constituent ontologies and granular partitions thomas bittner and barry smith ifomis – leipzig and...

63
Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Post on 20-Dec-2015

217 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent ontologies and granular partitions

Thomas Bittner and Barry SmithIFOMIS – Leipzig

and

Department of Philosophy, SUNY Buffalo

Page 2: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

http://ontologist.com

• User Ontologies for Adaptive Interactive Software Systems (with I. Nebel)

Page 3: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Adaptivity

<< Hypoglycaemia >> -- improves performance in recall and behavior

Our idea: User Ontologies, Competency Ontologies

vs. Statistical Stereotyping Methods

Page 4: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

User-Ontology

vs. User-Profiles

Page 5: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

To support adaptivity:

• Need for reasoning simultaneously with cross-cutting ontologies at different levels of granularity

Page 6: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Overview

• The method of constituent ontology• Levels of ontological theory• The hierarchical structure of constituent

ontologies• The projective relation of constituent

ontologies and reality• Relations between constituent ontologies• Types of constituent ontologies

Page 7: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

The method of constituent ontology:

• to study a domain ontologically is to establish the parts and features in the domain and the interrelations between them

Page 8: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Examples of constituent ontologies

Page 9: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent ontologies

I

M

W

ND

SD

N

I

M

W

Page 10: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent ontologies

• Database tables • Category trees

Page 11: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Nice properties

• Very simple structure

• Correspond to the way people represent domains– In databases– Spreadsheets– Maps

Page 12: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Meta-level relations between constituent ontologies

Page 13: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Meta level (sub-ontologies)

I

M

W I

M

W

ND

SD

N

x yx is sub-constituent-ontology of y

Page 14: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Meta-level (granularity)

Page 15: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Meta-level (granularity)

• Alabama• Alaska• Arkansas• Arizona• …• Wyoming

• West• Midwest• Northeast• South

Page 16: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Levels of granularity

• Alabama• Alaska• Arkansas• Arizona• …• Wyoming

• West• Midwest• Northeast• South

• USA

Coarse Intermediate Fine

Page 17: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Meta-level (themes)

USA physical• Mountains • Rivers• Planes

Page 18: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Meta-level (themes)

USA physical• Mountains • Rivers• Planes

USA political• Federal states

Page 19: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Levels of ontological theory

Constituent ontology1

Constituent ontology2

Constituent ontologyn

Page 20: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Levels of ontological theoryObject-Level • Formal relations: mereology, topology, location• Space and time• Basic categories: entities, regions, perdurants, endurants, …

Constituent ontology1

Constituent ontology2

Constituent ontologyn

Page 21: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Levels of ontological theoryObject-level (Taxonomies, partonomies)• Formal relations: mereology, topology, location• Space and time• Basic categories: entities, regions, perdurants, endurants, …

Meta-level• Granularity and selectivity • Relations between ontologies• Negation, Modality

Constituent ontology1

Constituent ontology2

Constituent ontologyn

Page 22: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Object-level

Page 23: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Levels of ontological theoryObject-level

• Formal relations: mereology, topology, location

• Space and time

• Basic categories: entities, regions, perdurants, endurants

Meta-level• Granularity and selectivity (Theory of granular partitions)• Relations between constituent ontologies

Constituent ontology1

Constituent ontology2

Constituent ontologyn

Page 24: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Formal relations

• Mereology (part-of) -- Partonomy• Mereotopology (is-connected-to)• Location (is-located-at)• Dependence (depends-on)• Subsumption (is-a) -- Taxonomy

Page 25: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent ontologies

• A constituent ontology is an abstract entity

• Has constituents as parts

• Constituents are abstract entities that project onto something that is not a constituent itself

Page 26: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent ontologies as

granular partitions

Page 27: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Levels of ontological theoryLevel of foundation• Formal relations: mereology, topology, location• Space and time• Basic categories: entities, regions, perdurants, endurants, …

Constituent ontology1

Constituent ontology2

Constituent ontologyn

Meta-level• Granularity and selectivity (Theory of granular partitions)

Page 28: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent ontologies have a simple hierarchical structure

Database tables Category trees

Maps

Granular partitions

Page 29: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Cell structures as Venn diagrams and trees

Animal

Bird Fish

Canary

Ostrich

Shark

Salmon

Page 30: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent structures (1)

• minimal cells: H, He, …• non-minimal cells:

orange area, green area,yellow area (noble gases)...

• one maximal cell: the periodic table (PT)

Page 31: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Cell structures (2)

• - subcell relation• He noble_gases (NG) • NG PT• Partial ordering

Page 32: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Remember:Constituent ontologies

• A constituent ontology is an abstract entity

• Has constituents as parts

• Constituents are abstract entities that project onto something that is not a constituent itself

Granular partitions: Theory B

Page 33: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Projective relation to reality

Page 34: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituents project like a

flashlight onto reality

P(c, bug)

Page 35: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

A constituent ontology is like an array of spotlights

Page 36: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Pets in your kitchen

Bug 1 Bug 2 Bug 3 Bug 4

Constituent 1 Constituent 2 Constituent 3 Constituent 4

Page 37: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Pets in your kitchen

Constituent 1

Constituent 2

Constituent 3

Constituent 4

Constituent ontology

RealityProjection

Bug 1

Bug 2

Bug 3

Bug 4

Page 38: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Projection of constituents

constituent ontology

Targets in reality

Hydrogen

Lithium

Projection

Page 39: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Projection of constituents (2)

Wyoming

Idaho

Montana

Constituent ontology

North AmericaProjection

Page 40: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Projection and location

Page 41: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Location

L(bug,c) Being located islike being in the spotlight

Page 42: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Projection does not necessarily succeed

John is not located in the spotlight!L(John, c)

P(c, John)

John

Page 43: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Projection does not necessarily succeed

Mary is located in the spotlight! L(Mary, c)

P(c, John)

JohnMary

Page 44: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Misprojection

Idaho

Montana

Wyoming

P(‘Idaho’,Montana) but NOT L(Montana,’Idaho’)

Location is what results when projection succeeds

Page 45: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo
Page 46: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Transparency

Transparency: L(x, c) P(c, x)

P(c1, Mary) P(c2, John)

L(Mary, c1) L(John, c2)

Page 47: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Projection and location

Hum ans A pes U n ico rns

M am m a ls

Humans Apes

Dogs

Mammals

),Humans''( HumansP

lysuccessfulproject

NOT does Unicorn'' cell The

???),'Unicorn(' P

recognized

NOT is species The

???)L(Dogs,

Dog

)Humans'',(HumansL

Humans'' cell by the recognized

is species The Human

Page 48: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Functionality constraints (1)

Location is functional: If an object is located in two cells then these cells are identical, i.e., L(o,z1) and L(o,z2) z1 = z2

VenusEvening Star

Morning Star

Two cells projecting onto the same object

Page 49: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Functionality constraints (2)

China

Republic of China

People’s Republic of China

The same cell (name) for the two different things:

Projection is functional: If two objects are targeted by the same cell then they are identical, i.e., P(z,o1) and P(z,o2) o1 = o2

Page 50: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Preserve mereological structure

Helium

Noble gases

Neon

EmptyNeonHelium

gasesNobleNeon

gasesNobleHelium

EmptyNeHe

NGNe

NGHe

Potential of preserving mereological structure

Page 51: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Well-formed constituent ontologies are granular partitions

which are such that:

• Projection and location are functions

• Location is the inverse of projection wherever defined

• Projection is order preservingIf x y then p(x) p(y)

If p(x) p(y) then x y

Page 52: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Mathematical Models for COs: (Z, P, )

FTM• Partial order• Unique root• Finite chain of immediate

subcells between every cell and the root

GEM

• Partial order

• Summation principle

• Extensionality

P: Z • x y P(x) P(y)

• (P(x) P(y) x y))

Page 53: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Constituent ontologies are mappings

Object-level

Meta-level• Granularity and selectivity (Theory of granular

partitions)

Constituent ontology1

Constituent ontology2

Constituent ontologyn

Page 54: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Relations between constituent ontologies (COs)

Page 55: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Relations between constituent ontologies

Object-level

Meta-level• Relations between constituent ontologies

Constituent ontology1

Constituent ontology2

Constituent ontologyn

Page 56: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Ordering relations between COs

• P1 << P2

• << is sub-partition-of• << is reflexive, transitive, antisymmetric

I

M

W I

M

W

ND

SD

N

Page 57: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

I

M

W

ND

SD

N

Ordering relations between LGPs (2)

I

M

W

Z1

Z2

P1 P2

ff is

• one-one• into• order preserving

• if x y then f(x) f(y)• (if f(x) f(y) then x y)

P1 << P2

Page 58: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

P2 is an extension of P1

I

M

W I

M

W

ND

SD

N

Page 59: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

P2 is a refinement P1

<<Z1

Z2

P1 P2

f

Page 60: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Composition of COs

composition operation• P1 P2 = P3 iff

– P1 << P3 and – P2 << P3

IM

ND

SD

N

=I

M

W

NDSD

N

Page 61: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Composition of COs

IM

W=

I

M

W

NDSD

N

°ND

SDW

N

Page 62: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

Composition of COs

IM

IM

W IM

W=

Page 63: Constituent ontologies and granular partitions Thomas Bittner and Barry Smith IFOMIS – Leipzig and Department of Philosophy, SUNY Buffalo

The End