<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Archiving and Interchange DTD v1.0 20120330//EN" "JATS-archivearticle1.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink">
  <front>
    <journal-meta />
    <article-meta>
      <title-group>
        <article-title>Annotation Classes: A Structuring Mechanism for OWL Ontologies</article-title>
      </title-group>
      <contrib-group>
        <aff id="aff0">
          <label>0</label>
          <institution>BBN Technologies 1300 N. 17</institution>
        </aff>
      </contrib-group>
      <abstract>
        <p>OWL provides AnnotationProperties as a means of providing extralogical information about ontologies, classes, properties, and individuals. We propose analogous AnnotationClasses as a means of adding hierarchical or other structure to an ontology without additional entailments, primarily for the benefit of human users. We use the RadLex vocabulary as a motivating example and also address the more general problem of understanding OWL ontologies.</p>
      </abstract>
      <kwd-group>
        <kwd>Mike Dean</kwd>
        <kwd>OWL</kwd>
        <kwd>AnnotationClass</kwd>
        <kwd>RadLex</kwd>
        <kwd>ontology visualization</kwd>
      </kwd-group>
    </article-meta>
  </front>
  <body>
    <sec id="sec-1">
      <title>1 Introduction</title>
      <p>
        OWL [
        <xref ref-type="bibr" rid="ref1">1</xref>
        ] supports AnnotationProperties which are specified formally but provide no
logical entailments. OWL defines some AnnotationProperties such as
owl:versionInfo, rdfs:label, and rdfs:comment. Users can also specify their own
AnnotationProperties, and might do so to specify information about how a class
should be rendered or ontology metadata such as dc:creator1.
      </p>
      <p>
        While working with the RadLex vocabulary [
        <xref ref-type="bibr" rid="ref3">3</xref>
        ], we identified a potential use for
AnnotationClasses. These are analogous to AnnotationProperties in that they appear
within the ontology and contribute useful information (particularly for human users)
but don’t lead to additional entailments.
      </p>
      <p>The rest of this paper is organized as follows: Section 2 introduces RadLex and
motivates AnnotationClasses. Section 3 discusses approaches for adding
AnnotationClasses to OWL. Section 4 discusses some more general issues in making
OWL ontologies understandable to humans. Section 5 discusses related work.
Section 6 concludes.</p>
      <sec id="sec-1-1">
        <title>1 http://dublincore.org</title>
      </sec>
    </sec>
    <sec id="sec-2">
      <title>2 RadLex</title>
      <p>RadLex2 is “a lexicon for uniform indexing and retrieval of radiology information
resources” being developed by the Radiological Society of North America. RadLex
was not developed using OWL, but is mappable to OWL.</p>
      <p>RadLex 2.0 consists of 11,962 terms related through 11,373 links. Table 1 shows
the distribution of these links. Links with relation_type “isa” correspond to
rdfs:subClassOf and a small number of rdfs:subPropertyOf relationships. Those with
relation_type "synonymof” correspond primarily to owl:equivalentClass relationships
(but could also include owl:equivalentProperty or owl:sameAs relationships).
“partOf” and the rest of the named relationship_types correspond to
owl:ObjectProperties that are themselves defined as RadLex terms. The 60 links with
an unnamed relation_type motivate our suggestion for AnnotationClasses as discussed
below. Links with a preferred value of 1 are displayed by the RadLex Viewer.</p>
      <p>The terms are arranged in a hierarchy rooted at RadLex term. Figure 1 shows the
top of this hierarchy as represented in the RadLex Viewer at radlex.org. The
hierarchy contains a mixture of isa, partof, containedin, and other relationships. This
works well for organizing major subhierarchies like anatomic entity by location
2 http://radlex.org
(abdomen, head, heck, etc.). Links from Table 1 with a preferred value of 1 are
included in the hierarchy.</p>
      <p>The 60 unnamed (but preferred) links from Table 1 represent relationships that are
meaningful to users but don’t have logical significance. Most of these involve terms
near the top of the hierarchy. Terms involved only in unnamed links led us to think of
AnnotationClasses. The unnamed links themselves could be represented using an
owl:AnnotationProperty.</p>
      <sec id="sec-2-1">
        <title>Radlex</title>
        <p>RadLex attribute
Relationship
RadLex synonym
contained in
contains
member of
member
origin</p>
        <p>Upon closer examination, we found that 62 terms are used in unnamed links.
Table 2 lists the 9 terms that are only used in unnamed links, and are thus candidates
for AnnotationClasses. Of these, the first four seem appropriate; the other five seem
to be included due to artifacts or errors.</p>
        <p>The use of a root class such as RadLex term is common in many ontologies. Often
such classes are meaningless or essentially synonyms for owl:Thing and could
effectively be represented as AnnotationClasses.</p>
      </sec>
    </sec>
    <sec id="sec-3">
      <title>3 Implementation</title>
      <p>We propose that OWL 1.1 include owl:AnnotationClass as an owl:Class. Such
classes would be included by ontology editors and other tools when rendering the
class hierarchy, but would not result in additional rdf:type or rdfs:subClassOf
entailments. This would also avoid logical “clutter” or worse in forward-chaining
reasoners.</p>
      <p>We don’t propose specific changes to the OWL semantics to incorporate
AnnotationClasses.</p>
      <p>A key decision heavily impacting the semantics would be whether to use
rdfs:subClassOf to associate AnnotationClasses and other classes, or to create a new
standard owl:AnnotationProperty. Such a new standard AnnotationProperty linking
classes in hierarchical but extra-logical ways might be more important than
AnnotationClass (e.g. in allowing links between non-AnnotationClasses), and might
be sufficient by itself.</p>
      <p>Like the developers of RadLex, we’re at a bit of a loss for what this new standard
AnnotationProperty should be named.</p>
      <p>
        The universal role introduced in SROIQ [
        <xref ref-type="bibr" rid="ref2">2</xref>
        ]. an implicit object superproperty
analogous to owl:Thing for classes, could potentially be used for this purpose,
although it has logical significance. Such an owl11:universal property has been
discussed3 but is not currently included in OWL 1.1 [
        <xref ref-type="bibr" rid="ref4">4</xref>
        ].
      </p>
      <sec id="sec-3-1">
        <title>3 http://www.w3.org/2007/OWL/wiki/Universal_Property</title>
      </sec>
    </sec>
    <sec id="sec-4">
      <title>4 Understanding OWL Ontologies</title>
      <p>The RadLex hierarchy points to a more general issue in making OWL ontologies
understandable. The mixture of relationships in the RadLex hierarchy seems to be
quite effective, although we would prefer that it be more explicit.</p>
      <p>As a graphically-based representation, UML4 (which is a competitor to OWL in
some respects) has the advantage that it can directly convey important extra-logical
attributes and relationships (as well as aggregation and composition). The most
important class(es) can be placed prominently at the center of the first class diagram,
and colors, subdiagrams, alternative views, and other visual mechanisms can also be
employed.</p>
      <p>Given the unordered RDF statements associated with even the relatively small
Friend-of-a-Friend vocabulary5, consider how difficult it would be to quickly
determine that foaf:Person is the most important class.</p>
      <p>Additional standard owl:AnnotationProperties to identify the key classes in an
ontology (such as a keyClass owl:OntologyProperty) and perhaps even represent
diagrams would also be helpful.</p>
    </sec>
    <sec id="sec-5">
      <title>5 Related Work</title>
      <p>Bijan Parsia’s OWL Annotation System6 proposes a modular means of defining and
grouping annotation properties, including “mustUnderstand” properties with logical
significance. It would provide an effective mechanism for defining additional
standardized annotation properties, including presentation properties as discussed in
Section 4, but does not directly address AnnotationClasses.</p>
    </sec>
    <sec id="sec-6">
      <title>6 Conclusions</title>
      <p>Based on our work with a major emerging vocabulary, RadLex, we propose the
addition of AnnotationClasses and/or a standard owl:AnnotationProperty to OWL 1.1
to hierarchically associate owl:Classes (primarily for the benefit of human users
seeking to understand the ontology) without producing additional entailments. We
believe that RadLex represents a use case for a large class of applications.</p>
      <p>Other extensions to make OWL ontologies more easily understood by humans are
also encouraged.</p>
      <sec id="sec-6-1">
        <title>4 http://www.omg.org/uml/</title>
        <p>5 http://xmlns.com/foaf/0.1/
6 http://www.w3.org/2007/OWL/wiki/Annotation_System</p>
      </sec>
    </sec>
  </body>
  <back>
    <ref-list>
      <ref id="ref1">
        <mixed-citation>
          1.
          <string-name>
            <surname>Dean</surname>
            ,
            <given-names>M.</given-names>
          </string-name>
          ,
          <string-name>
            <surname>Schreiber</surname>
          </string-name>
          , G., eds.
          <source>OWL Web Ontology Language Reference. W3C Recommendation 10 February</source>
          <year>2004</year>
          . http://www.w3.org/TR/2004/REC-owl-ref-
          <volume>20040210</volume>
          /.
        </mixed-citation>
      </ref>
      <ref id="ref2">
        <mixed-citation>
          2.
          <string-name>
            <surname>Horrocks</surname>
            ,
            <given-names>I.</given-names>
          </string-name>
          ,
          <string-name>
            <surname>Kutz</surname>
            ,
            <given-names>O.</given-names>
          </string-name>
          , and
          <string-name>
            <surname>Sattler</surname>
            ,
            <given-names>U.</given-names>
          </string-name>
          <article-title>The Even More Irresistible SROIQ</article-title>
          .
          <source>Proc. 10th Intl. Conf. on Principles of Knowledge Representation and Reasoning (KR</source>
          <year>2006</year>
          ), AAAI Press,
          <year>2006</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref3">
        <mixed-citation>
          3.
          <string-name>
            <surname>Langlotz</surname>
            ,
            <given-names>C.</given-names>
          </string-name>
          <article-title>RadLex: A New Method for Indexing Online Educational Materials</article-title>
          .
          <source>RadioGraphics</source>
          <year>2006</year>
          ;
          <volume>26</volume>
          :
          <fpage>1595</fpage>
          -
          <lpage>1597</lpage>
          . http://radiographics.rsnajnls.org/cgi/content/full/26/6/1595.
        </mixed-citation>
      </ref>
      <ref id="ref4">
        <mixed-citation>
          4.
          <string-name>
            <surname>Motik</surname>
            ,
            <given-names>B.</given-names>
          </string-name>
          ,
          <string-name>
            <surname>Patel-Schneider</surname>
            ,
            <given-names>P.F.</given-names>
          </string-name>
          ,
          <source>and Horrocks, I. OWL 1</source>
          .
          <article-title>1 Web Ontology Language: Structural Specification and Functional-Style Syntax</article-title>
          .
          <source>W3C Working Draft 8 January</source>
          <year>2008</year>
          . http://www.w3.org/TR/2008/WD-owl11
          <string-name>
            <surname>-</surname>
          </string-name>
          syntax-20080108/
        </mixed-citation>
      </ref>
    </ref-list>
  </back>
</article>