Re: [BIONT-DSE] Inclusion versus exclusion criteria

On 14 Sep 2007, at 14:12, Chimezie Ogbuji wrote:

[snip]
>> But I don't know of a way (at least a standard way) to go from first
>> order horn (without function symbols) to Datalog under LP semantics
>> for beyond ground entailments. My intuition sez it'd be pretty bad.
>
> I would agree about this being problematic.  But (thinking out  
> loud) at
> least where the evaluation is being performed by a production rule
> system,

I think you confuse the proof theory with the logic. If you are using  
a Rete system to evaluate positive datalog, then you can just talk  
about datalog.

> couldn't you simply skolemize the existential variables in the
> conclusion set?

I'm not sure what you have in mind.

> /me wonders what the appropriate forum is for these kinds of questions

public-owl-dev?

>>> is a very important contribution towards reducing the
>>> (currently) intractable nature of inference over large ontologies
>>> (especially EL++ ontologies
>>
>> It's not intractable currently. See CEL and Hermit.
>
> Okay then, how about http://www.cyc.com/2004/06/04/cyc ? :)
>
>> DLP is not really very helpful for dealing with EL++. EL++ is already
>> well known to be horn and there are several techniques for dealing
>> with it. The latest is hypertableau. See the link from my blog post:
>> 	<http://clarkparsia.com/weblog/2007/06/25/scaling-owl-two-new-ways/>
>
> My interest in DLP (and others as well), is not simply that it  
> gives you
> an alternative way to evaluate certain fragments of DL, but that it  
> does
> this by *mapping* to a rule language in such a way that if your  
> primary
> mechanism for evaluation is (for instance) a production rule system  
> you
> can use rules (which capture semantics outside of what DL is  
> capable of)
> in combination with ontologies.

This is true for the other horn fragments, pretty much. Consider how  
Boris came up with (and evaluates) DL Safe rules.

> This is incredibly useful for decision
> support (contrary to your claim below).

I think that *this* coupling with *production rules* (wherein you  
expect e.g., chunks of code to fire) is not clearly worked out. It  
certainly doesn't "just work" in the sense of you can reliably  
predict things. It's not much better than coupling with Java code,  
expect perhaps if you prefer the rules.

There are many ways of doing these couplings. One of my favorite is  
Eiterian hex predicates.

>   Bear in mind, DL is a KR with
> explicitly-engineered restrictions in expressiveness,

Chimezie, I think that it is perfectly safe for you to assume that  
not only am I *well* aware of that fact, but that whenever it is  
relevant I shall consider it.

You are aware that I make my living researching and teaching about KR  
formalisms?

You are aware that many rule systems such as the datalog variants  
trade off expressiveness for various computational properties?

> so by definition
> there will be semantics (useful to certain domains) that it cannot
> express.


>> I know you're really into it, but DLP is quite the uninteresting Horn
>> Description Logic, esp. for Life Sciences.
>
> That's a bit harsh

Sometimes the truth is harsh.

> (and perhaps unsubstantiated).

Major LS ontologies tend to fall into the EL family or hornSHIQ. Must  
I substantiate this further?

More to the point, why focus on DLP as *the* fragment? There are  
several interesting horn fragments.

>   On what concrete
> basis is it uninteresting, Bijan?

You claim that your interest in it is that it couples with rule  
systems (not the expressivity of the fragment itself). There are many  
other such fragments (e.g., EL family, hornSHIQ) and coupling  
techniques (e.g., DL safe rules, hex predicates, etc.). So this is  
not a unique feature of DLP, thus does not make it at all interesting.

Second, DLP is a contorted DL that, as far as I can tell, doesn't  
capture typical LS expressive needs as witnessed by such ontologies  
as Galen and SNOMED.

So to repeat, I never said that rules couldn't be helpful or weren't  
necessary. Dissing DLP has *nothing to do with that*.

Have you seen my extended series on DL Safe rules?

>   In the ACPP task group we found
> *many* guideline and workflow semantics that are trivial to express in
> rule languages and (in many cases) impossible to express in any  
> variety
> of DL.

This address the requirement, not the particular solution. I'm sorry  
that you continue to make this conflation, but DLP is not the sole or  
the most expressive way of combining rules and DLs. There's little  
evidence that it's particularly more scalable either. After all, it  
*forces* you to restrict the expressivity of the DL side.

> When you consider the combination of 'open-ended' rules with
> rules you can extract (while preserving meaning)

This is a bit dodgy, as we've discussed.

> from expressive
> fragments of Description Logic (which includes EL++),

What does this have to do with DLP? My point above was that of the  
Horn Logic fragments of DLs, DLP, specifically, was a rather  
uninteresting one. I know some people like it (see Pascal Hitzler's  
pubs), but I don't see that *that* has been substantiated.

> this results in a
> *very* versatile KR, IMHO.

It's one of several different ways of combining logics (e.g., rules  
with DLs). Of course, if you are unfamiliar with the rest, you don't  
have a good bases of comparison.

>> (I think the worst thing about the DLP paper is that damn shield
>> diagram which is one of the top 15 misleading diagrams out there.
>> It's very seductive but really overall hurts ones understanding of
>> the relationship between the various logics.)
>
> I happen to like that diagram,

Quelle surprise :)

> care to elaborate :)

Sure, it suggests that "DLP" *IS* the horn fragment of DLs (it's  
not). It suggests that LP is "the way" to get various non first order  
features (it's not). Isn't that enough, or should I go on.

Aren't we horridly off topic now? We can take if off list or to owl-dev.

Cheers,
Bijan.

Received on Friday, 14 September 2007 13:45:34 UTC