RE: [SKOS] A new proposal for ISSUE-39 ConceptualMappingLinks

Dear All,
Sorry to come in late. And sorry I'll continue to be here one day, gone
the next, but I guess we all have distractions to tear us away... This
response is built on one of Alistair's.

> 
> First, I'd like to try to summarise the minimum consensus 
> position. I.e. what's the least we can agree on? ...
> 
> Minimum consensus: Using SKOS, it should be possible to state 
> broader, narrower, related and exact (equivalent) semantic 
> links between concepts from different concept schemes. 
Agreed. These relationships have stood the test of time in the context
of ISO 2788, BS 5723, ANSI/NISO Z39.19, going back to 1974, for use
within a single concept scheme. They are widely used and understood.
They were always intended to be used for paradigmatic rather than
syntagmatic relationships (in other words, they should apply in a broad
range of contexts, not just on the basis of co-occurrence in a
particular document or set of documents) therefore they ought to be
applicable for use across a range of applications, vocabularies and
resource collections. There seems every reason to mirror them when
designing mappings between vocabularies. (But using a syntax that makes
it clear they are being applied across schemes, not within a scheme.)
> 
> Moving beyond this, I think there a number of issues with our 
> current proposals which need further discussion.
> 
> I'd like to identify three sub-issues here, which could be 
> discussed independently. I'll try to separate them out here, 
> then respond in more detail to each one in separate mails.
> 
> (ISSUE-39A) Should "grouping" constructs for mapping be 
> included, and if so, what are their semantics?

This topic seems to be about mapping from term A to a combination of
Term B and Term C, where the combination could be done with Boolean AND,
OR, or NOT. 
Somebody asserted that BS 8723 allows use of AND but deprecates OR. This
is only partly true.  BS8723-2, which applies only to relationships
within a thesaurus, indeed recommends avoiding the use of "A USE B OR
C", and explains how to do so. It encourages use of "A USE B + C" where
appropriate (but caution: in practice there is not much commercially
available  software that can handle this complex 3-way relationship) and
it says nothing at all about NOT. In fact, it steers clear of "AND" too,
preferring the symbol "+", so as not to get hung up on Boolean algebra.
BS8723-4, the part of the standard that has just been published this
December, deals with mappings between vocabularies and has a couple of
pages of discussion on one-to-many and many-to-one mappings (which it
does not simply equate to use of Boolean operators. Instead it
recommends use of symbols + and | ). One of the problems is that the way
they work  for conversion of search statements is different from the way
they work  for conversion of index terms. 
I think it was Margherita who made the point that many-to-one is hard to
handle, and effectively means that complex mappings usually work well in
one direction only (i.e. one-to-many). The SKOS community might want to
study some good use cases before reaching conclusions on this issue.

> 
> (ISSUE-39B) Is it necessary to have parallel vocabulary 
> (skos:broader // skos:broadMatch etc.)? If not, how do you 
> differentiate between intra-scheme vs. inter-scheme semantic links?

I'm not into the technicalities enough to recommend how; but certainly I
agree on the principle that it is important to find a way of doing so.

> 
> (ISSUE-39C) What's the difference between "related" and 
> "overlapping"? Is there enough precedent to justify a new 
> property for "overlapping"?
Three options have been discussed, namely related, overlapping and
inexact. I guess I am biased by long familiarity with "related". I feel
it is sufficient to choose just one, and that it should be "related". It
seems to me that "overlapping" is a subset of related, and harder for
simple people to use in a hurry. The definition of the associative
relationship (RT/RT) in a thesaurus has always been subjective, which
some people see as a failing (but I see it as a strength, so long as we
recognise that the fuzziness is present). 

Just by the way, I agree with what seems now to be the general choice to
drop majorMatch and minorMatch. I liked the sound of these when they
were first proposed, and they could still have uses in the context of
particular predefined document collections, where syntagmatic
relationships have applications. But they are not so workable if you
want the mappings to serve for collections that may grow in
unpredictable directions.

Cheers
Stella
*****************************************************
Stella Dextre Clarke
Information Consultant
Luke House, West Hendred, Wantage, Oxon, OX12 8RR, UK
Tel: 01235-833-298
Fax: 01235-863-298
SDClarke@LukeHouse.demon.co.uk
*****************************************************

Received on Friday, 7 December 2007 15:44:00 UTC