[sc34wg3] Representation versus Reification

Patrick Durusau sc34wg3@isotopicmaps.org
Sun, 22 Aug 2004 07:07:32 -0400


While I have often said to Newcomb that he is using private language
in the topic map discussions, I don't think that is the case with
regard to his use of reification.

Actually I think your use of the term "represent" illustrates what is
at its core a difference in views of how to implement topic maps, each
of which has advantages and disadvantages.

 From discussions at the meeting in Montreal, I have the distinct
impression (subject to correction by any Ontopians) that the Ontopia
strategy is one of data conversion (pre-topic map) that assigns PSIs
to subjects that will be represented in a topic map instance. Since
data conversion is never a clean process, the results are examined and
new rules made to refine the assignment of PSIs. But ultimately, the
merging process for topics is based upon a matching of the PSIs so

If that is correct, then yes, I think it is fair to say that PSI's
"represent" the subject, since all you have is the PSI on which to
base any merging decision. As far as the properties that lead to that
PSI, it is simply opaque.

Contrast that with reification as used by Newcomb and others (therefore 
it cannot be a private language, see:

A subject proxy 'reifies' a subject by standing in its place and
by virtue of that 'reification,' it has all the properties therefore
(both identity and other properties) of the subject itself.

Well, not entirely because it can only have the properties of the
subject that are also represented in the topic map but even so, this
is a significant difference from the notion of a PSI that 'represents'
a subject in a topic map.

For example, using reification (Newcomb, ISO 13250 and others) to mean
a subject proxy that stands in the place of the subject and therefore
has all the properties (in the topic map) of the subject, I can compare
two or more subject proxies on the basis of their identity properties
and not simply on whether an opaque representative of the respective
subjects matches of not.

Note that the difference in the approaches lies in reification being
understood as creating a 'stand-in' as it were for the subject that
has (for purposes of the topic map and limited to the properties so
represented) all the properties of the subject.

PSIs, on the other hand, which are a successful strategy for some
cases, is really an extension of the notion of the names of subjects
being unique and merging occurring on the basis of their names (or
representatives to follow the language of your post).

One other difference between PSIs and the subject proxy approaches is
that with subject proxies, the identity of a subject proxy can be
added to as more information becomes available in a topic map about
the subject that is being represented by a subject proxy. With each
merging operation, there is the potential for more information to be
added to the subject identity property of the subject proxy, allowing
increasingly finer grained comparison of subject proxies. Such as the
relationships of subject proxies to other subject proxies. Such fine
grained comparisons are not required but would be specified in a TMA
(Topic Maps Application).

Note that the notion of reification includes the idea of
representatives, since you are always free to decide on what basis
subject proxies will be compared. You could for very good and
pragmatic reasons decide that reified subjects (Newcomb, ISO 13250,
and others, sense) should only have one (or more properties) such as a
PSI as the basis for comparison.

But, representatives (very close to being names) do not carry with them
the notion of subject identity or other properties (because they
represent and do not 'stand in the place of' their subjects), meaning
that our topic maps are information poorer places. All the properties,
including relationships to other subject proxies, cannot be found at any 
given subject proxy.

It may not be appropriate or even useful, in some cases, to have such
rich information in a topic map, but I see no reason to foreclose the
potential for such topic maps to arise.

To me (no surprise), the choices come down to:

1.  Having subject proxies which stand in the place of subjects and
     have all the properties of the subjects they reify (to the extent
     those properties are represented in the topic map), or

2. Having representatives of subjects that offer an opaque string for
    comparison of those representatives.

The first includes the second, as appropriate. The second precludes
the first.

Hope you are having a great day!


Patrick Durusau
Director of Research and Development
Society of Biblical Literature
Chair, V1 - Text Processing: Office and Publishing Systems Interface
Co-Editor, ISO 13250, Topic Maps -- Reference Model

Topic Maps: Human, not artificial, intelligence at work!