RE: Are we losing out because of grammars? (Re: Schema ambiguitydetection algorithm for RELAX (1/4))

[Peter Jones <peterj@wrox.com>]

Hi,

V. interesting.

Please specify in more detail what you mean by graph distance.

Thanks,
Peter


-----Original Message-----
From: Bill dehOra [mailto:BdehOra@interx.com]
Sent: 31 January 2001 09:56
To: 'Thomas B. Passin'; Bill dehOra; Bullard, Claude L (Len); Rick
Jelliffe; xml-dev@lists.xml.org
Subject: RE: Are we losing out because of grammars? (Re: Schema
ambiguity detection algorithm for RELAX (1/4))


  
:Suppose I ask for the
:interest rate and I meant a simple rate, but you return the 
:yearly compounded
:rate instead because "interest_rate" was requested and you 
:could supply one.
:I know that's simplistic, I'm just trying to boil things down 
:to simplicity.

No, that's a good example. My impression (and if anyone wants to correct me,
please do) of one way the semantic web intends to deal with this is to point
to a machine readable dictionary definition of interest_rate. So when I
write a query to a Linda space (the intersection between these systems and
RDF is very interesting btw), I need the query to point to a namespace or a
computable definition for interest_rate. When something wants to fill the
interest_rate slot, it has to be sure that its definition of interest_rate
is in the same namespace, or that its interest_rate is equivalent, possibly
by traversing a graph of aliases. One other way is to have enough properties
hanging off interest_rate to compute a match: ultimately I think it boils
down to the first way, which is a lookup on a graph, for want of a better
description. So the fields here are not simple fields, they really stand for
graphs, implict or explicit, and the act of matching becomes an attempt to
merge two graphs. 

That of course means one has to supply an out of band definitions for their
flavour of interest_rate. The interesting thing being, with a graph lookup
you don't have to know that defintion ahead of time to reply. You only have
to know if you can link to its definition. 

I wonder then if there isn't a possibility of diluting the shared meaning as
the graph distance between the two interest_rate's grows? That is with
enough distance, the terms are not at all equivalent despite the linking.
The possibility of the semantic web equivalent of chinese whispers would be
a reasonably serious problem. Does anyone have an example of that?

-Bill
--
Bill de hÓra  :  InterX  :  bdehora@interx.com