Challenges for OWL 1.1
Peter
F. Patel-Schneider, Bell Labs Research
Copyright © 2008 Bell Labs
OWL 1.1 is an update to OWL DL and is currently being considered by W3C
for incorporation into a new W3C version of OWL. The original OWL was
the result of a number of compromises and non-compromises between
differing views of Knowledge Representation and the Semantic Web. OWL
1.1 takes differing stances on some of the issues underlying OWL,
resulting in a need for rethinking the some of the design for the new
OWL. In this talk, I will present the issues that underlie OWL and
OWL 1.1 and present my views on how the should be resolved in the new
W3C version of OWL.
What is OWL?
- The W3C Recommendation for a Web Ontology Language
- ... thus a Semantic Web language
- ... thus gets a lot of use and attention
- A KR language with two extant "species"
- OWL DL - which is a Description Logic for the Semantic
Web
- OWL Full - which is a same-syntax extension of RDF(S)
OWL Ontology Fragment
(Protégé 4 (a.k.a. Manchester) syntax)
Namespace: = <http://ex.com/owl/families#>
Ontology: <http://ex.com/owl/families>
Import: <http://ex2.com/owl/family.owl>
ObjectProperty: hasWife
Annotations: rdfs:label "moglie"@it
Domain: Man Range: Woman
Characteristics: Functional, Inverse Functional
SubPropertyOf: hasSpouse, loves
ObjectProperty: hasChild
Domain: Person Range: Person
SubPropertyOf: inverse hasAncestor
ObjectProperty: hasAncestor
Characteristics: Transitive
DataProperty: hasAge
Domain: Person Range: integer
Characteristics: Functional
Class: Person
SubClassOf: hasAge exactly 1
and hasGender exactly 1
and hasGender only {female , male}
Class: Man
EquivalentTo: Person that hasGender value male
Class: Parent
EquivalentTo: Person that hasChild min 1
Class: YoungChild
EquivalentTo: Person and not (Teenager or Adult)
Individual: Jeff
Types: hasChild exactly 2
Facts: hasWife Emily,
hasChild Ellen,
hasChild Jack,
hasAge 77,
loves Jeff
DifferentFrom: Ellen, Jack, Susan, Emily
Individual: Emily
Types: hasAge some {39 , 49}
Influences on OWL
- The Semantic Web
- OWL had to fit into the W3C vision of the Semantic Web
- Semantic Web brings notions of URIs, Semantic Web languages (RDF, RDFS)
- Description Logics
- Description Logics already proposed as web ontology
language foundation (OIL, DAML+OIL)
- Formal logics (syntax and semantics), inference, expressive power
- Users
- Various groups representing information in the Semantic Web
- FOAF (Friend of a Friend), DC (Dublin Core)
- Expectations that OWL would meet their needs (and
process existing represented data)
- Tools
- OWL needed to have implementations
- preferably efficient implementations
The Easy Part
- Semantic Web Names
- Uniform Resource Identifiers (URIs) make reasonable
names, e.g., http://whitehouse.gov/presidents#GWB
- URIs are just names
- don't require retrieval, e.g., of OWL document at
http://whitehouse.gov/presidents
- don't require believing any other content,
e.g., content of http://whitehouse.gov/presidents
- Living in the Web
- OWL ontologies are Web documents
- stored and retrieved just like other web
documents
- OWL ontologies can use (import) other ontologies by URI
- via web retrieval mechanisms
- import of http://whitehouse.gov/presidents
gets OWL document there and uses its content
The Hard Part
- The Semantic Web
- Some aspects of the Semantic Web are difficult to work with
- e.g., Resource Description Framework (RDF)
- Description Logics
- Description Logics have their own way of doing things
- Doesn't match up with RDF way well
- Users
- Users want it all!
- Efficiency, Expressive Power, Formality, Sloppiness, ...
- Tools
- Efficient tools require limiting expressive power
- What notion of "efficient" to use?
RDF vs Description Logics
- DL is FOL, RDF isn't (quite)
- RDF has a higher-order syntax (but isn't really higher-order)
- RDF can really treat classes and properties (predicates) as objects (individuals)
-
John rdf:type Person .
Person rdf:type rdfs:Class .
treats Person as both a class and an object
(individual)
- DL divides universe into objects, e.g.,
John, and data
values, e.g., 5 and "string";
RDF objects include everything
- RDF has only a few syntactic constructs, DLs have more
- RDF has only names (URIs), a kind of variable (blank nodes),
data (e.g., integers), and triples (binary non-ground facts)
- DLs have, e.g., descriptions of various kinds
- OWL in RDF:
Jeff rdf:type _:t .
_:t rdf:type owl:Restriction .
_:t owl:onProperty hasChild .
_:t owl:cardinality "2"^^xsd:integer .
- each triple is a fact
- can't forbid some malformed syntax
- making OWL work right requires presence of these syntactic
facts in all interpretations
Users vs Tools
- User desires
- Users want to be able to represent anything
- Users want tools to work fast
- Users are often sloppy, e.g., Dublin Core author is
either a person or a string
- Can't satisfy all user desires, something has to give
- Need to come up with a reasonable, implementable language
- Formal, because otherwise we don't know how to proceed
- Limit expressive power to achieve implementability
- Description Logics - SHOIN(D)
- Formal and expressive "ontology" language
- Decidable (but bad complexity - NEXPTIME)
- Nonetheless can be implemented using optimised tableaux
provers
OWL: A Grand Compromise
- Two extant species of OWL
- OWL DL - a Description Logic in Semantic Web clothing
- Based on SHOIN(D)
- Decidable, implementable
- OWL Full - an RDF vocabulary with added semantics
- Upward compatible with RDF and RDFS
- Undecidable
- OWL DL can't
- treat individuals or properties as classes in
the same way that RDF does
- have number restrictions on transitive roles
- have properties that relate objects to both
objects and data values
- use "variables" in arbitrary ways
- have malformed syntax
- Relationship between OWL DL and OWL Full
- OWL Full is written in RDF; any RDF document is an
OWL Full ontology
- Most DL ontologies can translated into RDF and thus
into OWL Full
- OWL DL inference is sound with respect to
OWL Full inference on OWL DL
ontologies that can be translated into RDF (and is
the same for nearly all these ontologies)
OWL: Ugly Underpinnings
- OWL Full semantics is very complex
- Has to take into account that OWL syntax is encoded
as facts
- Could easily have undiscovered bugs
- Fragile, can't be extended too far (liar's paradox)
- Translation between OWL DL syntax and RDF syntax is complex
- Parsing RDF into OWL DL is hard and slow
- Not all OWL DL ontologies can be translated into RDF
- Those that use a form of metamodelling called punning
- Some constructions don't fit into Description Logics well
- e.g., annotations, deprecation
- These constructions should have no meaning, but have to be facts in RDF
- They end up complicating OWL DL semantics
- They don't get fully implemented
- OWL DL is missing some expressive power
- Particularly qualified number restrictions (e.g., people
with at least
three children who are doctors)
Time Marches On
- OWL DL implementations built
- Description Logic systems based on optimised tableaux provers
- Sound and complete for OWL DL (or almost all of it)
- OWL Full implementations built
- Mostly rule-based, implement only a small fraction of OWL
Full
- Definitely not complete
- Users increase
- Use of OWL in bioontologies, etc.
- Want more expressive power (qualified number
restrictions, role chain inclusions)
- Description Logic theory advances
- Reasoning algorithms for qualified number restrictions (SHOIQ)
- Reasoning algorithms for certain kinds of role chain
inclusions (SROIQ)
- Let's update OWL to take into account (some of) this!
OWL 1.1
- Result of OWLED workshops bringing together users, developers, and
theoreticians
- An update to OWL DL in response to requests by users, but changes
limited to those with both theoretical underpinning and support from implementations
- Added expressive power (SROIQ(D))
- (ir)reflexive and (a)symmetric properties
- disjoint properties
- qualified number restrictions
- limited role chain inclusions
- self-restrictions
- enhanced datatypes
- Nicer syntax for some commonly-used constructs
- Annotations - no formal meaning, can be attached to more things
- Remove a few minor things that weren't used
- New formal syntax - closer to Description Logics syntax
- New translation to RDF - easier to parse and allows metamodelling
- Still a Description Logic - formal semantics, implementations, first-order
- Submitted to W3C
- Start for new working group to update OWL
OWL 1.1 Ontology Fragment
ObjectProperty: hasWife
Characteristics: Irreflexive, AntiSymmetric
ObjectProperty: hasDaughter
Domain: Parent
Range: Person that hasGender value female
SubPropertyOf: hasChild
SuperPropertyOfChain: f:hasSpouse o f:hasDaughter
ObjectProperty: hasSon
Domain: Parent
Range: Person that hasGender value male
SubPropertyOf: hasChild
DisjointWith: hasDaughter
Class: Adult
Annotations: rdfs:label "adulto"@it
EquivalentTo:
Annotations: rdfs:comment "This is a rather crude definition of adult"
Person that hasAge some integer[>= 21]
Class: Teenager
EquivalentTo: Person that hasAge some integer[>= 13 , < 20]
Class: Narcissist
EquivalentTo: Person that loves Self
Individual: Jeff
Types: hasChild exactly 2 Teenager
Individual: Ellen
Types: hasAge some integer[>= 15 , <= 19]
Individual: Jack
Facts: not hasAge "53"^^integer
W3C OWL WG
Working
group home page,
http://www.w3.org/2007/OWL/wiki/OWL_Working_Group
- Chartered to update W3C OWL specification
- Start with OWL 1.1
- Produce new documents defining and describing OWL
Should be easy!
OWL WG Activities
- Fixes to OWL 1.1
- Eliminate bugs
- Upgrade RDF syntax (Controversial!)
- Better description of importing
(Controversial!)
- Fend off requestions for even more expressive power
- They make OWL 1.1 undecidable or very hard to implement
- Non-formal documents (OWL Overview, User Guides, Language Description)
- Old documents very tied to RDF syntax
- Produce new documents or update old ones?
(Lots of controversy!)
OWL WG Activities (continued)
- New semantics for OWL Full (Hard!)
- Incorporate new expressive power
- Handle annotations
- What is the relationship between OWL 1.1 and OWL Full
- All(!) OWL 1.1 ontologies can be written in RDF
- OWL 1.1 inference is sound with respect to
OWL Full inference on OWL 1.1
ontologies (and is the same for almost all OWL 1.1
ontologies,
but fewer than before (Controversial!))
and annotations can't be inferred
(Controversial!)
- Fragments of OWL
- OWL Lite was supposed to be a lightweight fragment of OWL DL
- Wasn't lightweight, wasn't used
- Should it be replaced (with formal, tractable subset,
e.g., EL++, or with rule-based subset, e.g., Oracle
implementation)?
(Controversial!)
Sources of Controversy
Several views of OWL and OWL 1.1:
- OWL is an ontology language very close to Description Logics.
- OWL Full is just something required to get W3C
approval.
- Semantic Web orientation is just something required
to get W3C approval.
- OWL is a powerful extension of RDFS.
- OWL DL is just one of many fragments of OWL.
- The real Web Ontology Language is RDFS plus functional
properties (and maybe a bit of other stuff).
- OWL DL is a theoretical language and not ready to be
used in real applications, and OWL Full just shows a
potential way forward for when
tools get faster.
- Every feature of OWL must be preserved, OWL 1.1 can only add
a little bit vs OWL 1.1 should fix the broken parts of OWL.
How to Proceed
- Remove OWL DL and OWL 1.1 from the official Semantic Web,
terminating the uneasy relationship between OWL DL and OWL Full.
- W3C Web Ontology Language becomes OWL Full; or
- W3C Web Ontology Language becomes RDFS++.
- Go back to the previous syntaxes and relationship to OWL Full.
- Requires eliminating new annotations and not translating metamodelling,
at least.
- Can add new restrictions, new characteristics of
properties, and role chain inclusions.
- Relegate OWL Full to the ashcan of technology, where it
belongs.
- Use solution proposed (implicitly) in OWL 1.1.
- All OWL 1.1 constructs remain.
- Annotations have no meaning in OWL 1.1.
- New syntax for OWL 1.1.
- Deprecate deprecation and ontology properties.
- Better translation into RDF (and OWL Full).
- Similar (but not exactly the same) relationship to OWL Full.
Example OWL Ontology (Complete)
Namespace: =<http://ex.com/owl/families#>
Namespace: g = <http://ex2.com/owl/family.owl#>
Namespace: families = <http://ex.com/owl/families#>
Namespace: rdfs = <http://www.w3.org/2000/01/rdf-schema#>
Namespace: rdf = <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
Ontology: <http://ex.com/owl/families>
Import: <http://ex2.com/owl/family.owl>
ObjectProperty: hasWife
Domain: Person, Man
Range: Person, Woman
Characteristics: Functional, Inverse Functional, Irreflexive, AntiSymmetric
SubPropertyOf: hasSpouse, loves
ObjectProperty: hasHusband
Inverses: hasWife
ObjectProperty: hasSon
Domain: Person, Parent
Range: Person, Person that hasGender value male
SubPropertyOf: hasChild
DisjointWith: hasDaughter
SubObjectProperty: f:hasSpouse o f:hasSon f:hasSon **
ObjectProperty: hasDaughter
Domain: Person, Parent
Range: Person, Person that hasGender value female
SubPropertyOf: hasChild
SubObjectProperty: f:hasSpouse o f:hasDaughter f:hasDaughter **
ObjectProperty: hasGender
ObjectProperty: hasChild
Domain: Person Range: Person
SubPropertyOf: **INVERSEOF** hasAncestor
ObjectProperty: hasAncestor
Characteristics: Transitive, Irreflexive
ObjectProperty: hasSpouse
Characteristics: Symmetric, Irreflexive
ObjectProperty: loves
Domain: Person
DataProperty: hasAge
Domain: Person Range: integer
Characteristics: Functional
Class: Person
SubClassOf: hasAge exactly 1
and hasGender exactly 1
and hasGender only ({female , male})
Class: Man
EquivalentTo: Person that hasGender value male
Class: Woman
EquivalentTo: Person that hasGender value female
Class: Parent
EquivalentTo: Person that hasChild min 1
Class: Teenager
EquivalentTo: Person that hasAge some integer[>= 13 , < 20]
Class: Adult
EquivalentTo: Person that hasAge some integer[>= 21]
Class: Child
EquivalentTo: Person and not (hasAge some integer[>= 21])
Class: YoungChild
EquivalentTo: Person and not (Teenager or Adult)
Class: Marriage
EquivalentTo: CivilMarriage or ReligiousMarriage
Class: ReligiousMarriage
DisjointWith: CivilMarriage
Class: CivilMarriage
Class: Narcissist
EquivalentTo: Person that loves Self
Individual: male
DifferentFrom: female
Individual: female
Individual: John
Types: Person
Facts: hasWife Mary,
hasSon Bill,
hasDaughter Susan,
hasAge 33,
hasGender male
SameAs: Jack
DifferentFrom: Mary, Susan, Bill
Individual: Mary
Facts: hasSon Bill,
hasDaughter Susan,
hasAge 31,
hasGender female
Individual: Bill
Types: not (Narcissist)
Facts: hasAge 13,
hasGender male
Individual: Susan
Facts: hasAge 8,
hasGender female
Individual: Jeff
Types: hasChild exactly 2 Teenager
Facts: hasWife Emily,
hasChild Ellen,
hasChild Jack,
hasAge 77,
loves Jeff
DifferentFrom: Ellen, Jack, Susan, Emily
Individual: Emily
Types: hasAge some {39 , 49}
Individual: Ellen
Types: hasAge some integer[>= 15 , <= 21]
Individual: Jack
Facts: not hasAge "53"^^integer
SameIndividual: f:male g:masculine
SameIndividual: f:female g:feminine
EquivalentClasses: f:Adult g:Grownup
EquivalentObjectProperties: f:hasChild g:child
EquivalentDataProperties: f:hasAge g:age