Re: LANG: compliance levels

Ian,

At 11:40 am +0100 2/5/02, Ian Horrocks wrote:
>
>
>I didn't say that, and I was not talking about "modelling style"
>(whatever that is). I was talking about cases of real misuse, or at
>least of imprecision. Examples are easy to find in the daml ontology
>library. E.g., in the world fact book ontology all restrictions are
>universal. I would guess that this is the result of a relatively
>arbitrary choice when translating from some modelling "formalism"
>where there is no alternative (or no distinction between universal and
>existential quantification). In many cases it seems quite clear that
>existential quantification would be more appropriate. E.g., a property
>of countries is their totalArea. This is modelled as a universal
>constraint (if a country has a totalArea then it must be of type
>xsd;decimal) when it seems more reasonable/precise to use an
>existential (all countries have a totalArea and it is of type
>xsd;decimal). Note that this doesn't say that we know what the
>totalArea is or have to specify it, just that a value for the
>totalArea must exist. In this case I would say that the choice of
>universal quantification is not a question of style but simply of
>poor/imprecise modeling.


I think you have to be careful here. I would not say it is 
poor/imprecise: it is simply incomplete. Which is quite normal if the 
KR you have been given has limited expressivity.

At the same time you make a good point: although our proposal is 
always going to be of limited expressivity, it would be nice if we 
can provide enough power to cover these obvious cases.  Otherwise 
people won't be able to be precise even about relatively trivial 
statements.  This is why I think that although leaving everything out 
is a nice escape mechanism, which ensures that nobody loses,  at the 
end we will all lose, because we have not provided enough guidance to 
either implementors or users. If you look at the ontologies around, 
by and large what people do is to specify isa hierarchies, slots and 
provide some additional information about slots. We need to provide 
some decent mechanism to do this in webont level 1.

My view is that a combination of universal (local) range 
restrictions, cardinality, and some creative slot differentiation 
(when needed) allows us  to cover Ian's examples and also will 
provide something to write about for the GUIDE folks. As Deborah has 
pointed out, the opposite is not true.  However, this discussion is 
based on the premise that we cannot have both without excessively 
complicating the language and Ian has objected to this.  So, can 
Frank/Deborah please clarify this? If it turns out that we can have 
both for no great implementation cost, then I would definitely be in 
favour of having both universal and existential local range 
restrictions.

Enrico



>
>>
>>  Universal restrictions are exactly what is found in many OO languages:
>>  "if slot S has a value, then it is of type X"
>>  I find it hard to believe that entire communities don't understand what
>>  they are writing down.
>
>If in these languages universal quantification is the only available
>option (I suspect that in many cases the precise semantics isn't
>clearly specified), that would account for why people are using it a
>lot wouldn't it?
>
>Ian
>
>>
>>  Frank.
>>     ----

Received on Thursday, 2 May 2002 08:06:41 UTC