Transcript Ontology Learning for the Semantic Web

Ontology Translation
for the Semantic Web
by
Dejing Don, Drew McDermott, and Peishen Qi
The Problem Set
Ontology Translation
 Dataset Translation*
 Ontology Extension Translation
 Querying through different Ontologies
- “ontology translation by
ontology merging and automated
reasoning.”
Assumptions


Ignoring syntatic differences
A merged ontology is formed by:
– the union of the terms and axioms of
two ontologies
– an extension of said axioms to bridge
concepts from one ontology to the other
(called bridging axioms)
Software Involved

OntoEngine
– Involved with all three of tasks in the problem
set

OntoMerge
– Semi-automated (machine assisted expert tool)
– Online nexus

PDDAML
– Web-PDDL
Test Set One

Yale vs. CMU
Yale (yale_bib ontology)
Article
Inproceedings
CMU(cmu_bib onotology)
Article
Incollection
Test Set Two
Yale (yale_bib)
Booktitle
Collection
Proceedings
Incollection
(String)
Inproceedings
(String)
CMU (cmu_bib)
Booktitle
(string)
Test Set Three

European Geneaology Ontologies
bbn_ged vs. drc_ged
Translation vs. Mapping
As defined by this paper:
Mapping: finding or mapping the
relationships between two ontologies.
Translation: given a set of facts in
one vocabulary (the source), infer the
largest possible set of consquences in
another (the target).
Ontology Mapping


still human dependant
machine learning mapping
– Saves time
– Suggestions

Ontology translators can’t use
automated ontology mappers:
– Accuracy
– Complexity
Translation Choices
Global all-encompassing One True
Theory ontology that covers all existing
ontologies. All that wish to be apart of
the Semantic Web must write
translations from their ontologies to the
One True Theory. (Ontolingua)
The other strategy is to do on-thefly ontology translation directly
from a dataset in a (source)
ontology to a dataset in another
(target) ontology, on a dataset-bydataset basis. (OntoMorph)
Approach

3 Parts:
i. Syntatic translation parse
ii. Semantic translation
iii. Syntatic translation output
PDDAML - translator
the translator for parts one and three
built to be upgradable (originally built for
DAML+OIL, can now handle OWL)
internal language Web-PDDL
Web-PDDL


Internal representation used by OntoEngine to make
semantic inferences.
Extends PDDL with XML namespaces and additional
axiom notation.
(define (domain yale_bib-ont)
(:extends (uri "http://www.w3.org/2000/01/rdfschema#":prefix rdfs))
(:types Publication – Obj
Article Book Incollection Inproceedings - Publication
Literal - @rdfs:Literal)
(:predicates (author p - Publication a - Literal).....))
Putting it all together
Translation -> from source vocabulary to target
vocbulary:
Inference & Projection




Sentence : “The publication BretonZucker96 appeared in the
Proceedings of IEEE Conf. on Computer Vision and Pattern
Recognition"
In yale_bib:
(:objects ... BretonZucker96 - InProceedings)
(:facts...(booktitle BretonZucker96 "Proceedings of CVPR'96"))
In cmu_bib:
(:objects ... BretonZucker96 - Article proc38 - Proceedings)
(facts ... (inProceedings BretonZucker96 proc38)
(booktitle proc38 "Proceedings of CVPR'96") ...)
Bridging Axioms: to relate yale_bib & cmu_bib
(forall (a - Article tl - String)
(iff (@yale_bib:booktitle a tl) (booktitle a tl)))
(forall (a - @yale_bib:Inproceedings tl - String)
(iff (booktitle a tl)
(exists (p - Proceedings)
(and (contain p a)
(@cmu_bib:inProceedings a p)
(@cmu_bib:booktitle p tl)))))
Existentialism in Code
(forall (a - @yale_bib:Inproceedings tl - String)
(iff (booktitle a tl)
(exists (p - Proceedings)
(and (contain p a)
(@cmu_bib:inProceedings a p)
(@cmu_bib:booktitle p tl)))))


p - existential modifier
Skolem term?:
Skolem finalized set theory axoims
Another Skolem theory about countably infinite subset N of
larger set M for which N satisfies all ‘first-order’ sentences
accepted by M.
Theorem Proving OntoEngine


Concerns
Necessary inferences
– Forward chaining from source to target ontologies.
– Backward chaining for queries in one ontology to datasets
in another.
– Introduction of skolem terms and term-generating
functions as explained above.
– Use of equalities to substitute existing constant terms for
skolem terms.
OntoEngine Features

Stops:(protection against theorem loops)
• Hard coded limit to complexity of terms that
OntoEngine can generate
• Deductive Engines stops when it reaches
conclusions or goals, in the case of backward
chaining, in the target ontology.
 Good type-checking system making use of
the strong typed feature of Web-PDDL
 Translation Completeness:
traded completeness for efficiency
“Anything that can be expressed in the
source ontology can be expressed in the
target ontology.”
Experiment 2
bbn_ged -> 21164 facts
(3010 individuals & 1422 families)
(@bbn_ged:name @royal92:@I1248@ "Francis_II")
(@bbn_ged:sex @royal92:@I1248@ "M")
(@bbn_ged:spouseIn @royal92:@I1248@ @royal92:@F456@)
(@bbn_ged:marriage @royal92:@F456 @royal92:event3138)
(@bbn_ged:date @royal92:event3138 "24 APR 1558")
(@bbn_ged:place @royal92:event3138 "Paris,France")
drc_ged -> 26596 facts
instead of spouseIn, has husband and wife. bbn_ged infers from marriage
and gender.
map
(forall (f - Family h - Individual m - Marriage)
(if (and (@bbn_ged:sex h "M") (@bbn_ged:spouseIn h f)
(@bbn_ged:marriage f m))
(husband f h))…)
Exp 2 Results

85555 reasoning steps to generate all 26956
facts @ 59 seconds.
http://cs-www.cs.yale.edu/homes/dvm/daml/ontology-translation.html
(@bbn_ged:name @royal92:@I1248@ "Francis_II")
(@bbn_ged:sex @royal92:@I1248@ "M")
(@bbn_ged:spouseIn @royal92:@I1248@ @royal92:@F456@)
(@bbn_ged:marriage @royal92:@F456 @royal92:event3138)
(@bbn_ged:date @royal92:event3138 "24 APR 1558")
(@bbn_ged:place @royal92:event3138 "Paris,France")
(@drc_ged:name @royal92:@I1248@ "Francis_II")
(@drc_ged:sex @royal92:@I1248@ "M")
(@drc_ged:husband @royal92:@F456 @royal92:@I1248@)
(@drc_ged:marriage @royal92:@F456 @royal92:event3138)
(@drc_ged:date @royal92:event3138 "24 APR 1558")
(@drc_ged:location @royal92:event3138 "Paris,France")
Ontology Extension
Generation


If we know the relationships of existing ontologies (A & B)
and we want to generate subonotolgies of those, what
inferences can we assume about the relationships between
our created subontologies (c & d).
Automatic Updates (propagating changes to other levels
of ontologies) in PDDAML
Example provided using Congo’s delivery system, in
which they automatically created a fairly close match to what
their experts created manually.
Problems Future Work w/
Ontology Extension
When generating ontology extensions,
PDDAML can translate the types,
predicates and only those axioms that
are sub-properties of Os1 to
corresponding properties in Os2 and
not general axioms.
Querying through
different ontologies


Example: Genealogy drc_ged Find the name of King
Henry VI’s that married him on the date given in
the previous example.
Find those ontologies that can accept partial
drc_ged ontology queries that help us answer our
question.
(:query (freevars (?k ?q - Individual ?f - Family ?m - Marriage
?n - @xsd:string ?d - @xsd:date)
(and (@drc_ged:name ?k "Henry_VI") (@drc_ged:husband ?f ?k)
(@drc_ged:wife ?f ?q) (@drc_ged:name ?q ?n)
(@drc_ged:marriage ?f ?m) (@drc_ged:date ?m ?d))))











(@drc ged:name ?k "Henry VI")
(@bbn_ged:name ?k "Henry VI")
{?k/@royal92:@I1217@}
(@drc_ged:husband?f @royal92: @I1217@)
{?f/ @royal92:@F448@}
(and (@drc_ged:wife royal92:@F448@ ?q)
(@drc_ged:marriage@royal92:@F448@?m))
(and (@bbn_ged:sex ?q "F") (@bbn_ged:spouseIn
?q @royal92:@F448@)
(@bbn_ged:marriage @royal92:@F448@ ?m))
The bindings this time are
{?q/@royal92:@I1218@}, and
{?m/@royal92:event3732}.
(and (@bbn_ged:name @royal92:@I1218@ ?n)
(@bbn_ged:date @royal92:event3732 ?d))
The ultimate result is {?n/"Margaret of Anjou"} and
{?d/"22 APR 1445"}.
Backward Querying


Incomplete tool, a lot of research
being done in this area.
Not the focus of this paper
Related Work




This work relies on inference through
bridging axioms and layering logic on top of
RDF to get free of relying on descriptions.
Prompt & Chimera- name similarity and
taxonomic relation matching
GLUE- generates only simple mapping rules
Incremental Ontology building – more
simple rules, but more complicated
algorthims.
Conclusions

Ontology translation





Translating datasets
Generating Ontology Extensions
Querying through different Ontologies
Ontology translation thought of as Ontology
merging.
If all ontologies, datasets and queries can
be expressed in the same terms, semantic
translation can be automatic.


OntoMerge – Ontology translation server.
New avenues for automating production of
bridging axioms.
Future Development
 Tools to validate consistency of generated
bridging axioms.
 Tools to help experts build axioms through
dialogues about the form of the desired
relation between ontologies in question.