Part of the Workshop Session Notes


Rules Workshop -- Session 6 -- Candidate Technologies, Part 2

13 May 2005




<tlrDC> Scribe: josdebruijn

Common Logic

<tlrDC> http://www.w3.org/2004/12/rules-ws/paper/103

Speaker: Pat Hayes

PatH: Not sure if this is a candidate technology ...
... Common Logic is nearly fixed ...
... Common Logic (CL) is like HiLog-like with higher-order features ...
... CL covers FOL-based logics like DL ...
... origins in KIF, now called CLIF
... nothing about machinery (no back/forward chaining)
... defined in terms of an abstract syntax
... modality in syntax, but not in semantics
... no distinction between relation/function/etc... names
... CL seems higher-order, but is first order; talk to Pat offline
... conformance also defined for Horn-based reasoners
... CLIF is KIF-based concrete syntax for CL
... CLIF allows tail-recursion, like KIF
... CLIF treats XML Schema datatypes like RDF
... CL is being submitted to ISO
... XML syntax will allow including other dialects

<DanC_lap> (hmm... I wonder about using MathML as an XML syntax for CL)

MichaelK: why aren't propositions reified? HiLog does this.

PatH: Was not considered at first, but it seems a requirement now from the first community we encounter.

<DanC_lap> (the guy leading the XML syntax for CL is Murray Altheim, a long-time XML guy. I wonder what he would think of MathML)

DieterF: Great piece of work. What is the relation with the Semantic Web? RDF; rules, nonmon...

PatH: SemWeb languages have influenced design of the language. There is an XML syntax. Some SemWeb languages, such as SWRL, OWL DL, can be embedded. Nonmon will be harder.

TRIPLE - RDF Rule Language with Context and Use Cases

Stefan Decker: TRIPLE - RDF Rule Language with Context and Use Cases

<tlrDC> http://www.w3.org/2004/12/rules-ws/paper/98

StefanD: TRIPLE is an RDF rule language and has been experimented with

<DanC_lap> title slide definitely goes beyond ASCII... wonder if it fits in 8859-1 or if it needs unicode.

StefanD: "Rule language for interoperability" instead of "Rule language interoperability"

<DanC_lap> claim: it's faster to write rules than to write code

<DanC_lap> hmm

StefanD: Writing rules is faster than writing code to process RDF
... Multiple modeling languages, e.g. OWL, UML, XML Schema, etc.
... OWL is An ontology representation language, not THE ontology language
... Contexts; has already been identified is important in the use cases

<sandro> Stefan: does it come from whitehouse.gov or whitehouse.com....? I'm not saying which to trust.

StefanD: A triple is inside a model (context), an intersection of models or skolem function (creates new context)
... "object invention" is used for creating new resources, based on existing resources
... parameterized context for capturing semantics of different languages
... semantics of the language is partially captured using rules

<sandro> very intreresting way to talk about entailment in query

<bijan> queue note: Enrico wants me to proxy a question from him to Stefan

StefanD: examples: RDFS and UML

<DanC_lap> hmm... in N3, sorta like (<cars> <rdf-schema-rules).log:conjunction.log:conclusion

StefanD: Triple limited to Horn; interoperability with OWL using DLP or external reasoner

<DanC_lap> (hmm... how would the external reasoner thing mix with NAF? I'm puzzled)

StefanD: exists a mapping to Horn logic + NAF
... first implementation on top of XSB; URI in slides

<DanC_lap> (horn logic plus negation? I thought horn included negation. er... maybe only negated literals. Sigh. I can't keep all the marbles in my head)

StefanD: important aspects of Triple: contextualization and use of different semantics

Bijan: question from Enrico: UML has same complexity as OWL DL; how can you capture it?
... Enrico retracts question

StefanD: we captured what we can; there was at the time no official semantics for UML

DieterF: is there a semantics?

<efranconi> yes there is one!

StefanD: especially for Peter, there is a semantics, see slide

RDF/RuleML Interoperability

HaroldB: RDFizing RuleML
... before OO RuleML, RDF triples were represented as binary predicates
... OO RuleML allows compressing triples with same subject, like N3 and RDF/XML

<DanC_lap> (er... is he assuming we all have RuleML syntax in our head?)

HaroldB: RDF predicate is first child of RuleML slot

<DanC_lap> er... odd example... the director of W3C is not a web page; it's a person

<DanC_lap> er... and http://www.w3.org/ is W3C's homepage, not W3C

DanC: do you need Skolem functions?

HaroldB: no, we only need constants (nullary functions)
... for parameterized modules, like Triple, you need n-ary skolem functions
... different groups (also Gerd Wagner) came to the same semantics for different kinds of negation

<DanC_lap> (hmm... OWL's negation is pretty classical, no? what's limited about it? odd to say OWL has nothing for handling negation. it has nothing novel, but it has the ability to express/detect inconsistencies)

<DanC_lap> (note to self... got lost around slide 14 or 15)

<DanC_lap> "currently not registered"... registered in what? there's no reference to the list of rental contracts. I don't see how this allows working with partial knowledge.

<em> i'm not sure it does... its unclear to me this work considers this a requirement

PatH: what happens if wrong negation is used?

HaroldB: we did not distinguish when to allow which kind of negation. We have had some discussion about local closed-world assumption where you define which predicates you can negate using naf.
... has also been implemented in Prolog

TimBL: can you have a list which defines the extension of a predicate?

HaroldB: We do not tackle this problem
... any predicate corresponds with a list of ground facts

<DanC_lap> hmm... isAssignedtoRentalContract... is he using something like log:definitiveDocument ?

<timsHomePage> "You know, errors happen"

PatH: is the error syntactic or semantic?

HaroldB: semantic; we don#t say anything about relation names

BenG: We don't want to show two forms of negation to users
... rules can always we rewritten to one form of negation
... use negation under the covers

<DanC_lap> (I'm curious what the ineroperability/deployment expectations are with RuleML... they've embraced everything... where does the interoperability come?)

Sandro: time for closing discussion

Peter: your task is to provide sufficient information for a formal syntax and semantics or a way to produce an interoperable implementation

PatH: we have syntax and semantics

StefanD: we have an implementation
... agrees; there is a semantics

HaroldB: is family of incomparable languages
... paper is about RDF part; semantics of RDF applies to this part

Peter: what are you proposing as the candidate?

HaroldB: yesterday we already proposed datalog+naf

<DanC_lap> (hmm... datalog + NAF... so datalog doesn't have NAF? sigh.)

<efranconi> (I would say so...)

MichaelU: do you care about each other's systems?

<DanC_lap> (so "datalog" is ambiguous? there isn't a canonical paper that defines it?)

PatH: interoperability with RDF, OWL, UML

StefanD: Triple is working tool, not knowledge representation

<efranconi> (terminology standard is as follows: datalog has no negation, datalog^{neg} has negation, possibly stratified)

StefanD: you want to use intuitive semantics

<DanC_lap> ("datalog... imposes certain stratification restrictions on the use of negation and recursion per http://en.wikipedia.org/wiki/Datalog )

Bijan: sounds nice story, but does not seem a starter for a standardization effort; we need a specification which is understood by many people.

StefanD: misunderstanding. Horn logic provides semantic foundation; there is a model theory. Data which is processed might be messy, but language is properly defined.

Bijan: is there a document describing the model theory?

StefanD: Yes

AnthonyF: people present ever richer KR schemes. Is this the right way to go for standardization. An alternative is to standardize on a minimum subset on which we can all agree.

PatH: if you standardize on a language, you automatically start adding features. You need to keep many people happy by trying to keep it small and including features as long as they don't break the underlying model.

<DanC_lap> AnthonyF's question is a darned good one. I prefer semantic web architecture that tends toward very expressive formats, but as a WG chair, I like small, simple designs that have testable conformance.

?: choice made between which way to go. Do you accept sparql/n3 syntax or do you want a traditional syntax?

PatH: universal XML for different syntaxes. As long as we're not too ambitious we can stay with this syntax.

StefanD: information you create becomes part of the SemWeb and needs to be available in RDF. Surface syntax is more readable, but there is no contradiction.

HaroldB: agrees. XML has benefit of XSLT, which you can translate to different syntaxes
... about minimalism datalog is agreed for rule language; naf is not completely agreed upon

StanD: extensibility. During making MathML we received many feature requests; extensibility was very important

StefanD: you talk about extension of data representation mechanism. RDF has been designed for extensibility.
... for rules: you have basic logic and you need extensibility in area of built-ins; this is necessary

PatH: we looked at MathML. Problem with extensibility: framework for extension of model theory. I have no idea how to do that; would probably not be a good idea to ga beyond first-order

StanD: semantics identified by URI; allows systems to identify which semantics is used

PatH: this would render the logic useless

<DanC_lap> efranconi, I can't understand you. the panelists are struggling, and I'm not sure how much they got

Bijan: Enrico: Datalog as rule language. As soon as we add simple ontology language, the language becomes very expressive. There is some about this in the position statement.

HaroldB: I was talking about pure rule language. for combination with ontology language we can use DL-safe.

DanC: datalog+naf. Does datalog include naf?

HaroldB: datalog does not include naf officially

<efranconi> (Implemented disjunctive datalog systems such as DLV have both classical negation and NAF)

MichaelK: when it was introduced, it did not include naf

PatH: people should be precise

DanC: about the CLIF syntax: ... and []

PatH: [] is blank propositional variable
... ... are sequence variables
... is not first-order, because of sequence variables

Summary of Action Items

[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.122 (CVS log)
$Date: 2005/05/13 19:35:55 $