SHO-Place

My doctoral work in geographic information science (2010) entailed developing six modeling patterns for representing historical knowledge in spatial databases, and a ‘spatial history ontology’ accounting for all the entities and relations in those patterns.

Those patterns were for Event Objects and Participation, Groups and Membership, Historical Periods, Historical Processes, Attribution, and Place. Although development of the SHO was suspended, several conceptual elements of the six patterns have remained salientThe patterns were called “geo-historical information constructs” in that work, and referred to the ontology as “the SHO.” . The Place pattern modeled attestations about geographic features, whether physical (cities, mountains, etc.) or non-physical (e.g. political administrative bodies or constructed regions). This view of place corresponds with that of (Janowicz 2009), as “abstract entities used to structure knowledge and to ease communication.” Assertions about places can include any variant names, types, and geometric shapes associated with them over time. Key takeaways from this construct include (i) places are not simply locations, but sets of statements about the settings for events and features, potentially from multiple sources, and (ii) both places and their geographic referrents are dynamic—names, types, and spatial extents change over time and must be temporally scoped in historical gazetteers where that information is available.

The six patterns and the associated ontology were instantiated and evaluated in a relational spatial database using several exemplar datasets, however they had yet to meet the “real world” of software development for Digital Humanities research projects.

Topotime

In the course of designing Kindred Britain http://kindred.stanford.edu , an application connecting the lives of 30,000 Britons over many centuries, my colleague Elijah Meeks and I were challenged to represent a great many vague and otherwise uncertain temporal expressions (“about 1830,” “between 1825 and 1830,” “after 1900,” etc.). We began developing the modelhttp://dh.stanford.edu/topotime/ to handle these and many other cases, in order to visualize that uncertainty in timelines, and to enable temporal filtering.

A core feature of Topotime was that all temporal expressions are intervals. An expression could be of a single year, month or day, but could also be described with up to four elements: start, latest-start, earliest-end, and end. A single year in this case would be interpreted by Topotime-compatible software as having an earliest-start of 1 January and a latest-end of 31 December. If the assertion was for example that an event ended some time in the 1830s, the data creator could express that “fuzziness” with earliest-end and end, and it would be reflected in timelines and filtering functionality. This conceptual model (Fig. 1) is relatively simple and not novel. Visualizations of it can be found in timeline illusstrations dating at least as far back as the 19th century (Rosenberg & Grafton 2013), and our computational strategy derived from work by Tomi Kaupinnen and colleagues (2010). I also explored how temporal geometry derived from such data could permit some probablistic reasoning (Fig. 2).

Topotime grew quite elaborate, including for example operators on datetime expressions like “~”, “>”, and “<” for “approximately,” “after,” and “before” respectively. In time we discovered it was in fact too complicated to be readily adopted by historical researchers NB: a similarly elaborate set of operators has since appeared in an Extended Date Time Format (EDTF); https://www.loc.gov/standards/datetime/edtf.html proposed by the US Library of Congress, and in proposed extensions to the ISO 8601-1 standard (ISO 8601-2).. However, the four-part timespan can accommodate much of the uncertainty found in historical dating, and it appeared in subsequent modeling efforts.

Figure 1 - Timespan with uncertain start and end

Figure 2 – A timespan as a probability function, rendered as a curve

Setting

All phenomena occur somewhere, and for historical researchers that location is almost always a subject of concern or at least a dimension of interest. By and large, where and when go hand-in-hand as key dimensions of who (people, groups) and what (events, objects) data. It seems logical that our encoding of historical sources and corresponding data models should reflect this.

Temporal extents—the “when” of “where”—are in many cases described only by a named historical period, e.g. “Victorian Age,” or “Early Iron Age Anatolia.” There are now at least two gazetteers for named periods, PeriodO and chronOntology, resources that can resolve period names to the set of timespans asserted for them by multiple scholarsPeriodO (http://perio.do) and chronOntology (http://chronontology.dainst.org/) .

It is a striking fact that attested places and asserted periods share some basic modeling requirements. Both are social objects, the “abstract entities used to structure knowledge and to ease communication” Janowicz referred to (2009). Both can have multiple names and metric extents. Place records need to be temporally scoped, and Period records almost always have integral spatial coverage. For example, there are multiple assertions about the temporal extent of the Victorian Age, and the term is largely meaningless outside of Great Britain. The Neolithic, Iron, and Bronze Ages “took place” over timespans that vary by region.

Another way to put this is that both place and period have settings (Fig. 3), comprised of both spatial and temporal extents, described formally as an ontology design pattern in “Place, Period, and Setting for Linked Data Gazetteers” (Grossner, Janowicz, & Keßler 2016).

Figure 3 - A Setting pattern for Place and Period

GeoJSON-T

At this stage, the requirements for a generic model or format representing places as sets of assertions about dynamic attributes was becoming clearer. Furthermore, the solution might be applicable to certain events and to named periods as well.

Over the past decade, the GeoJSON http://geojson.org format for representing geographic features has been a defacto standard for web mapping applications. First specified formally in 2008, it was adopted as an Internet Engineering Task Force (IETF) standard in 2016

(RFC7946). Numerous software libraries can convert data to and from GeoJSON, and/or parse and render it for maps. Although GeoJSON records can have any number of “properties” of any kind, it lacks a recommended form for temporal attributes.

Although developers of the GeoJSON standard acknowledge there are “event-like geographic features,” a brief effort to account for them stalled in 2016. In response to this perceived need, and building on the modeling activity discussed above, I have developed a prospective extension to GeoJSON, GeoJSON-T. Its key addition is simple: an optional “when” element which can be included in any Feature as either (i) a sibling element of “geometry,” applying to the entire feature, or (ii) as a sibling of

“coordinates” within any feature geometry. In the latter case, a GeometryCollection for a single feature could thereby have any number of temporally scoped extents.

As indicated in Figure 4 below, the “when” object allows specifying either four-part timespans as described earlier, the URI for a named period from a period gazetteer, or both. Details of this work-in-progress format appear on GitHub https://github.com/kgeographer/geojson-t.

Figure 4 - (a) “when” added to a Feature; (b) example "when" detail

Linked Places

All of this modelling activity, undertaken in phases over a ten-year period, has informed the development in 2018 of a Linked Places (LP) format for contributions to both World-Historical Gazetteer (WHG) and Pelagios’ Peripleo project. Both WHG and Peripleo accept contributions of place data developed by historical researchers, and both are agnostic as to the models contributors use in creating and exposing that data for their projects. However, in order for aggregation systems to operate efficiently, contributors must make a serialization (i.e. export) to a standardized “interconnection format.”

A small working groupThe author was joined in this by Rainer Simon (Pelagios), and Graham Klyne and Arno Bosse (Oxford) has developed LP format for this purpose. It is both valid JSON-LD and GeoJSON (with the –T extension). Software libraries recognizing and parsing those formats will ignore GeoJSON-T “when” and all other foreign elements. As JSON-LD, it is also valid RDF. The full specification allows for a quite elaborate representation of place attestations, but most elements are optional. WHG will also support a compatible but less complex CSV alternative for contributors whose data is relatively simple.

As shown in Figure 5 (and described in more detail on GitHub https://github.com/LinkedPasts/linked-places ), LP format considers that a Place record may potentially have multiple temporally scoped names, types, geometries, and relations to other places. Each of these can therefore contain associated “when” elements. For example, a place might over time have changed not only names, but type (town to city), size/shape, and containing polity. Most contributors’ records will not include such detail, but it is typical for several large historical GIS projects. A Place record can also have any number of descriptions, depictions, and—critically—links. Links are assertions made by a contributor that a given record refers to the same place as a record in a name authority or another web-accessible Linked Data resource. WHG will provide reconciliation services that allow contributors to perform such matching and to review results for accuracy.

Figure 5 - Conceptual model of Linked Places format

Conclusion

Assertions that one place record is a close or exact match with another is arguably the key ingredient of Linked Data. If two research groups’ data refers to the same place, their data can be effectively linked, becoming part of a potentially very elaborate integrated description of the place. The context provided by temporally scoped attributes, including location, are also central to a successful matching process. Further details of forthcoming WHG functionality are beyond the scope of this paper. Its full version 1 release is scheduled for early 2020, and several interim Beta releases are planned.

References Grossner, K., Janowicz, K. and Keßler, C. (2016). . In R. Mostern, H. Southall, and M.L. Berman (Eds.). . Bloomington: Indiana University Press. Retrieved from http://carsten.io/Grossner_Janowicz_Kessler-Place_Period_Setting.pdf Grossner, K. (2010). . Ph.D. dissertation, University of California, Santa Barbara, United States — California. ProQuest Dissertations & Theses. (Publication No. AAT 3439428). Retrieved from http://kgeographer.org/pubs/Grossner_dissertation_2010.pdf Janowicz, K. (2009). . Paper presented at the Cultural Heritage of Historic European Cities and Public Participatory GIS Workshop, 17-18 Sep 2009, The University of York, UK. Retrieved 14 Jan 2010 Kauppinen, T., Mantegari, G., Paakkarinen, P., Kuittinen, H., Hyvönen, E., & Bandini, S. (2010). . , Volume 68, Issue 9, pp. 549-560 Rosenberg, D., & Grafton, A. (2013). . Princeton Architectural Press.