Re: Alan Ruttenberg: Re: notes from OWL and RIF datatype coordination meeting

Hi Bijan,

The problem is that the implementation considerations that lead RIF to
want disjointness lead to further consequences (primarily the desire
to use existing implementations of the xpath numeric operators) extend
beyond disjointness. As the report of my discussion with Jos revealed,
it extends into how facets can be used, requiring further changes,
and, because of, for instance, the lack of a precise, implementation
independent definition of them, yield consequences that undermine the
stated goal of interoperability with OWL and across the SW stack.

A discussion of our decision, in my view, needs to address all these
issues and I would be against a decision to choose disjointness if it
only brings an illusory gain.

On Wed, Feb 18, 2009 at 2:28 AM, Bijan Parsia
<bparsia@cs.manchester.ac.uk> wrote:
> On 18 Feb 2009, at 04:45, Alan Ruttenberg wrote:
> [snip]
>>
>> So I must respectfully disagree with your assessment here. While there
>> very well may be *some* engineers and scientists that think of
>> floating point numbers in the way you suggest, there is not a single
>> way in which floating point numbers are used nor a single way in which
>> decimal numbers are used. The representation issue you mention is not
>> unique to floating point numbers, either in their specification, or in
>> their use. I remain of the opinion that the proper way to specify
>> machine representations of numbers is to define their behavior in
>> terms of their mathematical properties.
>
> [snip]
>
> And there is not an univocal structure that underlies "float". There are
> many possible mathematical structures. One of them entails that decimals are
> disjoint from floats. Another does not.
>
> (I have to do some interpretation to figure out what you mean by "specify
> machine representations of numbers...". I presume you mean that we should
> define the number system of OWL via standard mathematical apparatus (we do)
> such that they align with common mathematical connotation of the natural
> language terms (for which either disjointness or non-disjointness is
> justifiable). There is no univocal set of mathematical properties associated
> with the term "float" even up to reasonable isomorphism.

We aren't working with the natural language definition of "float". We
are working with (effectively) the IEEE definition.

> So there is no
> "given" to adjudicate which mathematical structure we select. That is, you
> are *not* on the mathematical or philosophical high ground here, and I would
> appreciate it if you would stop writing as if you were.)

I'm sorry if you don't like my style of writing. However I stand by my
arguments.

> It is perfectly proper to appeal to all sorts of considerations to justify
> the choice of mathematical properties we want of our datatype system
> including mathematical well-behavedness, intuitive appeal, usability,
> implementability, and alignment with other efforts. AFAICT, the
> considerations overwhelmingly favor disjointness. The main reason we went
> *off* disjointness was a rather tendentious argument about usability,  a
> proven wrong argument about implementability, and fatigue. Implementability
> is firmly on the side of disjointness, the usability argument is further
> undermined by the usability costs of diverging from RIF (and standard
> interpretations of XML Schema), and I have a second wind :)

As I've pointed out, we are likely to diverge from RIF because of the
other consequences of them relying on built in numeric operators not
designed with reasoning in mind.

> Finally, I think it would be helpful at this stage to do some work to gauge
> the opinion/mood of the OWL WG. Given that the initial champion of
> disjointness has changed their mind in light of implementation experience

and fatigue

> (hence, new information), then, unless there are actual, new technical
> arguments in the offing

there are, as I detail in my recounting of the discussion with Jos.

, we should get a sense of whether there is consensus
> to continue with disjointness. If not, then we should see if there's "more"
> consensus to change.
>
> Cheers,
> Bijan.
>
> P.S. You may want to read Carnap's "Empiricism, Semantics, and Ontology" and
> Benacerraf's "What number's cannot be" -- just to pick two classic works
> that would, I hope, help dispel your confusions.

Received on Wednesday, 18 February 2009 13:48:28 UTC