Challenge-derived design practices for a semantic gazetteer for medieval and early modern places

3.1. Toponymy

Modelling the designation of places over time poses a challenge in itself. A town in the contemporary Federal Republic of Germany is called Münster by their inhabitants, but bears the official name Münster (Westf.). There are sources from the ninth century which refer to this place as Mimigernaford [38, p. 283]. Another source dated to the year 1244 calls it Monasterium [38, p. 284]. This simple example shows two problems one encounters when modelling historical places. Firstly, place names change over time. Secondly, even at one point in time multiple designations can refer to the same place, which is already the case when taking into account multiple languages.

To address these problems, a gazetteer must distinguish between the place as such and its designations.4

In order to describe places based on historical and contemporary human-made attestations, categories have to be developed. Especially when dealing with historical places, the development of these categories already implies an interpretation. This interpretation is either based on historical research or, when using concepts directly derived from historical sources, based on a specific worldview, held by the creators of these sources. When focusing on the latter, a gazetteer cannot be merely understood as a collection of places, but as a “cultural gazetteer” [58, p. 141]. In addition, there are two major aspects to consider when creating concepts for categorizing places.

Firstly, one can distinguish between fiat objects and bona fide objects (physical objects). Fiat objects are virtual spaces conceived by men [58, p. 135]. Mostly they are grounded in some legal or administrative concept, for example a diocese, a republic, or a kingdom. Physical objects are places that can be observed in the actual world, like buildings and towns (here understood as physical entities, not as political concepts) but also natural objects like rivers or trees. Especially in medieval and early modern times, it is easier to attribute exact geodata to these places than to fiat objects.

Although fiat objects and physical objects can be treated differently, it must nevertheless be possible to create dependencies and relations between them. A physical place like a church building can be located inside an administrative unit, which has its own properties. This means, that at least some of the properties concerning the administrative unit must be valid for the church as well. Therefore, the actual place and the meaning of the place attested by humans are distinguished by two concepts. Another example for this could be Vehmic oaks, which served as a court in medieval times. Modelling the tree object and the attested meaning as a place of jurisdiction as two concepts can also make it easier to represent changes to both aspects separately. The function of any specific Vehmic oak ended at some point in time, the tree itself on the other hand can exist much longer and even get another cultural meaning, attributed later in time.

Secondly, it is possible to use either specific/contextualized or general terms for place concepts. Place concepts like duchy, republic, prince-bishopric, or parish are specific terms, and their meaning is related to a certain historical context, while general concepts can be understood as concepts of broader categories, encompassing a multitude of place concepts. Examples for general concepts could be secular dominion, theocratic dominion and ecclesiastical administrative unit. Regarding the use of terms for concepts, a semantic gazetteer for historical places should not solely rely on one of these approaches. The assumption that a duchy and a republic can be grasped by only one general concept, like secular dominion, is a very broad historical simplification. Therefore a gazetteer always has to maintain relations between the place concepts it uses, depicting heredital dependencies between its domain oriented terms.

The given examples for specific fiat concepts illustrate an additional challenge related to medieval and early modern places in particular. Not all historic ways of ruling can be grasped the same way as contemporary administrative units. Going further back in time, a clearly defined authority ruling a certain territory is less likely to be found. Instead, administration existed in the form of various privileges, e.g. different forms of jurisdiction or the right to raise taxes. These, in a given territory, could be enforced by different agents – e.g. a prince, the church, a town, or a local nobleman. Sometimes the same or similar privileges could be claimed by multiple ruling actors for the same group of people. Therefore an accurate depiction of an administered space with a multitude of privileges based only on fiat concepts would require the creation of a large and heterogeneous number of such concepts. A more accurate way would be to model ruling as relations between agents, privileges and places. This way, the vast ambiguity of medieval administration, as well as small conflicts of interest, could be captured more effectively. An example for the connection of data about places and people to model rulership is provided by the Hull Domesday Project. The structure for the underlying database is mapped from the historical source text it represents, the Domesday book [30,40].

Place concepts are similar to the feature type attribute defined by Linda Hill for a basic gazetteer [29, p. 107]. However we would like to go further than structuring places based on a thesaurus of feature types developed from a certain point of view on (historical) places. When structuring historical places not only the historical context should be included, but also historical point of views about the places. That way a gazetteer about historical places can be a database trying to depict the world as it was presumably perceived by the people of the past (by using the terms and concepts from historical sources) while it is also a tool to structure our way of thinking about history in terms of research traditions. A simple feature type thesaurus is not sufficient for such a task [42, p. 45].

To better grasp these challenges about multiple assertions how a place is regarded, we can again take a look at the history of the city of Münster. During the Münster rebellion from 1534 to 1535 the city was under the rule of an anabaptist group. They regarded the city as the center of their new kingdom and therefore detached from the prince-episcopat Münster it was part of. In the eyes of the bishop, the city was still part of his territory. At the same time historians could want to classify the city for the years 1534–1535 as neither, but as theocratic dominion. A gazetteer for historical places should be able to include these multiple assertions.

3.3. Hierarchies

An extensive gazetteer should be able to structure not only the places as singular entities, but also model relations between them. We can distinguish semantic and geometric relations between places. In this section we will focus on the semantic relations between fiat places, while the issue of geometric relations has been discussed in Section 2. When considering semantic relations a historical gazetteer should especially be able to represent the membership of a place in an administrative structure. Modelling these relations has several advantages. When for instance the name of a town is not known to a user, the place can nevertheless be found by its context information e.g. by its association to a certain political entity. Furthermore, the position of a place in a hierarchy of rulership already delivers a basic understanding for the people associated with a place. Their rights, duties and status as ruled subjects can in part be derived from the legal status of the place they inhabit. This legal status can express itself in its relation to other places.

When modelling places as defined administrative units, one has to decide between distinguishing the levels in a general (administrative level 0, administrative level 1, administrative level 2, …, administrative level n) or in a specific way (parish, diocese, bishopric). However, while creating such a model, one should keep in mind that those apparently evident hierarchies are the product of a long and complex development. It would thus be necessary to keep the model of these hierarchies as flexible as possible.

Firstly, for the medieval and early modern period, we can distinguish between an ecclesiastical and a seigneurial hierarchy, which may be understood as different layers of society. On the ecclesiastical layer, a place could belong to a diocese, while on the seigneurial layer it is part of a certain county or duchy. Furthermore, there could be different territorial social groupings that can be modeled with additional hierarchies. An example for this would be the districts of certain craftsmanship.5

Until this point, we presumed that hierarchies were not in conflict with each other and therefore not contradictory. But places can also be part of two or more competing hierarchies. This can be the case when multiple authorities lay claim to a single place.6

To illustrate the problems described above we can take a look at the jurisdictional situation in the city of Münster from the 16^th to the 18^th century. As a city set in the prince-episcopat of the same name, it fell under the jurisdiction of the diocesan ecclesiastical court (Offizialatsgericht), which was part of the ecclesiastical jurisdictional hierarchy [26, p. 265]. At the same time the prince-episcopate employed a seigneurial court (Hofgericht) which was responsible for charges not concerning religious matters. On the other hand, the diocesan ecclesiastical court did claim non-religious jurisdiction as well [26, p. 265]. To make matters even more complicated, the (seigneurial) judge, responsible for the city of Münster, was admittedly appointed by the prince-bishop, but the city council could in some way control the administration of justice in the city by employing lay judges [26, p. 262]. One could argue that we have to distinguish between three layers of jurisdictional hierarchy of which the city of Münster was part of: ecclesiastical, seigneurial and the control by the city council. All these hierarchies would in general be in conflict with each other.

If territories are modelled in a gazetteer, it has to be asked how continuity and changeability of their borders and coverage are structured in the ontology. Furthermore, processes like merging or splitting of territories need to be modelled, so some relations for stating if there is any continuity between such operations should be created.

The representation of the territory as an area poses another problem. When dealing with territories that are encompassed by clearly defined boundaries, it is possible to store area information as polygons to maps. However, with medieval and early modern territories this is rarely feasible, although there were material (like using boundary stones, Landwehren or actual landscape descriptions) as well as symbolic ways (ritualistic statements of belonging, for example by staging processions) to mark the border of a territory [53, pp. 106–107].

Even if this kind of border demarcation is preserved, they are not necessarily an accurate mark of historical territories. Borders were often in dispute, so that this status has to be captured as well. These considerations only apply, however, when there was indeed a spatial concept of borders, thus when a dominion was linked to a territory. The polygon-based representation of administrative structures, which is used in modern maps, does not reflect the medieval or early modern situation. In the Middle Ages, ‘ruling’ did not mean ruling over space, but ruling over people. Homogeneous domains often did not yet exist in today’s form, but only developed over time into modern territorial states, demarcated by their clearly defined borders [53, p. 103]. Over the course of this development, the understanding and formation of territorial spaces were constantly changing [53, p. 100]. In cases when borders existed before modern times, they were an approximation and rather separated different spheres of influences than different territories.

Clear boundaries were more simply established for small spaces in which ruling agents were defined, and where they could unambiguously be marked, e.g. by walls, as it is the case for towns, or part of towns like cathedral immunities [52, p. 10].7

Because of the continuous changes of toponyms, place concepts, administrative affiliation or the mere existence of places, a historical gazetteer demands for a model of time. There are numerous possibilities to tackle this issue. Some projects, focusing on contemporary place data, simply use a historical or former tag for places and concepts that are no more. This is, of course, insufficient when using this kind of data for historical research. In the case of historical gazetteers, it is important to relate places and concepts to time in a precise and specific manner. With such an approach one can clarify exactly what place in what condition, and in what time frame is referred to. There are much more complex approaches for dealing with spatio-temporal data which go beyond the scope of this project. Still, this section provides a brief overview on the subject.

Using GIS practices as an example, the easiest way to model time is to understand the whole data set as a approximative representation of the world at a certain point in time. This temporal snapshot can be created for multiple points in time by copying the whole data set and assigning a new temporal index. Such a sequence of snapshots would allow for temporal disambiguation [69, p. 4]. Although easy to implement, this approach does not model relation of continuity between the places – these would have to be created by the user [13, p. 6]. Moreover, it would demand a completeness of data for the different temporal snapshots, which for historical data is hard to achieve due to the fragmentary nature of the historical evidence.

In fact, there are two general ways of conceiving a model for time: the first is based on the concept of timespan in which a statement is valid, the second is based on the point in time at which something has changed, in other words it is based on the event.

When every statement in the gazetteer is attributed with a valid time, it is possible to distinguish different aspects of a place on a high granularity. Not only the place as a whole, but also individual aspects (e.g. its population or predominant religion) can be distinguished by different time intervals. On the other hand, the valid time approach has the disadvantage of not modelling relations of continuity between two time spans, just as in the snapshot approach. There are some solutions to this, most notably the temporal spaceworm approach. Here, different intervals relating to the same object are encompassed by another object to state the relations between them [31]. Within a time span on the other hand, too much continuity can be a problem. A time span usually is not tangible in historical data, therefore creating a time span is bound to a certain research tradition in classifying historical events and epochs. The beginning or end of a time span is only seldom reflected precisely in historical sources. For example, there are seldom historical records that state the exact date when a populated place was founded. More commonly, we only have a terminus post quem for its existence.

This problem, as well as the problem of continuity, can also be addressed by using the second approach to model time – an event-based approach. Numerous ways for modelling events exist on a theoretical level. In general the valid-time approach assigns time spans as attributes to places while with the event-based approach event-objects with or without a time index are modeled. These can be associated with one or more places. One possibility to further distinguish the events in use is to build an additional ontology to represent the events in the gazetteer.8

A number of these problems results from the use of numerical dates, and arise from all the described approaches: Kauppinen et. al. distinguish three types of fuzziness regarding the representation of time: uncertainty (the exact date when a time span began or ended is not known), subjectivity (there are multiple diverging research traditions about the boundaries of a time span) and vagueness (expressions about a time span are made in reference to non-numerical concepts, e.g. “the event happened around the start of spring”) [35, p. 549]. When dealing with populated places in the pre-modern era, one almost always has to deal with uncertainty. The problem can, for example, be approached by using a known terminus ante quem (or terminus post quem). Using a day as a terminus ante quem is of course an approximation in itself.9

One also has to keep in mind that dates in historical sources may make use of different reference systems for dates. Like the Julian and Gregorian calendars, these can exist simultaneously. Although a standardization of these systems may be considered advantageous for comparability, a gazetteer drawing heavily from historical sources should be able to model different reference systems as well to keep the information provided by the sources.10

Designing an ontology as well as modelling a database for historical places is based on the study of historical sources. It is therefore important to state from which sources place data, concepts and properties are derived, since historical data are rarely unambiguous and undisputed. For historians, it is vital to distinguish between different sources that may present different perspectives on the same event or which have different levels of trustworthiness. Thus, for instance, there is a difference whether data derives from a charter certifying a certain act of law between different parties, or whether your data derives from a chronicle written many years after the event by one of the interested parties, for instance after a dispute about this act of law.

On the other hand, one should note that not only actual places but also mythological places played a role in historical sources. Fictitious places were occasionally even depicted in maps. The most popular example may be the Ebstorf Map, but fictional places were also used in maps that were actually used for navigation, like portolan charts [5, p. 106]. It is therefore necessary to make a distinction between fictitious and actual places in an ontology for a historical gazetteer taking into account early modern and medieval times [58, p. 139]. Since fictional places are not subjected to the same rules and as real ones, a semantic gazetteer should also put less constraints on mythological places than on real ones.

Furthermore, in the interest of citability, it might also be desirable to capture which person entered which data. This could become much more important in the future if a contribution to a database by a researcher should be counted as a form of publication [18,41].

4.1. CIDOC CRM

The CIDOC CRM11

OPEN IN VIEWER

Fig. 1.

UML representation of the CIDOC CRM classes discussed in Section 4.1.

The question of the diverging toponyms can be addressed by the property P87 is identified by (identifies): E44 Place Appellation. The class E44 Place Appellation has four subclasses to not only refer to the name of the place (E48 Place Name) but also to its coordinates (E47 Spatial Coordinates), its address (E45 Address) and or to smaller sections of a given place as for instance a gate or the deck of a ship (E46 Section Definition) [59, p. 23]. Describing multiple names can be achieved by assigning multiple Appellation objects to one place. If a gazetteer based on this structure should distinguish place names as less important, the superclass of Place Appellation (E41 Appellation) has the property P139 has alternative form: E41 Appellation to define a secondary appellation.

The problem of topological relations between different territories is tackled in the CIDOC CRM with three properties of the class E53 Place. These can model enclaves (P89 falls within (contains): E53 Place) as well as the fact that two or more territories overlap (P121 overlaps with: E53 Place) or border each other (122 borders with: E53 Place) – regardless of the actual spatial extension of the places. If this extension should be stored, the place-class also offers the property P168 place is defined by (defines place): E94 Space Primitive to assign geodata to an object.

To solve the issues of ambiguity and uncertainty with historical source data, Hiebel et. al. have provided a new approach in their extension of the CIDOC CRM, CRMgeo. They derive two subclasses from E53 Place to represent it: The first, SP2 Phenomenal Place, represents the place how and where it is actually located and shaped, regardless of known, unknown or uncertain geodata or any other human declaration, like its depiction on a historical map or a description of the place found in a source (thereby being regarded as a physical or bona fide object). The second one, SP6 Declarative Place, resembles the concept of fiat object described in Section 3.1. With regard to ambiguity of historical data it can be used as our description of a SP2 Phenomenal Place, based on historical source, archeological findings or guessed approximations. Only this place can be assigned geodata (or a Spacetime Primitive in terms of the CIDOC CRM) [28, pp. 4–6].

This distinction clarifies if one models the actual historical place, or if one talks about a contemporary and possibly flawed representation of it. A fictitious example to illustrate this could be a medieval village that does not exist anymore. Through descriptions in sources about it and archaeological findings we can make educated guesses about its true location and shape. But regardless of how exact our findings may be, we can never be sure that the data we acquire represents all the information of the village as it truly was in the past. To separate our incomplete findings from the historic object, the CIDOC CRM offers two concepts to describe a historic place.

Concerning the representation of temporal change and the matter of temporal disambiguation, the CIDOC CRM mainly focuses on the event-based approach. There are two different variations. Firstly, the use of the class E4 Period. A Period can be an E5 Event which can for example be further specialized by E63 Beginning of Existence or E64 End of Existence. Through E4 Period all these classes get the property P7 took place at (witnessed): E53 Place. Therefore, different events concerning a place, for example it’s foundation, can be modelled.

By using E7 Activity – another subclass of E5 Event – pre-modern privileges held on a place by actors can also be modeled. In the CIDOC CRM an E7 Activity can be carried out (P14 carried out by (performed)) by an E39 Actor. Modelling privileges as E7 Activity would take the ambiguity of medieval and early modern administrative structures and rulership, introduced in Section 3.2, into account.

With Version 6.2.2 some features from CRMgeo proposed by Hiebel et. al. [28] have become part of the CIDOC CRM. The concept centers around the class E92 Spacetime volume which represents a place at a given time period. Through its subclasses E18 Physical Thing and E4 Period a spacetime volume can either emphasize its physical or its temporal properties. The spatial properties of a spacetime volume are represented through the property P161 has spatial projection to E53 Place while the temporal information can be modelled with E52 Time-Span. With the subclass E93 Presence of E92 Spacetime volume, parts or snapshots of space-time-representation of a place can be modeled. With these changes it is possible to model a much more advanced interconnection between places and time.

To describe the provenance of any information modelled with the CIDOC CRM, it can be extended with CRMInf. This ontology provides a framework for making inferences and argumentation as well as modelling the credibility of statements. However, a full discussion of the CRMInf ontology would go beyond this paper. For an extensive discussion of the model – including the underlying concepts – see [9].

One should also note, that the classes, discussed in this section, are only meant to describe real objects. However the CIDOC CRM also provides the class E89 Propositional Object which can be applied for modelling mythological places as well. It’s property P67 refers to (is referred to by) allows such an object to be linked to any other entity of the CIDOC CRM, which can be used to make any statements about a fictional place [60].

OPEN IN VIEWER

Fig. 2.

UML representation of the GND classes discussed in Section 4.2.

For the Gemeinsame Normdatei (GND)12

OPEN IN VIEWER

Listing 1.

An extract about the town of Münster (Westf)

A conceptual disambiguation of places can be done in different ways. Firstly, there are a few subclasses for gndo:PlaceOrGeographicName, with no formal distinction between fiat objects and physical objects.15

OPEN IN VIEWER

Listing 2.

An extract about the Prince-Bishopric of Münster

As shown in line 1, the Prince-Bishopric of Münster can be understood as a gndo:AdministrativeUnit as well as a gndo:ReligiousTerritory. By using multiple inheritance, both the ecclesiastical and the secular aspect of dominion are captured for the territory.

The number of the subclasses provided by the GND ontology is rather limited. Therefore, objects can be further enriched with context by using the property gndo:broaderTerm or gndo:broaderTermInstantial, which means describing concepts not as part of a formalized ontology but as arbitrary descriptions. This can be seen in line 2 of Listing 2. In this case, the referred URI leads to the concept Hochstift (Prince-Bishopric) [8]. All subjects can be further specified by adding an information text via gndo:definition and related terms via gndo:relatedTerm. In this way, a basic human-readable conceptual disambiguation is possible. An automatic processing of concepts attributed this way on the other hand is not given.

The GND ontology allows for the modelling of multiple names for one place, may they be contemporary or historical. However, the model does not distinguish, for instance, between different languages. One of the alternatives is marked with gndo:preferredNameForThePlaceOrGeographicName as the name by which the place is mainly addressed. Other variations are stored by using the property gndo:variantNameForThePlaceOrGeographicName as can be seen in Listing 2 in line 4. There is no model for a temporal disambiguation of the names.

It is not possible to model territories as polygons with the GND ontology. A territory is defined as such by its conceptual and hierarchical structure. Coordinates can only be used to model the most northern, southern, western and eastern point of a territory. If a border itself is of interest for a gazetteer it is grasped as a place and can be modelled with the class gndo:WayBorderOrLine. This could for example be the case when one wants to model events taking place in context to the border.

For temporal disambiguation, the GND ontology uses the timespan approach. The dates of creation and termination can be modelled with the properties gndo:dateOfEstablishment, gndo:dateOfTermination and gndo:dateOfEstablishmentAndTermination. All three can be linked with either a literal describing the temporal information or with another node of the type xsd:dateTime. To capture information of continuity, e.g. when places are merged or split, the properties gndo:succeedingPlaceOrGeographicName and gndo:precedingPlaceOrGeographicName are provided.

The hierarchy model of the GND ontology does not rely on a predefined hierarchical structure for its place data, but on historically contextualized relations between objects. Places can be linked by the properties gndo:placeOrGeographicNameIsMember and gndo:relatedPlaceOrGeographicName. The only other way to arrange places into an administrative hierarchy is the usage of gndo:geographicAreaCode. Therefore, it is not possible to distinguish multiple hierarchy-trees and arrange the places in them. For the Prince-Bishopric of Münster, this can be seen in Listing 2 in line 3. The listed area codes stand for Deutsches Reich, Nordrhein-Westfalen and Niedersachsen. The first area code refers to a historical membership, the others are allowing for a simple spatial placing of the object, referring to the Federal states the Prince-Bishopric’s former territory is currently part of. A temporal disambiguation regarding the membership of elements in hierarchies is not possible with this ontology.

Since the GND ontology is mainly designed to standardize terms for cataloging literature, it does not provide any means to capture provenance of the information concerning historical place data.

In Wikidata, it is possible to describe general as well as specific objects and concepts with entities. As with the GND ontology it is possible to conceptualize a place by relating it to a number of concepts and using multiple inheritance. Using the Prince-Bishopric of Münster (with the identifier Q697254) as an example again, we can illustrate the conceptual inheritance.17

OPEN IN VIEWER

Fig. 3.

UML representation of the Prince-Bishopric of Münster in Wikidata.

Like the GND ontology, Wikidata can handle multiple name variants with one marked as the preferred name. Aside from that there is a property called Former place names (Q7458920). Because all values of properties can be attested qualifiers, the temporal disambiguation of historical names with the valid time approach is possible. However, in practice it seems to be more common for the Wikidata community not to distinguish between name variants and historical names. Both forms are instead stored as aliases to the preferred name in each language version. The alias field of Wikidata does neither provide the means for a temporal disambiguation nor is there a tag to mark a name as historical.

The primary intent of Wikidata is to provide structured information for the different Wikipedia projects. Therefore, it does not provide any means to store polygonal data. Territories can be depicted as SVG-images, but they can of course serve mostly for a visual representation and not an automatic analysis. Again like in the GND ontology, coordinates of the outermost points of a territory can be stored to approximate the extension of a place. On the other hand, a conceptualization with the classes shares border with (P47), enclave (Q171441) or exclave (Q933394) can be used to model basic topological relations between different territories.

Temporal data is also modeled in part with topological relations. The respective properties are called inception (P57) and dissolved or abolished (P576). To model continuity, Predecessor (Q16881385), replaced by (P1366) or followed by (P156) can be used. A further development of the concepts used in Wikidata are the Qualifiers. These are values that can be assigned additionally to each other value of a property. A qualifier can either be an integer or an interval [27]. With this feature it would be possible to distinguish temporal information of each aspect of a place.

Because of the wide scope of Wikidata, its properties for hierarchical arrangement of places are not contextualized for historical territories but rather for the use with general concepts. It is possible to distinguish between the membership of a place in a country, in an administrative and in a jurisdictional district.19

Wikidata offers a way to capture the provenance with its properties stated in (P248) or reference URL (P854). The first one relates to another Wikidata class or object while the second relates directly to another web resource.

The Pleiades ontology has been conceived to model places and spaces in a given historical time, in this case the Antiquity. Its purpose is to distinguish name variants taken from historical sources, and capture the time span in which they were used.

OPEN IN VIEWER

Fig. 4.

UML representation of the Pleiades data model.20

It thus uses specific place concepts that are embedded in a historical context. Although some of the concepts are more general than others, they are not related to each other in a structure of inheritance. Most concepts are defined in a documentation [65]. Because the project’s definition is based on Yi-Fu Tuan’s experience-based approach to places [63], the list consists not only of man-made objects like bridges, forums or villas, but also of natural objects like forests and rivers. Even fictional places are covered in Pleiades although there do not seem to exist specific concepts for them. Such places are marked as erroneous place as can be seen at the example of the mythical town Kikynethos.

The Pleiades data model22

OPEN IN VIEWER

Listing 3.

An extract about Istanbul

As one can see the properties pleiades:start_date (line 5) and pleiades:end_date (line 11) provide a valid time frame for the use of a certain name. One or more epochs in which the name was used can be assigned by using pleiades:during (line 4 in Listing 3).

Furthermore, the Pleiades ontology provides properties to model the accuracy as well as the completeness of toponyms that are taken from historical sources [16]. For both properties there exist three levels from false/fragmentary and not reconstructable to accurate/complete. Names can thus be used from a source that, for example, is known to use a wrong spelling of a name.

The Pleiades ontology does not provide a model to describe territories.

Concerning the time model, the properties pleiades:start_date and pleiades:end_date have already been introduced and can be seen in Listing 3 in lines 5 and 11. Statements in this ontology can not only be valid in time spans marked by dates, but also be assigned to an epoch or time period with pleiades:during. A time period also has an upper and a lower boundary in the form of timestamps. In terms of formal design, this feature bears no difference to the valid time model although it has the advantage that timespans are not arbitrary but contextualized with historiographic information.24

Since the Pleiades project focuses rather on the names of the places than on their underlying political structure, there is no elaborated hierarchical conceptualization. Basic hierarchies, like the membership of a city to a region, are simply referenced by using rdfs:seeAlso.

Finally, since the Pleiades project is based on the mentions of place names in literature and other source material, it also features elements to describe the provenance of Data. With the prov:wasDerivedFrom property taken from the Provenance Interchange Ontology (PROV-O) [50], a place object or a name can be referenced by literature or source material, as can be seen in Listing 3, line 8.

Whereas the Pleiades projects is dedicated to places and spaces in Antiquity, the GOV refers to a more recent epoch, from the 19^th to the 21^st century, with a particular focus on Europe. In 2014, the GOV contained approximately one million entries.25

The ontology of the Genealogisches Ortsverzeichnis (GOV) distinguishes between concepts for fiat objects and physical objects. Physical objects can be different buildings, towns, abandoned villages, churches or market areas while fiat concepts represent man-made virtual units, like secular, ecclesiastical and juridical administrative units. The general concepts are already contextualized. They model specific historical as well as contemporary ecclesiastical and secular administrative hierarchies.26

The GOV ontology also allows multiple designations for a place. The RDF representation of the village Roztoka28

OPEN IN VIEWER

Listing 4.

An extract about Roztoka

Listing 4 shows how two name changes of the village have been modeled in lines 3 and 9 by using gov:hasName. Besides the name of the whole object assigned by gov:value, every name gets a valid time assigned by gov:timeBegin and gov:timeEnd.29

The GOV ontology completely lacks a data model for territories. All places that can be described as a territory are distinguished as such by their arrangement in a hierarchical structure alone.

With the gov:timeBegin and gov:timeEnd properties, the genealogical gazetteer only supports a simple valid time approach.30

The hierarchy levels used are specific and historically contextualized. On a conceptual level, there is a distinction between an administrative, an ecclesiastical, and a jurisdictional hierarchy tree. Conflicting memberships at one of these trees can be resolved by the use of timestamps as is shown in Listing 5, depicting the memberships of the Freistaat Preußen31

OPEN IN VIEWER

Listing 5.

An extract about Freistaat Preußen

For a human user, it is possible to see at one glance the hierarchies a place was part of at a certain time. To achieve this, the whole hierarchy tree for an object is traversed and then visualized when the page of an object is requested. The whole tree for the town of Münster is shown in Fig. 5. Note that the nodes are named with historical terms to achieve a better distinction. The GOV ontology uses a similar concept as Pleiades for capturing provenance of property values. Here, the property gov:sourceRef can link an information to another resource as shown in the example in Listing 6,32

OPEN IN VIEWER

Listing 6.

An extract about Freistaat Preußen

Even if none of the projects discussed above is explicitly designed to take the specific problems of places in medieval and early modern times into account, a number of the challenges stated in Section 3 could already be considered resolved by them. Other issues yet remain untouched. (See Table 1 for an overview.)

Challenges concerning categorization of place have been addressed insufficiently. The distinction between fiat places and physical places as well as general and specific concepts is only done by the CIDOC CRM as well as the GOV ontology. Since the GOV focuses on the 19^th to 21^st century, the ontology would have to be extended for usage for early modern and medieval places. Because the hierarchy levels are also places and, therefore, conceptualized, comparison between new levels and already existing GOV levels would still be possible. Specific concepts using terms that apply to pre-modern times can only be found at Wikidata where they are contextualized through integration in an extensive class hierarchy.

When modelling territories, a geographical approach is used by the GND ontology and Wikidata through approximating the extent of the area with single pairs of coordinates. Topological relations are provided by Wikidata and the CIDOC CRM.

OPEN IN VIEWER

Fig. 5.

The town of Münster as an example for administrative hierarchies in the genealogical gazetteer.33

All ontologies introduced in this paper solve the problem of multiple names. However, only Pleiades takes into account that names retrieved from historical sources can be flawed data and should therefore be modeled as such. To model such cases with the CIDOC CRM, the CRMInf extension can be used, as stated in Section 4.1. As an authority file for names, the GND ontology does not allow to state when a variant of a certain name was being used. All other projects separate the name object(s) from the place object and are therefore able to describe such contextual information.

Problems of temporal disambiguation mostly remain unsolved. Except for the CIDOC CRM, which suggests an event-based approach, the valid time model is solely used for describing change over time.

By separating the historical source from the editor of a data set, only the Pleiades ontology and (with an extension of the CRMInf model) the CIDOC CRM allow in parts a model of provenance considering the needs of academic research. Different types of sources are not described by any of the data models above. The focus yet lies on web resources. Trustworthiness and completeness of sources is also ignored by others than the CIDOC CRM and the Pleiades ontology.

5.1. Place concepts

Finally, we summarize which of the different approaches discussed above are required for a gazetteer that covers medieval and early modern places.

With regard to place concepts, the most important question is the decision between using general or specific categories. General concepts allow for more interoperability and comparability, while specific concepts can be more historically accurate and allow for deeper levels of historical analysis. We propose the use of both in combination with an inheritance structure, so that the specific concepts are specializations of the general ones. The use of multiple inheritance guarantees the specification of more complex and ambiguous historical concepts. An enrichment with context information – for example through the property gndo:broaderTerm – should be avoided.

As shown in the GOV ontology, a distinction between physical objects and fiat objects is also necessary, which makes it possible to attribute different historical or historiographic classifications to otherwise clearly defined objects. An example in use for this could be the distinction between the category of a town and an ideal type that can be assigned to the same place (e.g. fortress town); see [67]. Furthermore, it is important to stress the need for modelling fictitious places. In the examined ontologies, this was mostly disregarded due to the temporal scope of the projects (with CIDOC CRM and Pleiades as an exception).

5.2. Toponymy

The importance of modelling multiple names for each place has been demonstrated. The most common and efficient practice to resolve this issue is to distinguish between the places and its names. The names can be single (place name) objects themselves, and therefore be provided with their own temporal attribution, shown in the GOV and Pleiades. As with other properties, it is important that the provenance of each name of a place can be modelled separately as shown by the GOV ontology in order to ensure the traceability of the used information as necessary basis of scientific work.

5.3. Territories

To be useful for qualitative reasoning, territories can be modelled by their spatial (like topological) relations in addition to (or instead of) their geometric representation. This enables to enrich the models by inferring new inherent relations, for instance relations between current and historical places (see [34,36]). While not offering a representation accurately e.g. by maps (as with coordinates), topological relations also better meet the requirements for representing places that are lacking clear defined boundaries. Wikidata as well as gazetteers based on the introduced properties of the CIDOC CRM, are in general capable of this feature.

If it becomes necessary to visualize the territory on a map, we propose that its extent will be approximated by choosing places whose positioning is less uncertain (like towns and villages) that belong to the respective territory. This provides a much more accurate depiction of medieval and early modern realities, in which authority was defined by power over towns, villages, farmsteads and rights to use forests or stretches of water.

In some cases, it can be useful to include the use of geometries, for example if a gazetteer serves as an object catalog for a GIS application. If territories are also capturing administrative authority in terms of dominion over spaces, it has to be considered to use a model for representing different degrees of administrative penetration. In this case, the overlap representation should also be possible.

The problem remains that most of the approaches looked at have been developed for a world mostly accessed by homogeneous administrative structures. The distinctive features of medieval and early modern dominion, which have been stated in Sections 2 and 3, are ignored by current gazetteer projects when it comes to model ruling structures on the level of towns or villages. To capture these features in a gazetteer for the medieval and early modern world, a new model that understands ruling as an interconnection between places, non-governmental institutions and people has yet to be developed. Such a model of ruling would have to be developed from scratch because of its high dependency from history as a knowledge domain. Nevertheless it should be applicable as an extension to every gazetteer that tries to take pre-modern political realities into account.

5.4. Temporal disambiguation

From the different approaches to model time introduced in Section 3.5, in practice, the valid time approach is used by most of the existing ontologies. Out of those approaches the CIDOC CRM allows a very flexible model of time. The complexity of the valid time model, however, is insufficient to model historical places because it does not capture, for instance, if and how two periods are connected to each other. When changes happen, marked by the transition from one valid time-interval to the next, it is unclear whether or not there is any linkage between the two time periods.

Furthermore, the valid time-interval in itself claims a continuity which is not covered by the sources. In most of the cases historical texts report events. In this case, the continuity is already an interpretation of the data from today’s perspective. Only an event-based approach can account for the problem of modelling continuity between the transition from one time span to another, and the creation of time spans at all. Since historical texts in general inform us about events that resulted in changes to the world rather than predefined time spans in which there was no change, event-based models bear a closer proximity to the source material. Therefore, the aspect of temporal disambiguation has to take a more prominent place in the design and usage of ontologies for historical data.

Finally, all aspects of a place have to be distinguished by temporal properties as shown in the GOV ontology. This includes all attributes of a place as well as its relation to other (place) entities, e.g. its relations in a hierarchy.

5.5. Hierarchies

The need for modelling hierarchies depends on the geographical, temporal, and conceptual scope of an ontology for pre-modern places. The GOV ontology shows how a model based on multiple hierarchy trees can work. This also allows for conflicting claims as well as different administrative structures to be modelled.

5.6. Provenance of data

Since the data in a gazetteer for historical places heavily relies on information drawn from historical sources, it is imperative that the provenance of all statements in such a database is made visible. Ideally, the data model used for capturing provenance is able to distinguish different concepts of historical sources. As shown with the Pleiades ontology, it is to be preferred to also represent the trustworthiness of a source. That way, historical data stored in a digital gazetteer becomes much more reliable and encourages for a wider use.