suggested upper merged ontology - hypotheses.org · –concept extraction (linguistic...

1(c) 2012 Adam Pease – [email protected]

Formal Ontology andthe Suggested Upper Merged

Ontology (SUMO)

Adam Pease, Articulate [email protected]

http://www.ontologyportal.org/


NLP Application of Formal Ontology

• Sentiment Analysis

• Information Extraction

• Question Answering/Textual Entailment

• Word Sense Disambiguation

• Document classification and clustering


Technology Map: Semantics

• Structure databases– Ensure that all fields are clear and consistent

– Database consistent with SUMO

• Expose features– Finding correlations only possible when features that correlate

are exposed• Abstraction based on superclasses etc

• Capture knowledge– Express in a computable language knowledge extracted from

text

• Inference– Deduce new facts from existing facts and rules


Sentiment Analysis• Emotional content of text

• Pilot project combining– Sentiment analysis (computational linguistics)

– Concept extraction (linguistic semantics/ontology)

• Note this is just a pilot project and the computational linguistic method used is really basic, not state of the art

• Applications: – Fine grained search by features

– Ratings by review, not by stars, and integrated across sources

– Merge hotel ratings from different services that have different scales by using sentiment


Meadowood, St. Helena: Restaurant:10“In recent years the elegant but unstuffy dining room has won rave reviews, becoming a destination restaurant.“

Marys Lake Lodge and Resort, CO: Roadway: -8“Not to mention it is very expensive and located in a place that doesn't get much sun so it's icy and cold; and the maintenance of roads is terrible in winter.”


Word Sense Disambiguation

• Brown corpus of English– One million words

– “balanced” corpus from many sources – newspapers, novels etc.

• WordNet SemCor– Marked up Brown corpus by hand with parts

of speech and word senses


WordNet SemCor

• “Bed” sense 1– air_mattress curtain sleep sleeping_bag slipper

– SUMOTerm: Bed

• “Bed” sense 2 – compost decayed manure pansy spade

spread_over yard

– SUMOTerm: CultivatedLandArea

• “Bed” sense 3 – dry face homely river tilt

– SUMOTerm: GeographicArea


Sentiment words

type word POS polarityweak abandon verb negativeweak abate verb negativeweak abdicate verb negativestrong aberration adj negativeweak able adj positive

Created at U. Pittsburgh by having a team of students score words in WordNet, (and then validated by inter-subject agreement measures).


Sentiment Finding

Scorer

WordSentiment

List

StopWordList

SentimentScore

Disambig.Corpus

Word SenseDisambig.

WordSenses

SUMO-WordNetMapping

SUMOConcepts

Reviews ConceptSentiment

"The guest rooms, ... are somewhat shopworn or downright shabby ..."

room (in a building) – hall 12, guest 10... room (space) – grow 22,

Associate word senses with frequency of co-occurring words

shabby -5shopworn -1...

room (in a building) → Room

SF Drake: Room (-6)

Application: Remove amenities with negative sentiment from hotel search


Outline

• Current Application

• General concepts

• SUMO and WordNet– Ontology mapping

– Why upper ontology

• Sigma

• Pitfalls (in ontology development)

• SUMO details

• Language to Logic


C.K. Ogden/I.A. Richards, The Meaning of MeaningA Study in the Influence of Language upon Thought and The Science of SymbolismLondon 1923, 10th edition 1969

Concept

Referent

Refers To Symbolizes

Stands For“Orange”

Terms and Concepts

from the slide of [Bargmeyer, Bruce, Open Metadata Forum, Berlin, 2005]

Slide adpated from (c) Key-Sun Choi for Pan Localization 2005

Term

Ontology work should be here,since logic is needed to substitute for

human thought.

Lots of “ontology” workhas really been here.


Imagine...your view of the web

CV

name

education

work

private

Joe Smith

BS Case Western Reserve,1982MS UC Davis, 1984

1985-1990 ACME Software,programmer

Married, 2 children


...and the Computer's View

name

CV

education

work

private

Εα σολυμ μινιμυμ ευμΝιβχ σολετ υβικυε εα σιθ

Ει κυις σιμιλικυε ρεφορμιδανς πρω, αν λαβωρε δισερετ μινιμυμ δυο, σεδ εα σαλυταθυς σορρυμπιθ. Ιλλυμ φασιλις ιν πρι.

Μεα συ ιψυμ υλλυμ Μελ ευ κυωδ μεδιοσριθαθεμ

Υταμυρ θραξιθ


But wait, we've got XML -

<job name=”Joe Smith” title=”Programmer”>


But wait, we've got XML -

<job name=”Joe Smith” title=”Programmer”>

<x83 m92=”|||||||||” title=”..............”>


But wait, we've got Taxonomies -

Person

Mammal

JoeSmith


But wait, we've got Taxonomies -

o4839

x931

i3729


Wait, we've got semantics -

Person

Mammal

JoeSmith

instance

subclass

implies

Mammal

JoeSmith

instance


Wait, we've got semantics -

Person

Mammal

JoeSmith

instance

subclass

implies

Mammal

JoeSmith

instance

u8475

x9834

p3489

r53

r22

implies

x9834

p3489

r53


Taxonomy

• What's an automobile?– truck or sedan

– Alone it might be taken as not including trucks

– Does truck include 18-wheelers?

automobile

truck

AdamsHonda

sedan


Automation

• if d is an a, a can't be a d (usually)a

b

d

c


Fixing Meaning

Horse


Fixing Meaning

Horse is a mammal


Fixing Meaning

Horse is a mammal that has four legs


Fixing Meaning

Horse is a mammal that has four legs and is

capable of carrying ahuman rider that largely

controls its actions


Fixing Meaning

Caballo


Language Formality & Expressiveness

Formality

Expressiveness

Human LanguageSUO-KIF

weak semanticsweak semantics

strong semanticsstrong semantics

Is Disjoint Subclass of with transitivity property

Higher Order Logic

Logical Theory

Thesaurus

Has Narrower Meaning Than

Taxonomy

Is Sub-Classification of

Conceptual Model

Is Subclass of

DB Schemas, XML Schema

UML

First Order Logic

RelationalModel, XML

ER

Extended ER

Description LogicDAML+OIL, OWL

RDF/S

XTM

Syntactic Interoperability

Structural Interoperability

Semantic Interoperability

Thanks to Leo Obrst, MITRE

Note, these are languages, not ontologies.

OWL+RuleML


Frame Restrictions• b is between a and c

– (between1 a betweenness1)– (between2 b betweenness1)– (between3 c betweenness1)– vs– (between a b c)

• Adam is not an accountant– (notOccupation Adam Accountant)– vs– (not (occupation Adam Accountant))

• Existential vs. Universal quantification• Similar problems for many description logics• Very efficient computation however


Existential vs Universal Quantification

• All farmers like tractors.

• Some farmer likes a tractor.

(forall (?F ?T) (=> (and (instance ?F Farmer) (instance ?T Tractor)) (likes ?F ?T)))

(exists (?F ?T) (and (instance ?F Farmer) (instance ?T Tractor) (likes ?F ?T)))


First Order vs Higher Order

• (believes Mary (likes John Sue)) – higher order

• Higher order logic is very useful, but much harder to compute


Digression: Implementation is Different from Representation

• Why lose meaning at design time just because of runtime issues?– We can’t reason with English definitions, but that

doesn’t mean we shouldn’t document our terms

• Many different implementations may be done from the same representation

• This does not mean that run time issues should be ignored at design time– If you represent information you know can’t be

reasoned with, it better not be essential in most conceivable applications


Open Source

• There’s too much knowledge for any one entity to capture and code it

• Network effort – the more people that use the ontology, the more valuable it is– Needed to remove barrier to adoption

• Can’t anticipate how it will be used– Testing theories of linguistic analogy!


Ontology vs Language and Knowledge

Ontology

- expandable- language independent- machine understandable

Language

- understood by humans- ambiguous

Knowledge

- changes rapidly- may be local to an entity


Suggested Upper Merged Ontology

• Initial versions: 1000 terms, 4000 axioms, 750 rules

•Mapped by hand to all of WordNet 1.6• then ported to 3.0 and continually updated

•Associated domain ontologies totalling 20,000 terms and 80,000 axioms

– Now linked with factbases including YAGO for millions of facts

– New ontologies of Hotels and Dining

•Free• SUMO is owned by IEEE but basically public domain

• Domain ontologies are released under GNU

• www.ontologyportal.org


SUMO (continued)

•Formally defined, not dependent on a particular implementation

•Open source toolset for browsing and inference

–https://sourceforge.net/projects/sigmakee/

•Many uses of SUMO (independent of the SUMO authors and funders)

–http://www.ontologyportal.org/Pubs.html

http://ontolog.cim3.net/

http://ontolog.cim3.net/file/resource/presentation/EIDX_Semantic-Harmonization_20041201/Ontological_Expression_of_E-Business_Standards--KurtConrad-BoNewman-BobSmith_EIDX-20041201.pdf

http://home.earthlink.net/~adampease/professional/cohen-aaai99.ps


SUMO Structure

Structural Ontology

Base Ontology

Set/Class Theory Numeric Temporal Mereotopology

Graph Measure Processes Objects

Qualities


SUMO+Domain OntologyStructuralOntology

BaseOntology

Set/ClassTheory

Numeric Temporal Mereotopology

Graph Measure Processes Objects

Qualities

SUMO

Mid-Level

Military

Geography

Elements

Terrorist Attack Types

Communications

People

TransnationalIssues Financial

Ontology

TerroristEconomy

NAICSTerroristAttacks

…

FranceAfghanistan

UnitedStates

DistributedComputing

BiologicalViruses

WMD

ECommerceServices

Government

Transportation

WorldAirports

Total Terms Total Axioms Total Rules

20977 88257 4730

Relations: 1280

Hotel

Food&Dining


SUMO Validation

• Mapping to all of WordNet lexicon– A check on coverage and completeness (at a given

level of generality)

• Peer review– Open source since its inception

• Formal validation with a theorem prover– Free of contradictions (within a generous time bound

for search)

• Application to dozens of domain ontologies


WordNet

• A dictionary for computational linguistics applications

• 100,000 word senses, hand-created

• Open source

• Concise - No need for OED-style etymology

• Precise data structures

• Semantic links–Aid in computation

–Verification of meaning during construction


Formal Ontology

• WordNet has synsets for “earlier” etc• But nothing in WordNet would allow a

computer to assert that the end of one event precedes the start of another if one event is earlier than the other

• This is not a criticism of WordNet

time(<=> (earlier ?INTERVAL1 ?INTERVAL2) (before (EndFn ?INTERVAL1) (BeginFn ?INTERVAL2)))

Interval 1 Interval 2


Internationalization

• Translation of SUMO paraphrases to diverse multiple languages– Some confidence there’s no cultural or linguistic bias– Chinese, Hindi, Tagalog, Czech, German, Italian,

Korean, Romanian, Arabic

• SUMO is linked to multiple very large lexicons (Euro WordNet, Balkanet, HowNet etc)– English, Chinese, Italian, Arabic


Example #1 – Simple statement

• Robert has an orange.(exists (?orange) (and (attribute Robert-1 Male) (instance Robert-1 Human) (instance ?orange OrangeFruit) (possesses Robert-1 ?orange)))

∃o attribute(Robert-1,Male) ^ Human(Robert1) ^ OrangeFruit(o) ^ possesses(Robert-1,o)


Example #2 – Simple Query

• Who has a fruit?(exists (?fruit) (and (instance ?fruit FruitOrVegetable) (instance ?who Human) (possesses ?who ?fruit)))

∃f FruitOrVegetable(f) ^ Human(w) ^ possesses(w,f)


Counter-example

• Brutus stabbed Caesar with a knife on Tuesday.

(exists (?S ?K ?T) (and (instance ?S Poking) (instance ?K Knife) (instance ?T Tuesday) (agent ?S Brutus) (patient ?S Caesar) (time ?S ?T) (instrument ?S ?K)))

(exists (?S ?K ?T) (and (instance ?S stabs) (instance ?K knife) (instance ?T Tuesday) (agent ?S Brutus) (object ?S Caesar) (on ?S ?T) (with ?S ?K)))

● Logical translation - “word translation”


Example #3 – Stative

• Dickens writes Oliver Twist in 1837.(and (authors Dickens OliverTwist) (exists (?EV) (and (instance ?EV Writing) (agent ?EV Dickens) (equals (YearFn 1837) (WhenFn ?EV)) (result ?EV OliverTwist))))


Example #4 – Another Stative

• Bob is a pianist.(and (attribute Bob-1 Male) (instance Bob-1 Human) (attribute Bob-1 Musician))


Example #5 - Counting• Bob kills 5 rats.

(exists (?CLNrats ?event) (and (attribute Robert-1 Male) (forall (?I) (=> (member ?I ?CLNrats) (instance ?I Rat))) (instance Robert-1 Human) (instance ?CLNrats Collection) (member-count ?CLNrats 5) (agent ?event Robert-1) (instance ?event Killing) (patient ?event ?CLNrats)))


Example #6 - Preposition

• Bob in on the boat.

(exists (?boat) (and (attribute Robert-1 Male) (instance Robert-1 Human) (instance ?boat Watercraft) (orientation Robert-1 ?boat On)))


Example #7 – Another Preposition

• The party is on Monday.

(exists (?party ?monday) (and (instance ?party SocialParty) (instance ?monday Monday) (during ?party ?monday))


PrepositionsPreposition Class SUMO relationat,in,on location locationat,in,on time duringfor person destinationfor, through time durationwith person agentwith object instrumentacross path traverseswithin,into object properlyFillsfrom object originfrom time BeginFnthrough object traversesuntil time EndFnafter time greaterThanbefore time lessThan


Example #8 – Quantification

• Some horses eat hay.

(exists (?event ?hay ?horse) (and (instance ?horse Horse) (instance ?event Eating) (instance ?hay Hay) (patient ?event ?hay) (agent ?event ?horse)))


Example #9 – Possessives• Tom's father is rich. Bob's nose is big.

Mary's car is fast.(exists (?father) (and (attribute Tom-1 Male) (instance Tom-1 Human) (attribute ?father Rich) (father ?father Tom-1)))

(exists (?nose) (and (attribute Robert-1 Male) (instance Robert-1 Human) (attribute ?nose SubjectiveAssessmentAttribute) (instance ?nose Nose) (part ?nose Robert-1)))

(exists (?car) (and (attribute Mary-1 Female) (instance Mary-1 Human) (attribute ?car Fast) (possesses Mary-1 ?car)))


Example #10 – Negation• Bob did not write a book. Bob did not

write the book.(not (exists (?book) (and (attribute Robert-1 Male) (instance Robert-1 Human) (authors Robert-1 ?book) (instance ?book Book))))

(exists (?book) (and (attribute Robert-1 Male) (instance Robert-1 Human) (not (authors Robert-1 ?book)) (instance ?book Book)))


High Level Distinctions

The first fundamental distinction is that between ‘Physical’ (things which have a position in space/time) and ‘Abstract’ (things which don’t)

Entity

Physical Abstract


High Level Distinctions

Partition of ‘Physical’ into ‘Objects’ and ‘Processes’

Physical

Object Process


Objects

ObjectSelfConnectedObject

SubstanceCorpuscularObject

RegionCollection


Processes

DualObjectProcess Substituting Transaction Comparing Attaching Detaching Combining SeparatingInternalChange BiologicalProcess QuantityChange Damaging ChemicalProcess SurfaceChange Creation StateChangeShapeChange

IntentionalProcess IntentionalPsychologicalProcess RecreationOrExercise OrganizationalProcess Guiding Keeping Maintaining Repairing Poking ContentDevelopment Making Searching SocialInteraction ManeuverMotion BodyMotion DirectionChange Transfer Transportation Radiating


Abstract

SetOrClassRelationPropositionQuantity

NumberPhysicalQuantity

AttributeGraphGraphElement


Case Roles

• Roles that entities play in a Process– agent, patient, instrument etc.


Case Roles

• “Brutus stabbed Caesar with a knife on Tuesday.”

A Stabbing

A Tuesday

A KnifeBrutus

Caesar

patient

agent

time

instrument


Case Roles

• “Brutus stabbed Caesar with a knife on Tuesday.”

(exists (?S ?K ?T) (and (instance ?S Stabbing) (instance ?K Knife) (instance ?T Tuesday) (agent ?S Brutus) (patient ?S Caesar) (time ?S ?T) (instrument ?S ?K)))

A Stabbing

A Tuesday

A KnifeBrutus

Caesar

patient

agent

time

instrument


Example Rules(=> (instance ?DRIVE Driving) (exists (?VEHICLE) (and (instance ?VEHICLE Vehicle) (patient ?DRIVE ?VEHICLE))))

“If there's an instance of Driving, there's a Vehicle that participatesin that action.”

Not just an English definition for humans to read, but a logicaldefinition that can be used in proofs.

suggested upper merged ontology - hypotheses.org · –concept extraction (linguistic...

Documents