[Fwd: [OEP,ALL] Potential FAQ: annotation schema problem

Hi all,

I'm not sure what TF this belongs to, but a recurring question people
have asked me (and I also found this question popping up in the
literature now and then) is this:

"What is the schema (and/or ontology) for my annotations?"

Alternative formulations of the same or related questions are:
  "When do I need an annotation template?"
  "Do I define my template in OWL or in XML Schema?"
  "I know what the output of an ontology editor looks like, but what does
   the output of an annotation tool that uses these ontologies look
  "How do I drive the GUI dialog in my annoation tool given the
   ontologies I need to use?"

I think these questions really reveal two related problems:

1. When people that come from the annotation/meta data angle (e.g. those
who think SemWeb == annotations/meta data/resource descriptions) look
into the literature for answers on this question they discover to their
surprise that many SemWeb papers are about knowledge representation in a
Web context, and do not even discuss how you use this knowledge to
describe you resources.

2. When papers do discuss annotations, it is often assumed that the
annotations are instances of the underlying ontologies (I'm not an
expert in this field, but this assumption may be inherited from the DL
A-box/T-box world view?).  This assumption is often an 
oversimplification: many applications have their own semantic and syntax 
requirements for the meta data.

For example, when annotating a large video, you might want to structure 
the annotations syntactically using the shot/scene _structure_ of the 
video (I mean _structure_ with all the associated syntactic hierarchical 
and ordering information you get in XML for free and is cumbersome to 
represent in RDF triples).

In general, you could say that even if the terms used by the annotations 
are defined by the underlying ontologies, that in itself doesn't make 
the annotations instances of these ontologies.

Another example: in Natasha's first simple example, it turns out that 
even when you already have an ontology defining animals, and an ontology
defining cross-domain information resource descriptors (such as Dublin
Core) you often still need some third schema to put the two together,
for example a third RDF schema that combines the animal and dc schemas
and defines a LionBook as an AnimalBook with dc:subject Lion:

   <AnimalBook rdf:ID="LionBook">
     <dc:subject rdf:resource="#Lion"/>

Note that I do not want to get into the "classes as values" discussion,
I want to discuss "what do my annotations look like and how do I define
this?".  For example, in [1] Guus uses the term "object-description
template" to refer to the structure that defines the annotation
template, that is, the placeholders in which the annotator needs to fill
in terms, the ontologies that are allowed to be used and the parts of
those ontologies that contain legal values.  Also note that the
resulting annotations might not even be in RDF, but, for example, could
also be in XML or some other grammar and only _use_ the terms from the
various relevant ontologies.  In this case, it becomes a question how
you relate the XML instances and/or the XML Schemas involved to the
RDF/OWL terminology they use. See, for example, the use of the
xx:semantics attribute in [2].

I'm wondering to what extent people recognize this problem and whether 
it is worthwile to describe possible solutions in a pattern or FAQ.


[1] Guus Schreiber, Inger Blok, Daan Carlier, Wouter van Gent, Jair
Hokstam, and Uri Roos. A mini-experiment in semantic annotation. In I.
Horrocks, editor, International Semantic Web Conference, Sardinia, June
10-12, 2002, Lecture Notes in Computer Science, Berlin/Heidelberg, 2002.
Springer-Verlag. http://www.cs.vu.nl/~guus/papers/Schreiber02b.pdf

[2] Jane Hunter. Adding Multimedia to the Semantic Web --- Building an
MPEG-7 Ontology. In: International Semantic Web Working Symposium (SWWS)
Stanford University, California, USA July 30 - August 1, 2001

Received on Wednesday, 28 April 2004 12:07:38 UTC