Jerome.Euzenat@inrialpes.fr

Jérôme Euzenat

655 avenue de l’Europe, 38330 Montbonnot Saint-Martin, France

Enabling technologies for semantic interoperability

semanticThe web

The semantic webs

Query

Conceptual graphs

Description logics

Object-based KR language

Escrire (INRIA)(with Orpailleur and Acacia projects)

Reasons for heterogeneity

• Language suitability (expressiveness…)

• Preferences

• Legacy knowledge

• Techno-diversity is good!

• Are the languages really mature?

Some reasons are arguable but some are good

Knowledge (ontology) representation language

OntologyLanguage

reasonner

semantics

Description

Meaning preservation

T()

KRL1

KRL

, T()|=KRL T() |=KRL1

, |=KRL1 T() |=KRL T()

T

Solutions

• Pivot languages

• Modular family of languages

• Transformation and properties

• Pattern-based language construction

Pivot language

• In the line of the Web: the HTML of knowledge

• Which one?

• Not problem free: loss of knowledge, confusion… (no round trip).

The modular semantic way

L1

r1

s1

Ln

rn

sn

L=ƒl(L1,…Ln)

r=ƒr(r1,…rn)

s =ƒs(s1,…sn)

This have been done in DLML for syntax and semantics andmainly in description logics for reasonners

Elementary operators

DLML: features

Modular conception of DL descriptions:

• Description (DTD,DSD) of 40+ constructors

• Automatic generation of 25+ logics by assembling constructors

• Compatible DTD for all logics.

Note: inference rules could be described the same way.

DLML: transformations

From logic specification to DTD and DSD

From logic to LaTeX

From ALC to ALUE

From ALC to AL

From FaCT to ALUNI and back

From Syllogistic to AL and back

The semantic-driven transformation

L1 Lp

L

r1

s1

rp

sp

Tp = ƒt(Lp,sp/L,s)

r

s

T1 = ƒt(L1,s1/L,s)

L1 Lp

L

r1

s1

rp

sp

Tp = ƒt(Lp,sp/L,s)

r

s

T1 = ƒt(L1,s1/L,s)

The semantic-driven transformation

ont1 ontp

Ont =T1(ont1)+Tp(ontp)

This is easily doable when LiL but can become really hard

DSD: principles

Coded by using MathML and Xpath conventions

Require the interpretation of the language.

I(CLASS) D I(@specialises )

i D; I( f@name)(i) I( f.DESCRIPTOR )

f FIELD

DSD: example…

<dsd:interpretation match="CLASS"/>

<subset/>

<apply><intersect/>

<dsd:domain name="D"/>

<dsd:apply-interpretation select="@superclass"/>

<dsd:apply-interpretation select="FIELD"/>

</apply>

</dsd:interpretation>

<dsd:interpretation match="FIELD">

<equal/>

<apply><intersect/>

<dsd:apply-interpretation select="*"/>

</apply>

</dsd:interpretation>

…

DSD: purpose

• pretty-printing semantics (XSLT);

• documenting a format;

• computing base interpretation (XSLT);

• proof-checking transformations;

• assisted proof (properties of transformations);

• automatic proof (p.o.t.);

• …

The ontology-pattern way(with Heiner Stuckenschmidt, TZI-Bremen)

L1

r1

s1

Ln

rn

sn

L=ƒl(L1,…Ln)

r =ƒr(r1,…rn)

s =ƒs(s1,…sn)

ƒr(.)

ƒs(.)ƒl(.)

DLML as a pattern?

ƒs(.)=the interpretation of a term is still the intersection of the interpretation of the constructors

ƒl(.)=the entry for roles must accept the constructor with its own syntax

Adding a role constructor

This is still part of the easy things to do…because DL have been neatly designed

The pattern-based transformation

ont1 ontp

’

’-1

ƒ(’-1 )

Conclusion

The interoperability issue is a serious problem at the web-scale.

There are many useful, doable, reasonable techniques (e.g. DLML);

The general case is difficult

Contact…

http://www.inrialpes.fr/exmo/

Jerome.Euzenat@inrialpes.fr