fabian m. suchanek ontological reasoning for natural language understanding1 ontological reasoning...

24
Ontological Reasoning for Natural Language Understanding 1 Fabian M. Suchanek Ontological Reasoning for Natural Language Understanding Fabian M. Suchanek Supervisors: Prof. Dr-Ing. Gerhard Weikum Dr. habil. Peter Baumgartner Colloquium

Upload: andra-fields

Post on 30-Dec-2015

221 views

Category:

Documents


0 download

TRANSCRIPT

Ontological Reasoning for Natural Language Understanding 1Fabian M. Suchanek

Ontological Reasoning

for Natural Language Understanding

Fabian M. Suchanek

Supervisors: Prof. Dr-Ing. Gerhard Weikum

Dr. habil. Peter Baumgartner

Colloquium

Ontological Reasoning for Natural Language Understanding 2Fabian M. Suchanek

Overview

1. Introduction to the Problem

2. Our Approach

1. SUMO-Ontology

2. Text Analysis

3. KRHyper

3. Conclusions and Outlook

Ontological Reasoning for Natural Language Understanding 3Fabian M. Suchanek

Text Entailment

Hypothesis: „Elvis is still alive!“

Text:

...Yes, Elvis is alive...

...Elvis is not alive...

...Elvis comes to Saarbrücken on January 1st...

...Elvis is living incognito in a small Russian village...

Ontological Reasoning for Natural Language Understanding 4Fabian M. Suchanek

Plan of Attack

KRHyper-Rules

subclass(funeral, socialinteraction).subclass(literacyattribute, traitattribute).subclass(mixedlanguage, spokenhumanlanguage).subclass(pidginlanguage, spokenhumanlanguage).subclass(creolelanguage, spokenhumanlanguage).

subclass(human,animal).

subclass(deafsignlanguage, manualhumanlanguage).).subclass(radio, electricdevice).).subclass(contractcarrier, transportationcompany).subclass(commoncarrier, transportationcompany).subclass(terminalbuilding, building).subclass(trainstation, transitterminal).subclass(representationalartwork, artwork).subclass(artwork, artifact).subclass(singlefamilyresidence, permanentresidence).subclass(hotel, commercialagent).subclass(hotel, temporaryresidence).subclass(hotel, residentialbuilding).subclass(residentialbuilding, residence).subclass(residentialbuilding, building).subclass(temporaryresidence, residence).subclass(permanentresidence, residence).subclass(residence, stationaryartifact).subclass(room, stationaryartifact).subclass(building, stationaryartifact).

comesTo(elvis,saarbruecken).subclass(watercraft, vehicle).subclass(train, railvehicle).subclass(wagon, landvehicle).subclass(railvehicle, landvehicle).subclass(motorcycle, roadvehicle). subclass(artificialsatellite, satellite).subclass(communicationdevice, engineeringcomponent).subclass(contractcarrier, transportationcompany).

SUMO-Ontology

Text

Ontological Reasoning for Natural Language Understanding 5Fabian M. Suchanek

SUMO-Ontology – Concept Hierarchy

Entity

PhysicalAbstract

ConstrictorSnake

Anaconda

Human

Ontological Reasoning for Natural Language Understanding 6Fabian M. Suchanek

instance(P, intentionalProcess) agent(P, A)=> holdsDuring(when(P), attribute(A, awake))

SUMO-Ontology – Axioms

Ontological Reasoning for Natural Language Understanding 7Fabian M. Suchanek

instance(P, intentionalProcess) agent(P, A)=> holdsDuring(when(P), attribute(A, awake))

SUMO-Ontology – Axioms

Functions

Ontological Reasoning for Natural Language Understanding 8Fabian M. Suchanek

instance(P, intentionalProcess) agent(P, A)=> holdsDuring(when(P), attribute(A, awake))

SUMO-Ontology – Axioms

Functions Nested Formulae

O

Ontological Reasoning for Natural Language Understanding 12Fabian M. Suchanek

SUMO-Ontology – Nested Formulae

thinks(bob, alive(priscilla) alive(elvis)).

alive(elvis) alive(priscilla)

<=> alive(priscilla) alive(elvis)

thinks(bob, alive(elvis) alive(priscilla) ).

Ontological Reasoning for Natural Language Understanding 13Fabian M. Suchanek

SUMO-Ontology – Nested Formulae

thinks(bob, false beautiful(priscilla)).

alive(elvis) ¬alive(elvis)

<=> false

<=> false beautiful(priscilla)

?

thinks(bob, alive(elvis) ¬alive(elvis) ).

O

Ontological Reasoning for Natural Language Understanding 15Fabian M. Suchanek

Propositional Equivalence

p(t1,… q(t'1,…t'n'),… tn)

~~~~~> x p(t1,…,x,… tn) q(t'1,…t'n',x)

p(t1,… not(q(t'1,…t'n')),… tn)

~~~~~> x p(t1,…,x,… tn) q(t'1,…t'n',neg(x))

says(bob,a)

says(bob,not(b))a b

Ontological Reasoning for Natural Language Understanding 16Fabian M. Suchanek

Propositional Equivalence, Example

thinks(bob, alive(elvis) ¬alive(elvis) )

=> thinks(bob, beautiful(priscilla) )

thinks(bob, wife(elvis,priscilla)) wife(X,Y) <=> husband(Y,X)

=> thinks(bob, husband(priscilla,elvis))

knows(bob, alive(elvis)) knows(X,Y) => believes(X,Y)

=> believes(bob, alive(elvis))

Ontological Reasoning for Natural Language Understanding 17Fabian M. Suchanek

Plan of Attack

KRHyper-Rules

subclass(funeral, socialinteraction).subclass(literacyattribute, traitattribute).subclass(mixedlanguage, spokenhumanlanguage).subclass(pidginlanguage, spokenhumanlanguage).subclass(creolelanguage, spokenhumanlanguage).

subclass(human, animal).subclass(deafsignlanguage, manualhumanlanguage).).subclass(radio, electricdevice).).subclass(contractcarrier, transportationcompany).subclass(commoncarrier, transportationcompany).subclass(terminalbuilding, building).subclass(trainstation, transitterminal).subclass(representationalartwork, artwork).subclass(artwork, artifact).subclass(singlefamilyresidence, permanentresidence).subclass(hotel, commercialagent).subclass(hotel, temporaryresidence).subclass(hotel, residentialbuilding).subclass(residentialbuilding, residence).subclass(residentialbuilding, building).subclass(temporaryresidence, residence).subclass(permanentresidence, residence).subclass(residence, stationaryartifact).subclass(room, stationaryartifact).subclass(building, stationaryartifact).

comesTo(elvis, Saarbruecken).subclass(watercraft, vehicle).subclass(train, railvehicle).subclass(wagon, landvehicle).subclass(railvehicle, landvehicle).subclass(motorcycle, roadvehicle). subclass(artificialsatellite, satellite).subclass(communicationdevice, engineeringcomponent).subclass(contractcarrier, transportationcompany).

SUMO-Ontology

Text

Ontological Reasoning for Natural Language Understanding 18Fabian M. Suchanek

Text Analysis

Elvis comes to Saarbrücken

Noun Verb Prep Noun

Comp

VP

S

O

Ontological Reasoning for Natural Language Understanding 20Fabian M. Suchanek

Mapping a Sentence to SUMO

Elvis comes to Saarbrücken

?

Ontological Reasoning for Natural Language Understanding 21Fabian M. Suchanek

Mapping a Sentence to SUMO

Elvis comes to Saarbrücken

BodyMotion

Ontological Reasoning for Natural Language Understanding 22Fabian M. Suchanek

Mapping a Sentence to SUMO

Elvis comes to Saarbrücken

BodyMotion

??

AgentDestination

O

Ontological Reasoning for Natural Language Understanding 25Fabian M. Suchanek

Mapping a Sentence to SUMO

Elvis comes to Saarbrücken

BodyMotion

Agent Destination

instance(b,bodymotion).agent(b,elvis).destination(b,saarbruecken).

Ontological Reasoning for Natural Language Understanding 26Fabian M. Suchanek

Plan of Attack

KRHyper-Rules

subclass(funeral, socialinteraction).subclass(literacyattribute, traitattribute).subclass(mixedlanguage, spokenhumanlanguage).subclass(pidginlanguage, spokenhumanlanguage).subclass(creolelanguage, spokenhumanlanguage).

subclass(human, animal).subclass(deafsignlanguage, manualhumanlanguage).).subclass(radio, electricdevice).).subclass(contractcarrier, transportationcompany).subclass(commoncarrier, transportationcompany).subclass(terminalbuilding, building).subclass(trainstation, transitterminal).subclass(representationalartwork, artwork).subclass(artwork, artifact).subclass(singlefamilyresidence, permanentresidence).subclass(hotel, commercialagent).subclass(hotel, temporaryresidence).subclass(hotel, residentialbuilding).subclass(residentialbuilding, residence).subclass(residentialbuilding, building).subclass(temporaryresidence, residence).subclass(permanentresidence, residence).subclass(residence, stationaryartifact).subclass(room, stationaryartifact).subclass(building, stationaryartifact).

comesTo(elvis, Saarbruecken).subclass(watercraft, vehicle).subclass(train, railvehicle).subclass(wagon, landvehicle).subclass(railvehicle, landvehicle).subclass(motorcycle, roadvehicle). subclass(artificialsatellite, satellite).subclass(communicationdevice, engineeringcomponent).subclass(contractcarrier, transportationcompany).

SUMO-Ontology

Text

Ontological Reasoning for Natural Language Understanding 27Fabian M. Suchanek

KRHyper

First-Order-Reasoning System ر

Developed at the University of Koblenz/Germany ر

by P. Baumgartner and C. Wernhard

Given some facts and rules, calculates a set of ر

entailed facts

Facts+Rules

entailed

O

Ontological Reasoning for Natural Language Understanding 30Fabian M. Suchanek

attribute(A, alive) ← agent(P,A) instance(P,process).

subclass(bodyMotion, process).

instance(b,bodyMotion).agent(b,elvis).

Application

SUMO-Ontology

Text

attribute(elvis,alive).

instance(b,process).

Hypothesis false ← attribute(elvis,alive).

Ontological Reasoning for Natural Language Understanding 31Fabian M. Suchanek

Results

It is possible to enhance question answering ر by a kind of common sense reasoning

There is still a long way to go to practical applicability ر

O

Ontological Reasoning for Natural Language Understanding 33Fabian M. Suchanek

Outlook

Idea: Learn the Ontology from the Web

Web(Wikipedia)

Free Text+ HTML+ Ontological Knowledge

Ontological Reasoning for Natural Language Understanding 34Fabian M. Suchanek

Conclusion

References:

KRHyper: http://www.uni-koblenz.de/~wernhard/krhyper

SUMO: http://www.ontologyportal.org

Comp. Linguistics: http://www.coli.uni-saarland.de

Thesis: http://www.mpii.mpg.de/~suchanek/personal

Elvis is alive, but there‘s still much work to do.