In this paper we describe recent work in adapting various new OWL and ontology standards to ontology development for the IC and DoD. We present work done to adapt the Universal Core Semantic Layer (UCore SL) standard ontology to support intelligence analysts. We show how new features in the OWL 2 standard can be used to make such ontologies simpler and more readable, and how they facilitate modeling the relationships of concepts across models. We present a proposed standard security model using OWL 2. We conclude with planned future ontology development using these standards.
8283 Greensboro Drive
Standards
In our most recent work, we were tasked with supporting a group of analysts by devising a consistent and inter-related set of models for their wide range of data sources and analytical processes, covering the usual assortment of people, organizations, and places, as well as numerous kinds of materials, equipment, and processes. We elected to create a set of ontologies, mapping to OWL each of: the schema for the desired subset of each data source (MIDB, TIDE, Artemis, …) the Palantir ontology we developed with the analysts the common organizational models called PMESII and CTAF We also created a “master” OWL ontology, based on UCore SL, which covered all the concepts of interest to the analysts, and provided the OWL declarations needed to relate the concepts across all the other ontologies, for data mapping and correlation purposes.
In order to do this, we of course needed to extend UCore SL, adding whole
sublattices of concepts under various of its concepts. For example, we have a handful of
new classes refining UCore SL’s ActOfCommunication. Similarly, we have new classes
under its Vehicle and Sensor. In doing this, we borrowed heavily from SUMO [
<rdfs:subClassOf rdf:resource="#Device"/> </owl:Class> <owl:Class rdf:about="&meped;Equipment">
<rdfs:subClassOf rdf:resource="#Device"/> </owl:Class>
We also found it useful to borrow from SUMO to impose a bit more structure and detail in other areas, such as Geophysical and Geopolitical concepts.
The above examples follow the UCore SL practice of carefully declaring all the disjointWith relationships, including declaring each pair (redundantly) in both directions. One of the new features in OWL 2 is a pair of constructs for declaring this information in a cleaner, more compact fashion. Since some of the classes above are allowed to overlap (for example, Weapon can overlap both ExplosiveDevice and Vehicle), we don’t have a nice clean partition which would enable removing all the disjointWith’s, but using the new AllDisjointClasses still helps somewhat: <owl:AllDisjointClasses> <owl:members rdf:parseType="Collection"> <owl:Class rdf:about="#Equipment"/> <owl:Class rdf:about="#ExplosiveDevice"/> <owl:Class rdf:about="#Sensor"/> <owl:Class rdf:about="#Vehicle"/> </owl:members> </owl:AllDisjointClasses> <owl:Class rdf:ID="Equipment"> <rdfs:subClassOf rdf:resource="#Device"/> <owl:disjointWith rdf:resource="#Weapon"/> <rdfs:subClassOf rdf:resource="&art;Equipment"/> </owl:Class> <owl:Class rdf:ID="Sensor"> <rdfs:subClassOf rdf:resource="#MeasuringDevice"/> <owl:equivalentClass rdf:resource="&ucsl;Sensor"/> <owl:equivalentClass rdf:resource="&pal;Sensor"/> </owl:Class> <owl:Class rdf:ID="Vehicle"> <rdfs:subClassOf rdf:resource="#Device"/> <owl:disjointWith rdf:resource="#CommunicationDevice"/> <owl:disjointWith rdf:resource="#MeasuringDevice"/> <owl:equivalentClass rdf:resource="&ucsl;Vehicle"/> <owl:equivalentClass rdf:resource="&sumo;Vehicle"/> <owl:equivalentClass rdf:resource="&pal;Vehicle"/> <owl:equivalentClass rdf:resource="&avrs;Conveyance"/> <owl:equivalentClass rdf:resource="&tide;Vehicle"/> <rdfs:subClassOf rdf:resource="&meped;Equipment"/> </owl:Class> … etc …
One of the principles in our modeling work was to represent all multi-faceted things as first-class objects, with classes in the ontology. In particular, it was clear that Locations should be represented in that way. By attaching properties to a Location, such as location containment (address contained in city contained in etc), location adjacency, location position (coordinates), even the Political, Military, Economic, etc circumstances of a location, the door is opened to reasoning about locations and the things at those locations. Some of the RDB models we worked with made the same decisions on first-class objects, but many did not. For example, to relate a Location to some Person, Organization, Event, et al, the value of the relationship (birthplace, residence, affiliation, destination, et al) would often be, not a pointer to a Location record, but simply a string naming the location (often, just a country name). Since one of our goals was to relate concepts across models, these string-vs-object differences were a problem. Again, OWL 2 introduces a handy construct which makes it possible to relate the two approaches. If, say, model a represents birthCountry as simply the name of a country, whereas model b represents birthCountry as a link to a country which has a name, we can indicate the equivalence via: <rdf:Description rdf:about="&a;birthCountry"> <owl:propertyChainAxiom rdf:parseType="Collection"> <owl:ObjectProperty rdf:about="&b;birthCountry"/> <owl:DatatypeProperty rdf:about="&b;name"/> </owl:propertyChainAxiom> </rdf:Description>
When the new OWL 2 model was discussed at the 2008 Semantic Technology Conference, it was noted that the new annotation property capabilities were suited for capturing information such as security, provenance, and confidence, all uses of great interest to this community. We have accordingly mapped the IC’s recently-released XML security model, IC-ISM v3, into an OWL ontology called ISM3 using the new constructs.
We have defined a property for each of the ISM v3 XML attributes, a Security class as their domain, and a security annotation property to relate a Security class instance to anything. We have mapped each of the “CVEs” (Controlled Vocabulary Enumerations) defined by the IC-ISM v3 XML specification into the OWL equivalent. For example: <owl:Class rdf:ID="CVE_Classification_US"> <rdfs:label>CVE: Classification (US)</rdfs:label> <rdfs:comment>allowed values for a classification, USonly</rdfs:comment> <owl:oneOf rdf:parseType="Collection"> <owl:Thing rdf:about="#U"> <rdfs:comment>UNCLASSIFIED</rdfs:comment> <ism:security rdf:resource="#U-USA"/> </owl:Thing> <owl:Thing rdf:about="#C"> <rdfs:comment>CONFIDENTIAL</rdfs:comment> <ism:security rdf:resource="#U-USA"/> </owl:Thing> <owl:Thing rdf:about="#S"> <rdfs:comment>SECRET</rdfs:comment> <ism:security rdf:resource="#U-USA"/> </owl:Thing> <owl:Thing rdf:about="#TS"> <rdfs:comment>TOP SECRET</rdfs:comment> <ism:security rdf:resource="#U-USA"/> </owl:Thing> </owl:oneOf> </owl:Class> <ism:Security rdf:ID="U-USA">
<ism:classification rdf:resource="#U"/>
In contrast to the usage of Security above, which annotates each entry in the enumeration with its security markings, we note that usual practice would be the use of annotated axioms, each of which simultaneously asserts and annotates a triple: <owl:Axiom> <owl:annotatedSource rdf:resource="#ID1"/> <owl:annotatedProperty rdf:resource="http://example.com/example.owl#memberOf"/> <owl:annotatedTarget rdf:resource="#ID2"/> <ism:security rdf:resource="#Sec1"/> </owl:Axiom>
We should note that ICS500-21 "Tagging of Intelligence and Intelligence-Related Information" directs that all XML documents shall use the ISM XML standard for security markings. This is of course impossible for XML languages such as RDF/XML. But the rationale for that directive is obvious, and applies to OWL data as well. We urge the community to agree on a standard OWL ontology for security, so that it can be approved as an alternative, and provide the same benefits for OWL use that agreeing on ISM XML does for XML use. We offer this as a possible approach for that standard. We suggest that a similar standard for provenance (sourcing) would be beneficial as well.
We plan to: incorporate mappings to UCore SL into the METS ontology return to the other project to model and map additional data sources and concepts continue retrofitting OWL 2 constructs in both continue devising ontologies such as IC-ISM v3, ideally in coordination with others across the community