Re: [Fwd: OWL reasoning with rules]

Hi Tanel,

Tanel Tammet wrote:
> Axel Polleres wrote:
> 
>>>
>>> I have noticed that many of the attempts at OWL reasoning using rules
>>> run into problems on dealing with OWL-DL.
>>>
>>> As I understand it, this is because OWL reasoning gets complex when we
>>> include things like nominals and some cardinality.
>>
>>
>> Yes, rules and OWL DL are not straightforward compatible due to 
>> decidability issues.
> 
> I am not sure what you mean by compatibility here.

Ok, that the combination with monotonic rules is farily clear if you 
treat them as Horn rules like SWRL, the papers I mentioned cover more 
subtle aspects as well, such as different types of combinations

- combination via entailment vs. per model
- UNA vs. non-UNA
- etc.

> It is certainly true that once you make your logical formulas (in OWL DL,
> some rule language, etc: does not really matter) sufficiently complex,
> you will no longer have an algorithm capable of deciding the formula.
> 
> Prop logic is decidable, predicate logic is not. Description logics walk 
> the fine line,

Which does not need to be crossed when extending by (even nonmonotonic) 
rules, as the mentioned papers demonstrate.

> trying to be relatively expressive while still being decidable. OWL full is
> over the line: it is expressive enough to be undecidable. Any nontrivial
> rule language is also undecidable. Clearly, if you add an undecidable rule
> language to something, the end result is still undecidable.

> When people speak about "rule languages", it is typically unclear
> what they mean: is it pure predicate logic, a subset of latter, does
> it contain nonmonotonic logic, does it contain closed world assumption?
> They all behave hugely differently.

Even when combining with SWRL alone (which is basically Horn, even less, 
i.e. purely monotonic) you run into decidability issues. The articles 
mentioned in my last mail treat all these aspects in more depth.

In th RIF WG, we are currently trying to classify diffferent rule 
systems/languages... but we are only at the beginnning, since, as you 
say, this is a quite complex task:

http://www.w3.org/2005/rules/wg/wiki/Rulesystem_Arrangement_Framework

Comments to the RIF list public-rif-comments@w3.org
very welcome!

> On one hand, this decidability stuff is all basic logic and algorithm
> theory: owl and specific rule languages are in no special position
> here. On the other hand - from the practical standpoint - decidability
> is not that important in itself: we have much more important issues
> where we behave miserably, like existence of clear semantics
> (rdfs semantics is a horrible mess, and owl - inevitably - likewise),
> ease of understanding (all our stuff is hugely complex when
> compared to, say, sql or prolog), efficiency on large datasets
> (none present when compared to sql engines or, say, swi prolog)
> etc.

Agreement on this last paragraph from my side ;-)

axel


-- 
Dr. Axel Polleres
email: axel@polleres.net  url: http://www.polleres.net/

Received on Wednesday, 4 October 2006 11:04:09 UTC