Search Mailing List Archives


Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

[protege-owl] Modeling change, source, uncertainty, contradiction?

Johann Petrak johann.petrak at chello.at
Thu Aug 23 23:11:15 PDT 2007


Thank you, lots of interesting stuff and interesting pointers!

I will have to let this settle and also see where a practical
way to compromise is in my application.

Getting back to OWL I wonder if it would be possible to
have the cake and eat it: would it be possible to
introduce triples that give additional information
about properties (making a property the domain of another
property) which can also be searched,
but with the protege-owl framework ignoring
those triples for reasoning (e.g. when computing
a transitive closure)? I am aware that his would again
also provide limited functionality but in my case it
probably would be sufficient.

Thomas Russ wrote:
> On Aug 22, 2007, at 6:11 AM, Johann Petrak wrote:
> 
>> Matt Williams wrote:
>>> As a very simple approach to modelling time, you could use
>>> time-interval-valid versions of the ontology. Not pretty, but  
>>> might be
>>> enough.
>>>
>> For my purpose, I would be more interested in searchability than
>> deductability. In other words, I do not want to make deductions based
>> on time -- it would be sufficient to find properties of instances
>>  that are valid at a specific time. (So time would not be an issue
>> for classes, just for instances)
>> Naively I want an attribute "valid during" for each property/relation
>> between instances.
> 
> This is generally tough to get without some sort of temporal  
> reasoning system to keep the information separate.  One approach that  
> I worked on integrated such reasoning with a description logic system  
> (LOOM, starting with version 2.1).  The only reference to that is in  
> the release notes, though:
> 
>    <http://www.isi.edu/isd/LOOM/documentation/loom2.1-release- 
> notes.html#Time>
> 
> As Matt Williams indicated, sometimes a simple partitioning or  
> context system will be sufficient, as long as the time granularity  
> does not need to be too fine.
> 
> Although you suggest that you don't care about deduction, in fact  
> your desire to find items "valid at a specific time", which implies  
> some form of reasoning that understands notions of temporal extent  
> and persistence.  Unless you are able to specify ALL of the specific  
> times at which your fact is supposed to hold.  But that quickly gets  
> unwieldy.  That generally means you want to have some type of (at  
> least) interval reasoning capability so that you don't have to code  
> the temporal constraints directly in any query that you write.
> 
>> More generically I want several attributs: "valid during", "source"
>> etc.
>> The only thing I would need to do is to get the value of these
>> attributes out of the knowledge base and to search for triples
>> where these attributes match a specific pattern.
> 
> This is where things start to get a bit trickier, since the searching  
> interface will either need to understand the conventions for the  
> temporal or attributive relationships or else you will need to do  
> some reified version of the propositions.  But if you reify them,  
> then you will end up having to do any domain level reasoning on your  
> own.
> 
> It's a tricky problem.
> 
>>> DL ontologies will not handle conflicting information. To do that,  
>>> you
>>> need to use a defeasible formalism. There is a little bit of work on
>>> ontologies & defeasibel reasoning.
>>>
>>> I have used argumentation & ontologies; there's a paper at
>>> http://acl.icnet.uk/~mw/WillliamsHunterICTAI07.pdf which also has
>>> references to the other approaches.
>> Thank you for the reference. I am not sure I need it at such a
>> complex level though, since I do not need a reasoner to
>> come up with sets of entailed or conflicting facts.
>>
>> I think it would be sufficient for me to be able to model
>> something like
>>   "source A indicates instanceX Rel1 instanceY"
>>   "source B indicates instanceX notRel1 instanceY"
>> or put differently
>>   instanceX Rel1-withattr: from source A instanceY
>> etc.
> 
> This requires either some form of the frames-world facets, or more  
> generally, a higher order logic, since you are making assertions  
> about particular sentences rather than just making domain-level  
> assertions.
> 
> The straight-forward approach involves using higher-order logics.   
> This allows attaching source and certainty values directly to  
> sentences in the logic.  But there are generally few such logics  
> available, and they have somewhat limited reasoning.  Some examples  
> are CyC (<http://www.cycorp.com>) and PowerLoom (<http://www.isi.edu/ 
> isd/LOOM/PowerLoom/>).  [Disclaimer:  I work on the latter system]
> 
> The alternative is to reify the relationships as instances in the  
> logic.  The reification will also work, but by doing so you give up  
> any understanding by OWL of the fact that these reifications are  
> properties of particular instances.  It will also slightly complicate  
> your querying, because instead of a query something like the  
> following (using made-up pseudo-syntax):
> 
>     SELECT ?y WHERE (instanceX Rel1 ?y)
> 
> you would need to query for something like
> 
>     SELECT ?y
>    WHERE (?rel isa Relationship) and (subject ?rel instanceX) and  
> (predicate ?rel Rel1) and (object ?rel ?y)
> 
> With reification, you have a individual that represents the  
> relationship, and that allows you to make arbitrary additional  
> assertions about that relationship.  This uses a technique similar to  
> that used by OWL for representing N-ARY relationships.  (Ref: <http:// 
> www.w3.org/TR/swbp-n-aryRelations/>)
> 
> 
>> So it comes down to attatching searchable arbitrary attributes
>> to properties/relations again.
> 
> Well, they wouldn't be properties per se anymore if you used  
> reification.  Instead, the relationship would be an instance of  
> something like Relationship with its own properties such as subject,  
> predicate object.  In other words, it would look like an RDF triple,  
> about which you could express something else.
> 
> In that sense, perhaps Matt is correct in that you might want to look  
> at less constrained modeling systems such as RDF/RDFS instead of  
> OWL.  You can then make arbitrary assertions about any triple, as  
> long as you were content to do any reasoning yourself.
> 
>> Could it be that I am missing something totally elementary
>> here because it seems that should be something that is needed
>> all the time?
> 
> Well, it's a hard problem, especially with regard to producing  
> reasoners and query answering services over the more complicated  
> logics.  For one thing, having both the base relations and  
> reifications of them introduce issues of keeping the two items  
> synchronized.  The usual solution to that is to use only the reified  
> form, since that is the simplest.  But then you don't get to treat  
> those particular relationships as relationships that, say OWL, would  
> understand.  They would be instances that you would have to provide  
> all the reasoning for.
> 
>> Cheers,
>>   Johann
>>> If you want to discuss this in more detail, email me off-list.
>>>
>>> HTH,
>>>
>>> Matt
>>>
>>> Johann Petrak wrote:
>>>> I am pretty new to using Ontologies for knowledge represenation so
>>>> most of the tutorials and examples I have seen only are about
>>>> modeling some consistent set of unchanging facts.
>>>>
>>>> However, in real word situations it is often necessary to deal
>>>> with knowledge or information that has one of the following
>>>> complications:
>>>>
>>>> * a fact might change over time. More exactly, a property
>>>>   might be valid during some period of time but not another.
>>>>   Is it possible to model this in OWL ontologies and if yes,
>>>>   what are common design patterns to do it?
>>>>
>>>> * The fact that some instance has some property might be
>>>>   known based on sources A and B but might not be confirmed
>>>>   from source C. More problematic, it might contradict
>>>>   information from source D. So instead of some fact just
>>>>   "existing" we would like to model that it exists
>>>>   "according to source A" but "not confirmed by source C"
>>>>   and "not, according to source D"
>>>>   Is it possible to model this?
>>>>
>>>> * Sometimes it would be useful to attach a level of belief
>>>>   to a fact. E.g. some instance having some property might
>>>>   be likely but not certain.
>>>>
>>>> These things are probably differently hard to model, if at
>>>> all.
>>>> My biggest concern at the moment is change over time: for
>>>> most applications where I need some knowledge representation
>>>> it would be extremely important to be able to know that
>>>> e.g. some name was used during a certain time or that some
>>>> property existed during a certain period but not another.
>>>>
>>>> I would be thankful for any hints you could give me or
>>>> any papers or sources you could point out where these
>>>> issues are discussed.
>>>>
>>>> Johann
>>>>
>>>> _______________________________________________
>>>> protege-owl mailing list
>>>> protege-owl at lists.stanford.edu
>>>> https://mailman.stanford.edu/mailman/listinfo/protege-owl
>>>>
>>>> Instructions for unsubscribing: http://protege.stanford.edu/doc/ 
>>>> faq.html#01a.03
>>> _______________________________________________
>>> protege-owl mailing list
>>> protege-owl at lists.stanford.edu
>>> https://mailman.stanford.edu/mailman/listinfo/protege-owl
>>>
>>> Instructions for unsubscribing: http://protege.stanford.edu/doc/ 
>>> faq.html#01a.03
>>>
>> _______________________________________________
>> protege-owl mailing list
>> protege-owl at lists.stanford.edu
>> https://mailman.stanford.edu/mailman/listinfo/protege-owl
>>
>> Instructions for unsubscribing: http://protege.stanford.edu/doc/ 
>> faq.html#01a.03
> 
> _______________________________________________
> protege-owl mailing list
> protege-owl at lists.stanford.edu
> https://mailman.stanford.edu/mailman/listinfo/protege-owl
> 
> Instructions for unsubscribing: http://protege.stanford.edu/doc/faq.html#01a.03 
> 



More information about the protege-owl mailing list