Review Comment:
This article makes a nice contribution to the domain of temporal modeling and I gladly support its acceptance. I have a few comments explaining why, list some challenges that may be out of scope here, plus a couple of relatively minor suggestions. I think the paper stands on its own but I'd welcome thoughts on, or flagging of, those challenges.
By articulating the classes of temporal reference systems (TRSes) from ISO 19108, the Time-plus and Time-new extensions to OWL-Time offer a promising start to helping that ontology better support historical and geo-historical knowledge representation. The author properly constrains the paper by making a "stub" for others' formal representations of non-Gregorian calendars and so forth, but demonstrates nicely how named historical periods can be defined, and start points for any calendar also using a form of year/month/day structure (Hebrew, Buddhist, e.g.) can be defined.
I don't think these extensions address a couple of cases I'm especially interested in, but I allow that those needs might be met at a higher, application layer (or by another extension). That is: 1) how do we describe an instant in a multi-year span? For example, how can we express born in (1901 or 1902)? If only "time:year '1901'" is given for a dateTimeDescription, and unitType is unitDay, we can I suppose rely on an as yet unavailable reasoner to know it's some time during that year and not throughout. But how can this be done for multi-year intervals?; and 2) historical data so often arrive as "about" or "circa," a case that might be handled by adding hasLatestBeginning and hasEarliestEnd to hasBeginning and hasEnd, forming a quad that can be used to construct a curve (or ignored if they're not given). This "interval bounded by either instants or intervals" pattern has been used well in Simile Timeline, and in analog fashion for a very long time.
The paper is cogent and highly readable, apart from a couple of minor points:
- Another sentence or two explaining the "complications [which] arise from the involvement of datatypes from XML Schema that are tied to the Gregorian calendar" would be helpful I think. It took some digging to get a grasp of those, which would have been aided by a little elaboration.
- not sure why both "time:intervalBefore geotime:Ordovician" and "time:intervalMeets geotime:Ordovician" appear in the example, aren't they distinct?
Given the nature of historical data, the answer to whether two individuals were spatial-temporal contemporaries is often 'possibly.' Modeling instants, intervals, and named periods such that we can compute probabilities to e.g. drive comparisons or rank search results is an unmet challenge. It could be argued that case and others like it are uncommon, and don't warrant attention in highly general standards. In any case, the extensions in this paper, by permitting Period definitions and topological relations between Periods, while remaining compatible with OWL-Time, constitute real progress. I plan to work with some of its constructs in my own development of the Topotime model and software.
|
Comments
Location of ontologies
Apologies - the ontologies described in the paper are at
http://def.seegrid.csiro.au/ontology/time/plus
http://def.seegrid.csiro.au/ontology/time/new
html, rdf, ttl versions available
(in the text the last "/" appears as a "-")