Re: ISSUE 77 and postcoordination [and ISSUE-40!]

Jakob,

 > "Semanticists generally recognize two sorts of meaning that an
 > expression (such as the sentence, "John ate a bagel") may have: (1) the
 > relation that the expression, broken down into its constituent parts
 > (signs), has to things and situations in the real world as well as
 > possible worlds, and (2) the relation the signs have to other signs,
 > such as the sorts of mental signs that are conceived of as concepts."
 > Every RDF graph has only semantic in the second sense unless human
 > beeings use it for some purpose in the real world. A KOS encoded in SKOS
 > does not carry any practical meaning unless you know that it can be used
 > for indexing resources.

I would not agree here, although I don't understand what do you mean by 
'practical meaning'. My view is that KOS itself establishes the meaning 
of a concept within the system. Hence SKOS or any other carrier of KOS 
is not likely to strip KOS of the meaning that can be implied from 
relationships between concepts in the scheme. Concept in KOS has meaning 
whether this is attached to a resource or not.

Semantic field of KOS concept and semantic field of resource content 
(that is described by KOS concept) could be represented with two circles 
that may overlap to any degree or contain each other). This is because 
indexing process can never be 100% accurate and also because in real it 
is not likely to find resources that can be described with one concept 
only.

My feeling is that we have problem here, not on the philosophical level 
(Frege,  Wittgenstain, Saussure, Pierce, Chomsky etc...:-) but rather on 
agreeing of application scenarios and defining functional requirements 
with respect to the following

a) vocabularies that are not born digital or XML/RDF - linked to a large 
amount of legacy data. They are developed with various level of semantic 
rigour - they are applied with a different level of sophistication - 
they are likely to continue to be used for some time

b) emerging vocabularies: created by humans, extracted by 
auto-categorisation and data mining tools or extracted using new 
semantic technology applications (RDF. For these we need a good standard 
framework that would impose rigour and automate application

On this list we  very often stumble over the hurdle of reconciliation 
between two types of requirements corresponding to a) and b)
My first assumption, and the one that anyone could hastily make reading 
SKOS guide is that SKOS aims at supporting both scenarios

But maybe this issue will have to be revisited upon completion - in case 
SKOS is neededed more for b) than for a).

The field of a) vocabularies is well established it has been extensively 
written about and researched in IR context past 50 years. Terminology 
used in this field is well established and to a great extent unambiguous.

I admit I am guilty of not knowing much about b) and I found learning 
about this kind of applications rather hard as there is no well 
established corpus of literature  - but mostly because authors use 
terminology by approximation and shift it from the field a) to field b) 
with similar but different meaning (examples: taxonomy, collection, 
pre-coordination).

It sometimes causes less confusion and prevents obfuscation if new 
concepts or data elements are given new and very different names.


 > If you ignore this "layers above" then SKOS vocabularies are irrelevant
 > at all. What do you want to do with a vocabulary encoded in SKOS but
 > indexing resources?

The use of vocabularies is a matter that should be considered 
irrespective of whether these are expressed in SKOS, Topic Map, MARC or 
some proprietary schema. I can think of the following:

- maintenance and development of KOS
- extension of KOS (mapping)
- translation of KOS
- distribution/publishing/exachange
- re-use in building of other new KOS
- support of automatic indexing and author NLP based tools


cheers

aida

Received on Monday, 17 March 2008 16:39:28 UTC