Building a charter database 3: places and their relationships

Posted: Oct. 25, 2014, 8:51 p.m. by Rachel Stone

One of the main categories of entity in most historical databases is place: where are particular people, institutions and objects located and where did particular events happen? The three main problems in dealing with such historical places are also common to many databases. How should places be identified and mapped and how should hierarchies of places be represented?

Carolingian place names

The Charlemagne project started designing data structures for places by considering the nature of the place data that occurs in Carolingian charters, using a selection of charters from different collections. Place names appear in Latin (or in a Latinised form of the vernacular, such as “Wizenburg” for Wissembourg). The spelling of such names is often inconsistent even within a single charter. While many of the places mentioned had been identified by the editor of each charter, some identifications are uncertain and some places have not been identified at all. Some identifications that are made refer to settlements that have subsequently vanished or been absorbed by other locations.

In addition, the charters contain references to many different types of places. The places mentioned are of various different scales, from a single hamlet or villa up to an entire country or region. As well as names of settlements, there are numerous references to natural features, such as rivers and mountains (and even a few named menhirs). Some places are said to be within other places, e.g. “in pago Texandrisse in villa qui dicitur Dissena” (in the pagus of Tesierbant in the villa which is called Diessen). The use of a wide variety of descriptors associated with place names is also common, e.g. “pagus” and “villa” in the phrase above. These descriptors, however, are not necessarily consistent between charters, e.g. Mainz is called a “civitas” in some charters and an “urbs” in others.

Finally, the place names we are interested in stretch across a number of different modern countries. To make things more complex, at the time that some of the charters were edited, they were in different countries. For example, the MGH edition of Charlemagne’s charters was produced when Alsace-Lorraine was a part of Germany, rather than France.

This consideration of some sample data led to our initial decisions on data structure. We wanted both standardised modern and medieval names for places, but also a way of including all spellings of a place’s name within a specific charter. We also wanted geo-referencing, both to allow mapping and to provide a permanent and “neutral” framework that would endure even if modern administrative boundaries changed. Although settlements can disappear into the sea and new land can be created, specified geo-coordinates can always be entered onto a map to produce results.

We knew we required some mechanism for indicating uncertain identifications of places. We also decided that we needed to attach one or more charter-specific place descriptors to places, since some places are referred to by several different descriptors in the same charter (e.g. from Fulda 216 (ed. Dronke): “trado ... in Teinenheimo marcu vineam unam ... et unum agrum in illa marca id est in Teinenheim ... Acta traditio ista in vico publico Teinenheim” (I hand over one vineyard in the marca of Dienheim...and one field in that marca, that is in Dienheim...This  handing-over was enacted in the public vicus of Dienheim). Finally, we knew we would need some method for dealing with place hierarchies: both modern hierarchies (Aachen is in Germany) and medieval ones (Diessen is in the pagus of Teisterbant ).

The method we devised is shown by this early schematic:

Place data model - schematic 

The key point is that charter-specific information on the left-hand side is separated from standardised place data on the right-hand side. In particular, we decided to regard information about medieval place hierarchies as charter-specific and separate this out from modern place hierarchies. Both PASE and POMS were able to use a relatively stable medieval place hierarchy: that of English and Scottish counties and shires, which persisted well into the twentieth century. In contrast, mapping of Carolingian “counties” (pagi) is a far murkier business. Indeed, some historians have argued that these pagi were not fixed territorial units, covering the whole of the landscape, but instead were patchy or unstable concepts (1). Thus one charter might describe a place X as in pagus Y and another referring to the same place might not mention pagus Y, or might even describe X as in pagus Z. We therefore chose to store such information on place hierarchies in place relationship factoids (PRFs), which will allow users to explore the conceptual space of Carolingian charter scribes, and may reveal more about how pagi and the like were imagined.

Our basic place model, therefore, contains the following main fields:

Modern place name

Standardised medieval place name


Modern place hierarchy

Identification certainty

Because we were recording geo-location information, it has been possible to include maps for many of the places mentioned. A screen-shot of an example place shows the information we include:

Screenshot of Gondreville 

Reference sources

One of the early decisions we made was that we were not going to attempt to second-guess identifications of places by the editors of charters, unless they were obviously wrong. We felt that we did not have the time or local knowledge to do so. Therefore our first reference source is always the edition itself: the majority of these include place identifications, linking the medieval name in the text to its modern equivalent.

For standardised modern place names, place name hierarchies and geo-locations, our main source is GeoNames, although we also make some use of other sources such as Wikipedia and Google Maps. Because the vast majority of our charters are very vague in their references to places, we do not attempt to be completely accurate when geo-locating. For the mapping we are providing, an approximate location is all that is feasible: e.g. Mainz is geo-located to its modern centre, without us attempting to pin-point the exact spot of the Carolingian town. Similarly, vanished or absorbed settlements are geo-located to the nearest modern day settlement.

Settlements are included in a modern place hierarchy, which usually includes both the country and one layer of lower administrative authority. There is some inevitable inconsistency here between different modern countries, because of their differing administrative structures. Thus in France we go down to department level, but in Germany only to Länder level, since including the districts of Germany would make the hierarchies unmanageably large.

We have not attempted to geo-locate natural features, such as rivers or forests, and they are also excluded from the main modern place hierarchies, since many of them cross country boundaries. Instead they are subsumed under the top term of “Natural Landmarks”.

Our main reference source for medieval place names is Orbis Latinus (the 1909 version), since this is freely available on the web. When we began the project, we were not aware of the Pleiades project; this might have been a better reference tool for us, but there are potential problems given that the two projects use different conceptual models of place. The other problem with potential reference tools, as with those for personal names, is that the charters include a lot of names of insignificant places, which are too unimportant to appear in the main medieval gazetteers: almost 60% of them aren’t in Orbis Latinus and I suspect we might draw even more blanks with other sources. Our compromise at the moment is indicating with an asterisk medieval names that we have taken directly from a charter and haven’t been able to standardize: it may be possible in future to use regional studies of medieval place names to standardize these further. However, since we hope to include the option to search for any spelling of the place found within an edition (place name instances), users should still be able to locate the place they are seeking.


Unknown places

Around a quarter of the places we currently have in the database are unknown: we have a medieval name for the place, but we do not have a secure modern identification for the location. The basic rule here is that the place is recorded in the database with its medieval name (which can rarely be standardized) and with the modern name “Unknown XXX”, where XXX is an ID number generated by the database itself. Thus, for example, “Sulzibah” from Charlemagne’s diploma DKAR 1:169 is Unknown 700.

We are not attempting to record unnamed medieval places except in the rare cases where we can physically locate them accurately, such as the “locus quoddam in Avaria” in Charlemagne’s diploma 212, which is said to be at the meeting of the Pielach and Danube rivers, and can thus be geo-located to Melk, the nearest modern settlement.

Our place model is deliberately made flexible enough to incorporate several different types of unknown places, depending on how much information the editor of the charter provides. One possibility is that the editor does not attempt to locate the medieval name precisely or provides only “modern” names that in turn cannot be found in our reference sources; this is a particular issue with older editions. For example, the editor of Charlemagne’s diplomas thinks Sulzibah/Unknown 700 is somewhere on the Salzbach river, a tributary of the Krems, but gives no further details.

With such a vague location, we obviously make no attempt to geo-locate Sulzibah. Nevertheless, even this rough indication of its location does allow us to place it within a hierarchy of modern-day regions. Sulzibah is somewhere within Austria; specifically, it’s within Upper Austria. This information can then be recorded in the place model, so that those interested in a particular modern region can have an indication of events somewhere within it. While we often don’t know the specific modern region for an unknown place, in the vast majority of cases we do at least know the country: we therefore have separate categories in the modern hierarchy for “Unidentified places in France”, etc.

Often, however, editors provide one or more suggested locations for medieval place-names and we have constructed data structures that allow us to link such places together. Any place in the place-model can be linked to multiple alternative places by “place identifications” – the statement that unknown medieval place A either possibly or probably is modern place B (sometimes with extra information provided on the source of this statement). Such modern places can in turn normally be geo-located and thus mapped. We can then construct a place display that represents this information, such as for the place “Teginga” in Mondsee 124:

Screenshot of Teginga 

Since these modern places are created in the database as places and then mapped, users who are browsing areas via zoomable maps can also be alerted to charters that may possibly be associated with a particular modern place, without us needing to make a definitive identification in any particular charter.


Medieval regions

As already mentioned, the charters provide not only information on individual settlements and natural features, but also medieval regions and countries. Entries for medieval regions are created in much the same way as entries for other places, although they are entered into a separate category in the modern place hierarchy and obviously we do not attempt to geo-locate them. The medieval name of the region is taken from Orbis Latinus or from the charter itself. The “modern” forms of medieval region names take a variety of forms, reflecting the variable status of such units. While in West Francia and Italy, medieval regions are often centred on a town or city (e.g. Anjou as the region around Angers, or the Reatino area around Rieti), in other parts of the empire, they are more likely to reflect natural features of the landscape, e.g. the Moselgau in the valley of the Mosel river. If there is a form of the name that is frequently used in the research literature, such as the Moselgau, we will tend to use that. Otherwise, we will often use a modern place name and add the phrase territory to it, e.g. “Bologna (territory)”.

One of the trickiest initial issues was deciding what actually counted as a region. In particular, there was heated argument about the meaning of the term “marca”, since it can be applied to units of all kind s of scale: a vicus or villa can have a marca around it or associated with it, but the Marca_Hispanica covered several medieval counties on the Spanish border. Our division of names between regions and individual locations has probably not been entirely consistent, but the recording of place descriptors means that users interested in the particular meaning of a Latin term for a type of place should be able to retrieve all the examples.


Place relationships

Place relationship factoids (PRFs) were a new concept developed for this project, but the basic principle was fairly simple. They would be charter specific assertions that “Place 1 is in Place 2”, and would also include place descriptors for the two places concerned. Here is a screenshot of the resulting display:

Screenshot of place relationship factoid


The difficulties we have found with PRFs have not concerned the concept so much as deciding what types of information to include within them and also how to extract such statements from a charter text.  We decided early on that we would not record the relationships between places and natural features in such factoids. A large number of charters concerning Wissembourg monastery, for example, mention that is on the Lauter river. Since neither the monastery nor the river have moved subsequently, it was not thought useful to record these statements separately. Instead, the Lauter river is recorded with the place role “Reference point”. This means that anyone particularly interested can check which Wissembourg charters mention the river and which do not, but the user interface is not overloaded with repetitive statements.

We also decided that we would not develop the PRFs or other structures within the database to record boundary clauses, listing the exact extent of property transacted. Our sampling suggested that such boundary clauses were rare and the extra work in setting up structures to record them was not justified. However all named places (hills, rivers, marshes etc) within boundary clauses are recorded as reference points within the database, and so can be retrieved by users. Boundary clauses which refer to the property of other owners are recorded using neighbourhood relationship factoids, which I will discuss in a later post.

We did not impose any restrictions on what types of place could be input for either Place 1 or Place 2. In the majority of cases, Place 1 will be an individual location and Place 2 will be a region, e.g. “Wissembourg is in Speyer (territory)”. But the factoid structure is flexible enough to incorporate two individual locations, one of which is inside the other e.g. a church “in fundo Curia in loco qui nuncupatur Furtunes” (in the fundus of Chur, in the place which is called Fortunes” or record that one region is within another, e.g, that the “ducatus Baioarie” is part of the “regnum Francorum”.

In some cases, a place may be stated as being within several other different places, e.g. the location of Farfa monastery, Acutianus, is often referred to be terms such as “in ducatu Spoletano territorio Sabinensi.” In such cases, we made the decision that we would create two separate PRFs: “Actianus is in Spoleto” and “Actianus is in Sabina (territory)”, but would not attempt to make any statements about the relationship between the duchy of Spoleto and the Sabina territory. Given how little we know about medieval territories, it seems rash to try and place them within any fixed hierarchy.

One of the biggest problems about recording PRFs was deciding when the charter text was actually implying them. For example, how should a statement like that in Fulda ed Stengel 144: "quicquid proprium habeam, quod est in pago Grapleld in marcu Hengistorple et in Hrannungen" (whatever property I have, that is in the pagus of Grabfeld in the marca of Pfersdorf and in Rannungen) be understood? Pfersdorf is clearly in the pagus of Grabfeld, but it's not certain from the syntax that Rannungen is. Because of the need to interpret such statements, an interpretation which cannot be reduced to standardised rules, our use of place relationship factoids is therefore inevitably inconsistent. We have not yet developed the full mapping abilities in the user interface to allow us to “build up” a pagus from the places within it: when we have done that and also input more data, we may be able to make fuller use of PRFs. But the possibility of recording such statements in a structured way already suggests interesting possibilities for other projects.



(1) See e.g. Borgolte, Michael, Geschichte der Grafschaften Alemanniens in fränkischer Zeit. Vorträge und Forschungen Sonderband, 31 (Sigmaringen, Thorbecke, 1984)