Foundations of the Semantic Web

Download Report

Transcript Foundations of the Semantic Web

Foundations of the Semantic Web:
Ontology Engineering
Building Ontologies 2
Ontology Patterns
& Top Ontologies
Alan Rector & colleagues
Special acknowledgement to Jeremy Rogers & Chris Wroe
1
Common patterns & top level ontologies
• Patterns –
– what links with what by what property
• High level ontologies
– The top level view of the categories that can exist
• One approach to high top level ontologies
– The constraints for the common patterns are expressed in the top
level ontology
– Top Level ontologies are most easily expressed as a series
dichotomies1 which capture the distinctions needed for critical
patterns.
• Therefore work from ‘bottom up’, looking at patterns and deriving top
the level ontology
1Dichotomy:
split into two, from “di”: two; “chotomy” cut.
2
Basic Assumptions
• All primitives disjoint
– With specified exceptions for “ValueTypes” to be explained
• Primitives separated into a
– “Top Ontology” which are about broad abstractions
• “Thing”, “Process”, “Substance”, etc.
– Corresponds to our most abstract schemas and distinctions (dichotomies)
» Only some of which are relevant to most domain ontologies
– A “Domain Ontology Skeleton” – the things we really want to talk
about
• Modules, Universities, People, Exams, etc
– Domain primitives form disjoint pure trees
» Every domain primitive has exactly one domain primitive parent
» If not, requires “untangling”
3
Top Ontologies
• There are many proposed top ontologies
– See OntoClean, SUO, SUMO, Cyc, John Sowa, …
• What is presented here is an amalgam
– Mostly in plainer English, even at the cost of some loss
of precision
• Although I will present some of the technical words
– Motivated by common engineering issues
4
Common Patterns Summary (1)
• Independent things can have modifiers
– The sky is blue
– The exam is difficult
• Properties may be “reified”
– “The food is hot”
– The food has a temperature which is hot but cooling.
• NB OWL allows descriptions of concepts only, not properties
5
Common Patterns Summary (2)
• Things participate in processes / processes act on
things
– Alan participates in the process of lecturing
– Food participates in digestion
– Student participates in learning
• Agents are responsible for Acts
– Doctors order tests
– People buy and sell things
– Students take exams
6
Common Patterns Summary (3)
• Forms & ‘oeuvres’ are manifest in things
– “Hamlet” is manifest in a performance of “Hamlet” and
in script for “Hamlet”
– “The handouts for this course are manifest in both the
presentation of power point slides and the physical
paper handouts”
• Things and processes can be Physical or
NonPhysical/ “Abstract”
– A ball, the concept of a ball
– An plan; an physical action motivated by that plan
7
Common patterns Summary (4)
• Things are made of stuff/substances
– Books are made of paper; The desk is made of wood.
– Bones are made of bone_tissue; …
– A Lecture is constituted of speaking, gesturing, pacing about,
writing on the board,
• Things have parts in many different ways –
– Part-whole relations are distinct from containment, connection,
adjacency, branching, and other spatial relationships
• The study of “merology” – the study of parts and wholes
• Part-whole relations are often ‘reciprocal’ as well as
having inverses
– We want to say both
• “All fingers are parts of some hand” and
“All hands have parts some fingers”
8
Common patterns Summary (5)
• There are many different kinds of collections/aggregations
– Collections of small things become substances to make up big things
• Molecules make up substances
• Cells make up tissue
• Many things come in duals
– Some are clearly distinguished
• The “Bacteria causing the infection” is different from the “Infection caused by the
Bacteria”
• The “Degree earned by the study” is different from the “Study earning the degree”
– Some are often identified
• The “process of erosion” and the outcome of an “area of erosion”
• The “process of building” and the “building built”
• Some of this is a matter of language, but we often use the same word differently in
the same sentence.
– “The erosion has been going on for decades and covers thousands of square miles”
9
Independent Entities can have modifiers
Entity
Refining
Entity
Indepenent
Entity
Selector
Feature
State
ValueType
10
Ontology Patterns 1
Independent entities can haves modifiers
• Examples
–
–
–
–
–
–
“Tom is tall”
“CS646 exams are easy”
“Dolphin’s swimming is fast”
“Night is dark”
“This is my left hand”
“My left hand is resting on the right side of the desk”
• (We’ll work on this one properly later)
– The flow in the river is fast
• Modifiers/refiners
– Usually expressed by adjectives or adverbs in English
• Can be turned into nouns but this often makes odd reading, e.g.
– “The tallness of john”
11
Common dichotomy 1:
Independent vs Dependent
SelfStanding vs Refining
• Some things are “self-standing”
– We can talk about them on their own
– We cannot give an exhaustive list of their kinds
• People, books, modules, buildings, ideas, …
• Some things are “refining”
– They only make sense when applied to something else
– Their kinds are arbitrary or measured
• height – short, medium, tall – or 6ft
• quality – good, indifferent, bad
• laterality – left right
• “Self-standing” vs “Independent”
– Engineering justification vs philosophical justification
• treat as synonyms for now
– Explain more in advanced course
12
Refining concepts include “ValueTypes” & are
linked to self-standing concepts by properties
• SizeValueType
small
medium_sized
large
• Department hasSize someValuesFrom SizeValueType
• Values normally are
– disjoint
– Cover value type
• SizeValueType = small or medium_sized or small
• Refining properties are normally functional
(aka “single valued”, “unique”)
– It does not make sense to have two sizes
13
Example 2
• LateralityValueType = Left or Right
• Example
– Hand hasLaterality someValuesFrom LateralityValueType
– Screw hasThreatHandedness someValuesFrom LateralityValueType
– Hmm… perhaps we have a poor name for LateralityValueType
• “LeftOrRightValueType”?
14
Technical Pattern 1:
Reification1 of Properties
• We can re-express
Department has_size someValuesFrom Large
as
Department hasFeature someValuesFrom
(Size hasStateValue someValuesFrom Large).
• Any expression of the form
Object property someValuesFrom value|
can be re-expressed as
Object hasFeature someValuesFrom
(ReifiedProperty hasStateValue someValuesFrom state)
– The process of making real. In knowledge representation, making
something a object which can have properties in is own right and be a value.
15
Closely related to the linguistic phenomenon of “nominalisation” – forming a
noun from an adjective, adverb, or verb. (NB There is a second usage to be covered later)
1 Reification
Technical Pattern 1 (cont)
• We must be careful of cardinalities, when we re-express
Department has_size someValuesFrom Large
as
Department hasFeature someValuesFrom
(Size hasStateValue someValuesFrom Large).
• A Department should only have one Size – but can have many other
features.
– Solution 1: Qualified Number Restrictions – works in OilEd but not in official
OWL (yet)
• restriction Continuant1 hasFeature max-cardinality=1 Size
• restriction Department hasFeature someValuesFrom
(Size hasStateValue2 someValuesFrom Large)
“Continuant” is a very high level notion in the top ontology which subsumes
Department
16
2 “hasStateValue” is functional
1
Why Reify Properties
• In order to say something about the relationship
– i.e. in order to restrict the relationship
• Restrictions only apply to classes, not properties
– Example
“Departments that are large by comparison with Computer Science
Departments”
– Department and hasFeature SomeValuesFrom (Size and
(restriction hasValue someValuesFrom Large) and
(restriction hasTrend someValuesFrom Increasing))
17
Value, Trend, Temporal Pattern, Quantity
• Many features have a qualitative value, quantitative value,
trend, and temporal pattern, e.g.
• “Trains that are delayed, becoming more delayed, whose delay
is now 60 minutes, and whose delay is recurrent”
– Train and hasFeature SomeValuesFrom (Timekeeping and
hasQualitativeValue someValuesFrom Delayed
hasQuantitativeValueInMinutes value 601
hasTrend someValuesFrom Increasing
hasTemporalPattern someValuesFrom Recurrent)
1
NB “Concrete domains” – e.g. numbers and literals –
are in OWL standard but not currently implemented in
OilEd FaCT or Racer.
18
Exercise
• Add the difficulty of exams and size to the
department ontology
– Allow for an exam’s difficulty to be given a reason, e.g.
too long, material obscure, questions ambiguous, etc.
• Add size, publication output, and research income
to the Department
– Allow for trends
19
Dichotomy
Feature-State pairs vs Intrinsic Properties (“Selectors”)
• Some modifiers can change – “Being late”, “tall”,” beautiful”, “being to the right of something”
– Usually thought of as ‘accidental’
– Often something useful to say about the relationship
therefore useful to reify
• Making reification a consistent pattern often helpful – otherwise
difficult to remember which are reified.
– We have called the reifications “Features” 1
• values are often called ‘states’ because they are true during some time
period
– See temporal discussion later.
1
Guarino calls them “qualities” and some other authors use “feature” to mean
other things. However “feature-state pairs” is fairly common usage.
20
Dichotomy (cont)
Feature-state pairs vs Intrinsic Properties
• Some modifiers are intrinsic and help identify the object
–
–
–
–
being a right hand, a philips screwdriver, a flatbed truck, the first rib,…
Usually thought of as ‘intrinsic’
Do not change during the life of the subject
Nothing to say about the relationship, therefore
No value in reification
– Nomenclature unsettled. Called “selectors” in GALEN
• StateValues can have further modifiers
– “A very severe cold”
– “A moderately good wine”
• Intrinsic properties cannot
– “A very right hand”?
– “A very third finger”?
21
relates_to
Things
(Continuants)
PhysicalSructures
Buildings
Computers
…
owns
Independent
Entity
has_part1
participate_in
Processes
(Occurrents)
Act
Agent
Lecture
Organisation
Exam
University
isAgentOf
Module
Department
…
Research_group
Person
1
Should strictly
distinguish “monads”
22
Common Pattern
Processes act on things
Things “participate in” Processes
“Continuants” participate in “Occurrents”
“Endurants” participate in “Perdurants”
•
•
•
•
•
Person participatesIn Journey
Student participatesIn Study
Water participatesIn Flow
Lecturer and student each participatesIn Lecture
Ball and Footballer each participatesIn Kick
•
•
Numerous different kinds of participation
Can distinguish discrete and continuous “Occurrents”
– discrete – Journey, Kick, Lecture
– continuous – Flow, Study, Speaking, Listening
But not always easy or useful to do so
Lenat termed continuous Occurrents “Activities”
23
Basic Dichotomy
• Continuants vs Occurrents
– Roughly Things vs Processes
• Processes change the state of things
– Exact relationship left to philosophers
• “Continuants”  “Endurants”  “Things”
– Entities which endure in time
– Their parts likewise endure in time and are separated in spatial dimensions
• Books, people, ideas, software, buildings, metal, plastic, …
– May have “versions” and be modified by “feature-state” pairs at different periods of
time, but have an identity which endures through time
• “Occurrents”  “Perdurants”  “Processes & Events”
– Occur at or during some point or period of time
– Their parts are separated in time (as well as space)
• “Subprocesses”, “subevents”
24
Important observation
Not All Ontological Patterns are relevant to all Ontologies
Not of the Top Ontology are relevant to all applications
• The pizza ontology didn’t have any processes or events
– The pizza menu is about things not processes
• A model of baking pizzas would need processes
• A model of pizza recipes would need processes
• A model of pizza menus does not need processes
• Most practical ontologies need only a bit of the top
ontology
25
False Dichotomy
Events vs Processes
• “Process” vs “Event” a point of view rather than intrinsic
– “John ate dinner at 6:00 pm. He went to the movie at 7:30pm. After the
movie he went home”
– “During the dinner the telephone rang at 6:15. The movie lasted two
hours. During the trip home he stopped to buy petrol”
• “Eating dinner”, “Attending the movie”, “The trip home”
– Each treated sometimes as occurring at a definitive time point– an event
– Each treated sometimes as occurring over a time interval - a process
• NB the way in which verb moods mark processes and events varies even in
Indo-European languages.
• Best treated as “points of view” rather than properties of the Entity. (See
Davis)
– Not all philosophers agree, but we shall treat “event” and “process” as equivalent in
this course.
26
Many procssses have “Actors” (“Movers) and
“Objects”/”Targets” and other roles (cases)
•
Actors and Objects
– The River erodes the valley
– The Person eats the food
– The stone breaks the window
•
ActorContinuant isActorOf someValuesFor
(ProcessOccurrent hasTarget someValuesFor ObjectContinuant)
•
We use “hasTarget” or “actsOn” rather than “hasObject” because it avoids confusion
with linguistic usage
– Which verbs are transitive and what prepositions/case they require varies from
language to language and construction to construction within a language;
•
•
•
•
•
I am waiting for John
I await John’s arrival
J’attends Jean
I expect John’s arrival
J’attends à l’arrivée de Jean
27
Some actors are “responsible initiators”:
Agents1 vs nonAgents
• “Agent” – a key notion for most organisational ontologies
– Entities that can initiative actions and take responsibility for them
•
•
•
•
•
•
People
Governments
Committees
Companies
Animals?
Gods?
– Aristotle’s & the Deists’ notion of God was as “First mover”
– “Mover” another alternative label for “Actor”
• Processes which have responsible agents are called “Acts”
1
NB: There is no agreement on the use of the labels “Actor” and “Agent”. In linguistics
“agentive” is a case roughly corresponding to “Actor” here We make the distinction in
this way because many top ontologies use “Agent” in the sense of “responsible initiator”.
28
However, in other ontologies, I.e. Schank, “Agent” is used as we use “Actor”.
Being an agent may carry legal
responsibility
• An adult is a legal agent in most situations and has legal
responsibilities for their acts
• A child is not a legal agent in most situations
• Organisations may or may not be legal agents
• If needed
– Distinguish between de facto and de jure (legal) agents
• de jure agents often derive their agency from an authority
– The University, the Government, the Bar, the General Medical Council,
…
– Agency is complicated in many organisational schemes
• Best ‘reified’ so that it can be described
– However, in this course we will keep it simple and make it just a simple
property isAgentOf
29
Many processes have a source or a
recipient
• Consider
–
–
–
–
John gave the book to Mary
John received the book from Peter
John received the book from Peter and gave it to Mary
The surgeon transplanted the kidney from John to Mary
• Often does not matter – useful to have abstractions
– hasSourceOrRecipient
hasSource
hasRecipient
30
Many processes have outcomes
• Consider
–
–
–
–
–
John wrote the book
The architect created the design for the house
The gene was expressed as a protein
The student wrote the exam paper
The river eroded the canyon
• Outcomes can in general be more general – i.e. states
– The storm washed away the beach
• In this course we will stick to simple cases where outcomes are things
– Let’s us run together several intervening steps
31
Role hierarchy for participates so far
• Property hierarchy provides means of abstraction
• Relations between occurrents and continuants
hasParticipant
hasActor
hasAgent
actsOn
hasTarget
hasSourceOrRecipient
hasRecipient
hasSource
hasOutcome
participatesIn
isActorOf
isAgentOf
isActedOnBy
isTargetOf
isSourceOrRecipientOf
isRecipientOf
isSourceOf
isOutcomeOf
32
33
Common Dichotomy
Physical vs NonPhysical/Astract
• Physical things are in the physical world
– Have mass or energy – or both – or be part of the physical
surroundings
• We’ll stick to a fairly naïve physics for this. We don’t need quantum
mechanics to describe the University!
– NB not necessarily “tangible” (touchable)
• Energy, signals, magnetic fields, … are not tangible but are certainly
physical
• Space & Time (although require such special handling that it hardly
matters)
• Includes both “continuants” and “Occurrents” – “Things” and “Processes”
34
Common Dichotomy
Physical vs NonPhysical/Astract
• Nonphysical things are in the psychosocial world
– We won’t bother too much about how
• leave that part for the philosophers
– Some separate the Psychological from the Social World
• But then a story which is composed in one person’s mind (Psyche)
must become something different when it is told to and remembered
by someone else and enters the social realm)
– Separating Psyche and Social means we must make this decision
» Do our potential applications require it?
If not, best blur it.
» an example of an “Ontological Commitment”
• NB: The word “Abstract” is badly overloaded
– Often used in OO programming for classes that are not allowed to
have direct instances
• Including many ontology tools
– So I will just use “NonPhysical”
35
Abstract Patterns are manifests in Physical Entities
Nonphysical/
Abstract
Entities
Patterns
Natural Patterns
Shapes etc
Oeuvres
Literature
Music
Art
Theories
…
Physical/
Entities
manifests_in
Copies of Books
Performances
Paintings
Texts
Resources on the Web
36
Common Pattern
Abstract Patterns are manifests in Physical Entities
• Consider
– “Hamlet” manifests in a performance of “Hamlet”
– “Hamlet” manifests in the “Script for Hamlet” manifests in “This copy of
the script for ‘Hamlet’ held in the University library”
– “Hamlet” was written by “Shakespeare”
– “Your project” manifests in the paper copy you hand in or the electronic
copy you email.
• What is “Hamlet”
– The abstract “Pattern” or “Form”
• Exists in psycho-social space somehow
– Engineers need not care about the details
– Abstract “Patterns” which are the outcome of “Acts” by “Agents” we call
“Oeuvres”
• From the French for “works” but less ambiguous in English than “Works”
which has 17 meanings in my dictionary.
37
Patterns, Oeuvres and their Manifestations
• Our biomedical ontologies ignore the distinction most of
the time
• Interested either in bodies or pieces of paper / messages in wires
• Librarians must be very careful of these distinctions – and
their refinements
– Must distinguish
• the abstract notion of a novel, play, symphony, poem, etc.
• their specific editions (versions)
• the copies on paper, microfiche, digital media held in their stores.
– Copyright lawyers and IP specialists likewise
• Important manifestation on the Web:
“You may link to this but not copy it to your web site”
38
More NonPhysical Things
• Mental things
– ideas, feelings, perceptions, …
• Social things
– organisations, governments, languages, …
• Patterns
– The patterns manifest in physical things
• either composed or abstracted from them
– For most purposes can keep it simple
– Roles – social_roles, physiologic_role, functional_role, etc.
• Collections and aggregations – (see later in presentation)
– Aggregations, multiples, and collections other than mathematical sets
• Symbolic things
– Mathematical and logical objects
1
A significant proportion of philosophy is devoted to the status o of abstract things such as mathematical objects,
lost works of Aristotle, etc. We shall content ourselves with a five-way split. In practice, this is actually with four,
since most symbolic things will end up in “concrete data types”. Since OWL explicitly excludes concrete data
types from from the general classification lattice, their status has no engineering significance
39
Question?
• Is CS646_2003 a process in the physical world or
an abstract pattern for that process?
– How would we know? What questions should we ask?
– We could probably distinguish
• a specific instance of a “named pattern” – the thing that
appears in the syllabus – clearly something nonPhysical
• the actual course itself with its lectures, labs etc – clearly
physical
• Would it be helpful to do so?
– Can you think of examples
» Do they matter for this application?
40