject. While both
originate in a study about elite networks of the Late Bronze Age
The case study presented in this paper deals with elite burials of the late
Urnfield period (c. ninth century BC), which constitutes the last period
of the Central European Bronze Age in an area that reaches from Eastern
France to the entrance of the Carpathian Basin and from the Alps to the
German Mittelgebirge. Grave goods and other characteristics, such as grave
architecture, are analyzed in the form of a network consisting of burials and
the grave goods and features that they share (Figure 1). As the construction
of burials and the selection of grave goods were most likely intentional
acts that served to represent intersecting identities, statuses, and sources
of power of the deceased or their successors, the patterns evident in the
network reveal which strategies the elite(s) pursued to gain prominence at
the beginning of the transition from Bronze to Iron Age, a time of marked
socio-political unrest.
The burial of a member of the elite also constituted a focal point for
the construction and negotiation of communal as well as individual
identities in a manner analogous to Giddens’ theory of the duality of
structure, which states that “the continuous recreation and re-articulation
of identities through burial rite [...] feeds back into the structuring of
societies”
For the purposes of this study, 82 previously published burials were
selected on the basis of the presence of objects that most likely represented
prestige goods in the late Urnfield culture, as well as additional signifiers of
high status that can be assumed to have been characteristic of the funeral
presentation of elites, such as the presence of an abundance of ceramic
vessels and elaborate grave architecture
The data was stored in a graph database1 and exported into network
analysis software2 in various configurations in order to analyze it from diefrent
1https://neo4j.com/release-notes/neo4j-3-4-10
2http://visone.info
network perspectives. The underlying data model as well as the ones
presented in this paper are mainly based on the CIDOC CRM
As discussed above, the question of how to model the burials themselves was of central importance to the study – the goal being not only to be able to store, analyze, and retrieve data quickly and eofrtlessly in and from the graph database, but also to facilitate the export of specific one- and twomode networks that figured prominently in the course of the network research. In this regard, the structure of the model directly aefcted which inferences could be drawn from the subsequent analysis. To illustrate this point, two models are presented below. The first one is based on the research question outlined above, namely: how is elite identity constructed through elements of the burial?
As can be seen in Figure 2, in this case, burial as a general term is
characterized as E19 Physical Object, which, according to the specification, is
comprised of “aggregates of objects made for functional purposes”
While this model serves its purpose and is supported by the specification
of the CRM, it does present some problems, especially when it comes to
the assignment of the class E19 Physical Object to the burial itself. By
designating the individual physical structures of the burial classified as E25
ManMade Feature as distinct elements, the burial node no longer embodies an
actual physical object, but rather an abstract container that is defined by the
association of the diefrent types of entities described above. At the same
time, it can be argued that a burial in general should be seen as a conceptual
entity comprised not only of material remains, but also of (mainly ritual)
actions – a point to which I will return in my discussion of the second model.
The CIDOC CRM oefrs the class E28 Conceptual Object for “non-material
products of our minds and other human produced data that have become
objects of a discourse about their identity, circumstances of creation or
historical implication”
By way of contrast, the second research question that I wish to explore in this paper focuses less on the material background and more on the concept of social practice as an integral building block of societies: which social processes constitute a burial?
Analogous to the first example, the data model constructed in response
to this research question is based on the CIDOC CRM. As the CRM is
explicitly described as an “event-centric model”
Even at a cursory glance, this model appears to be much more complex
than the first (Figure 3). Adding a meta level in the form of actual persons
and their activities to the objects that are typically the focus of
archaeological research expands each relation by two or three additional steps. Here,
too, the burial is understood as an E7 Activity; E20 Biological Objects, E22
Man-Made Objects, and E25 Man-Made Features are now connected to it
through their use in the burial rite. Added to this model are the E39 Actor,
representing the successor of the deceased, and the E74 Group, which – as
the burial community – provides the social framework for all actions taking
place in the context of the burial. As such, the model directs the researcher’s
attention not to the objects themselves, but rather to the events of
production and modification that made them a part of the burial, their purpose,
and the relationships encoded within them. While it is certainly possible to
derive status-related inferences from the interplay of these actions, the focus
shifts away from the intra- and supra-regional significance of the social
persona of the deceased
Another aspect that emphasizes function and social practice over typological
and chronological attributes is the arrangement of artifacts in the grave. As
Martina Löw states, space represents a social construct that materializes and
impacts perceptions and interactions of communities (
Within this context, diefrent conceptions and encodings of space answer diefrent research questions. A geospatial coordinate system, for example, seems most useful for the exact documentation of the location of an object, as in the case of an excavation. This attributive absolute understanding of space as topology can be easily expressed through the CRM (Figure 4): graves, cemeteries, or other types of locations of finds – in this case expressed through a generic E27 Site node – can be assigned an E53 Place through the property P53 has former or current location. The E53 Place class can then be enriched by adding an E44 Place Appellation, coordinates in the form of an E94 Space Primitive, and – if necessary – an E55 Type. Hierarchies of these P 4 6 is c o m p o s e d o f f o t r a p s rm o f 6 4 P
E19 Physical Object f o t r a p s m r o f 6 4 P P 2 h a s t y p e E55 Type
E25 Man-Made
Feature
P 4 6 f o r m s p a r t o f E21 Person places can be constructed through the property P89 falls within. In theory, the same approach could be taken for individual elements of a burial, such as grave goods or features.
E55 Type
E27
Site
E7 Activity architectural features, and human remains, based on the CIDOC CRM socE Relationship
E55 Type E53 Place
E20 Biological Object socP to o t P c o s
socE
Relationship architectural features, and human remains that focuses on underlying social processes, based on the CIDOC CRM and CRMsoc
E69 Death E21 Person
P 1 0 0 w a s d e a t h o f s d n i b P c o s socE Ownership
E94 Space Primitive
E44
Place
Appellation ifnds from Grave 119 in Franzhausen-Kokoron, based on the CIDOC CRM and extended through additional E42 identiefirs corresponding to Figure 5
Apart from documentary purposes, the assignments of geometric information also serves as a general frame of reference that can be translated into other, less normative conceptions of space. For example, the location of objects in relation to the architectural features of a burial pit or chamber can answer questions regarding the standardization of burial activities, including deposition patterns, the construction of spaces within the grave as zones that give material form to diefrent identity aspects of the deceased, and their relationship to each other, or even the role of space as an actor in the perception and negotiation of the burial process.
In this paper, however, the focus lies on yet another aspect: the placement of objects in relation to each other. On a very basic level, the analysis of these kinds of spatial patterns allows inferences about the function of objects that might otherwise remain obscure, and, in a broader sense, about intraand supra-regional meanings of individual artifacts or groups of artifacts in connection or opposition to each other. Grave 119 of the cemetery of Franzhausen (Nußdorf ob der Traisen, Lower Austria) provides a comparatively simple example, particularly when it comes to the bronze and iron knives it contained.
As can be seen in Figure 5, the assorted ceramic and metal grave goods
found in Grave 119 are systematically distributed over the area of the
rectangular burial pit. The metal objects (no. 9, 10, 12, 13, 15) are divided up
and placed into several of the ceramic vessels. Of particular interest here are
one bronze (no. 10) and two iron knives (no. 13, 15), whose contexts are
very diefrent: while the bronze knife lies across the remains of a vessel that
unfortunately cannot be reconstructed and is accompanied by animal bones
(most likely a meat oefring or remains of a funeral feast), the iron knives
were found inside of an urn, alongside the ashes of a cremated body
While these findings might seem trivial at first glance, the diefrent
treatment of the same type of object depending on its material ties into a wider
cultural context: the transition from Bronze to Iron Age, and the
increasingly widespread adoption of iron as a production material. Especially with
regard to the way in which objective sources of power
It can be assumed that many more insights into the function and
meaning of objects, as well as the social construction of funeral space, could be
inferred from the respective arrangements of grave goods. In particular,
incorporating spatial relations into the data model of a corresponding knowledge
graph that extends one of the models shown above could allow automatic or
semi-automatic queries that would in turn point researchers to other
potentially fruitful constellations. At the moment, such socio-theoretical
understandings of space have not been integrated into the CIDOC CRM. While
compatible models such as CRMarchaeo
In the absence of suitable standards, Figure 6 represents an initial attempt to construct a simple model of the use case outlined above. It incorporates only the section of the inventory connected to the three knives and their possible functions, focusing on two types of relationships or properties: those that describe the placement of an object in relation to another, and those that provide an interpretation of this connection. Regarding the latter, type and certainty are expressed through attributes. As the model is merely a tentative ifrst step towards a possible structure for such data, some caveats apply: for example, the plainly labelled relations of ‘next to’ and ‘above’ are in need of a more formal treatment, preferably in the form of a controlled vocabulary or even a hierarchical thesaurus. Likewise, the question of which connections should be incorporated into the model requires further clarification: should the spatial relations between all objects be codified? Or does it sufifce to provide an explication of the ones deemed to be ‘important’?
In contrast to the CRM, space or spaces as such do not appear as nodes of
their own in this concept, at least not at this point in the modeling process.
Drawing on the two processes involved in the creation of social space
proposed by Löw, spacing and synthesis (
These two short examples show how the practice of data modeling constitutes an explicit part of the research process that influences the conclusions that can be drawn from the material at hand. At the same time, the theoretical frameworks or premises that are present in the primary design of a given case study exert a strong influence on the resulting model. For this reason, the goal of data modeling in the humanities, especially in the context of research-driven projects like the ones discussed above, cannot be to ‘streamline’ data structures. Instead, what is desirable is for the diversity that is to be found in study design to be mirrored in the respective modeling approaches. It is incumbent upon humanities researchers to address crucial questions such as the following: how can the task of modeling be further integrated into the scholarly process? How can one eficiently and clearly disclose the choices that lead to particular models? And which possible implications and biases could arise from these data structures when analyzing specific data-driven research questions?
Beyond these lines of inquiry, which mainly concern communication
between humans, there also remains a question about the communication
between machines, or in other words, interoperability. As Arp et al. note,
[o]ne central goal of the annotation of data using ontologies is to enable
what is called ‘semantic interoperability’ between heterogeneous
computer systems, defined as the ability of two or more such systems to
exchange information in such a way that the meaning of the information
generated by any one system can be automatically interpreted by each
receiving system accurately enough to produce results useful to its end
users.
Today’s digital humanities emphasize the linking of information that enables such exchanges, as the steady growth of fields such as knowledge graphs, the Semantic Web, and Linked Open Data shows. Yet, especially in the case of research-driven databases, the question of how to navigate expectations between the poles of universal interoperability and case-specific models, and how to integrate such models in the wider network of the knowledge domain, can be dificult to answer. Nevertheless, such considerations are of fundamental importance, particularly in cases where researchers are publishing previously unknown material as part of their projects (e.g. recently excavated archaeological finds or newly edited historical sources), or where they aim to integrate data from other publicly accessible databases.
On a very fundamental level, a partial solution to these challenges lies in the thorough publication and documentation of datasets. In the case of research-driven projects (e.g. in the form of quantitative analysis), databases should customarily be published along with the findings. For these as well as for curation-driven databases, data models should be readily accessible and thoroughly described. In the accompanying documentation, the underlying logic and purpose should be made explicit – in general, researchers should aim to develop a critical approach to practices of modeling.
Concerning the interoperability of databases, it seems an easy solution to call for more generic data models that can be easily mapped onto each other, or to dismiss the contribution of research-driven data collections to the overall record in general – especially since, at present, the online publication of these collections can be dificult to manage for individual researchers or small teams without access to the necessary infrastructure.
Indeed, the task of mapping two data models or schemata onto each other
– in the sense of “a suficient specification to [allow the] transformation of
each instance of schema 1 into an instance of schema 2 with the same
meaning”3 – involves numerous challenges. On a structural as well as a semantic
level, data models can diefr significantly from one another. Some factors
have already been discussed, yet more can arise, concerning data types,
naming conventions, the level of detail or completeness, inconsistencies, as well
as fundamental discrepancies in the design of the underlying ontologies (cf.
For research-driven projects, these approaches might not be immediately
usable. Yet it is precisely in these cases that the exact goal of the mapping is
particularly relevant. As pointed out above, the objective must be to produce
a system that is “useful to its end users”
In the first case, the data model might matter less than the careful enrichment of individual entries with standardized metadata, such as identifiers from the Getty Thesaurus of Geographic Names6 or the integrated author3http://www.cidoc-crm.org/short-intro-mappings 4http://www.cidoc-crm.org/mapping-methods-technology, http://www.cidoc-crm.org/mapping-tools, http://www.cidoc-crm.org/reports_mappings 5https://kalliope-verbund.info/en/index.html 6http://www.getty.edu/research/tools/vocabularies/tgn/index.html) ity file (GND) of the German National Library. 7 Even if certain standard schemata are required, the identification of basic entities such as persons and places is likely to pose only a moderate challenge, particularly if the respective target database is well documented.8
In the second case, researchers must first identify to what degree the two
databases in question overlap in terms of their theoretical and structural
scope. A complete mapping of concepts and relations might not be a
sensible approach if, for example, certain terms have diefrent meanings or ranges.
Particularly in the case of graph databases, the question of which entities are
presented as nodes and which as attributes, and whether events or
circumstances are modeled as nodes or edges, can also distinguish individual models.
If approaches vary considerably, the exploitation of ontological hierarchies
or the use of upper ontologies9 can build bridges to connect diverse datasets.
As the top level classes of the CIDOC CRM form one such upper
ontology
In conclusion, data modeling presents a necessary and fundamental part of the scholarly process, and can provide insights into research topics beyond the immediate goal of database implementation. Data modeling also confronts researchers – creators as well as users of research data – with its own set of challenges: this paper has shown how data models and the (implicit) premises and theories built into them a priori can shape the study of phenomena in the humanities; how they influence researchers in terms of the analytical approaches, interpretations, and conclusions that are available to them on the basis of specific modeling choices; and how they can obstruct or block entire paths of inquiry, whether it be on account of technical barriers or one-directional thought processes. To address these issues in a critical and deliberate manner, the publishing process must include clear and expli7https://www.dnb.de/EN/Professionell/Standardisierung/GND/gnd_node.html 8For example, the web service correspSearch of the Berlin-Brandenburg Academy of Sciences and Humanities (https://correspsearch.net/index.xql?l=en, that collects metadata from scholarly editions of letters gives detailed information on which formats and authority files need to be provided to add information to the service (https://correspsearch.net/index.xql? id=participate_steps&l=en, and which can then be exported even from highly specialized databases with a reasonable amount of eofrt.
9Many thanks to Cristina Vertan (University of Hamburg) for her helpful suggestion regarding upper ontologies in the discussion that followed the presentation in Vienna on which this article is based. cit documentation which outlines the reasons for selecting specific models, discusses the implications of these choices for subsequent research activities, and thoroughly describes the data models, schemata, and authority files used. If, however, mapping approaches carefully consider the scope and goal of the datasets in question, they can facilitate interoperability even in the case of research-driven databases, and enable both individual researchers and largescale projects to share their data, analytical approaches, and research results with a much broader audience.