Integration of Semantic Web Technologies
Dr David Mott, Dave Braines, Gareth Jones (IBM UK)
Dr David Mott, Dave Braines, Gareth Jones (IBM UK)
International Technology AllianceIn Network & Information Sciences
International Technology AllianceIn Network & Information Sciences
Context• Research Focus
– Collaborative problem solving across a network– Shared understanding between a team– How semantic web technology may use and integrate sources of information
• Hypothesis: shared understanding and collaboration facilitated by:– Standard set of shared concepts for building solutions (e.g. CPM)– ITA Controlled English for human expression of facts, rules and rationale– Rationale for showing how conclusions were arrived at– Argumentation for guiding how rationale is explored
• Demonstrate:– The planning of FIRE support, collaborating with Brigade Commander, exposing
hidden assumptions• “this problem is endemic in planning” LWS
– The planning of an NGO, integrating this planning information with public government information
Planning of FIRE Support
FIRES Planner
Brigade Planner
“I need fire support to cover my troops”
Plans must synchronise
Brigade Plan
Cross bridge, defeat, assuming peace talks fail Rationale for planStandard Set of Concepts
FIRES support requirement
FIRES receives problem (in CPM) Must supply Fire Support Sees rationale from the level up
FIRES must allocate resource
FIRES has 3 gun batteries Request to satisfy task fire_support Why must fire support start by 6?
To Task “fire_support”
Rationale for support latest start 6
Rationale shows reasoning We want users to see this!Graph of CE premises to conclusions
All rationale for latest start 6
Brigade and FIRES rationale Dependent on assumptionHuman and machine rationale
Plan fragment for latest start 6
Rationale mapped onto plan Alternative view for user Dependent on assumption
Fire support unachievable
A too small, B unavailable, C out of range Problem not solved But rationale for C Bty is ROUGH
Start by asking terrain ready reckoner
Travel time on LONG? Calculation (3 hrs) includes rationale Uses SemWeb technology (CPM)
Start to build rationale for “C Bty out of range”
FIRES constructs case for “C out of range” Why cant we get there? GUI or Controlled English
Construct full rationale
Analysis is complex Because BG1 using SHORT and LONG too slow How explain to BDE Cdr?
Divide into areas of reasoning
Abstract irrelevant detailNeed full detail for validation BUT need to summarise for Cdr
FIRES: key lines of reasoning
Main areas abstracted to single fact Full picture in 1 page Linked to detail if required
BDE Cdr: rebuts claim
BDE Cdr reviews Argues against doctrine by tactical imperative FIRES hidden assumption revealed
FIRES problem no longer unachievable
“Doctrine” Assumption unmade Knock on effects calculated C available to complete plan
NGO Planning Task
• To look after schools and other local services• To ensure that the schools are evacuated as
required for the current or future operation.
Source of data - data.gov.uk
UK Government initiative Publically available Uses Semantic Web RDF
(The previous planning map)
For reference, the planning map Geographic areas correspond (but colours not the same)
Demonstration …
Schools in the area
Map-based “Mashup” Obtain schools data from Web Overlay in area of operations
Overlay areas exported from plan
Previous Plan data “published” Overlay operational areas Uses Semantic Web RDF
An area has information based on assumptions
Area data from “published” plan Start and end times Plan rationale – the assumption
Need to Contact school
Schools in affected area Contact schools for evacuation Assumptions useful in decision making
Suppose the time is 2, and peace negotiations have
not yet broken down, might be worth waiting to see if peace is established
before evacuating
Suppose the time is 3.30, and peace negotiations
have not yet broken down, then probably need to
evacuate
Suppose peace negotiations have broken
down, then must evacuate
It is assumed that “peace negotiations broken down by 4”
NGO demonstration summary
• Semantic Web allows common representation of information and meaning– Ways to reference information– Ways to define common models
• Planning data made available in Semantic Web form:– private access– includes rationale and assumptions
• Existing social data (schools) available through pre-existing sources in Semantic Web form
• Easy to integrate these sources to provide new functionality:– What about road control? hospitals… weather …– Could be used within military too
• Achieves a shared understanding between the military and other organisations– within limits (e.g. security)
“CURIOUS” Demonstration Architecture
CPM(rules)
Brigade Plan (+Rationale)
FIRES Plan(+Rationale)
Map, terrain
BDE Planner
FIRES PlannerTerrain speed
the L118 Light Gun moves at 20 km on desertthe L118 Light Gun moves at 40 km on metalled
the L118 Light Gun moves at 10 km on woodland
SPAR
QL
endp
oint
“Mashup” applicationfor geographic social
effects
e.g. Hospitals, Schools
SPARQL endpoint
Mapping Data
Tasks, Areas, Rationale
OWL/RDF/CE
Engineer
NGO
Plan Visualiser
Plan Visualiser
Argumentation Visualiser
BDE Plan
Argumentation
FIRES Plan
Some Discoveries
Collaborative Problem Solving Model
basic logic and rationale
Agent, Assumption, ConceptualSpace, Container, Entailment, Inconsistency, PossibleWorld, Proposition, PropositionIndex, Quantity, ReasoningStep, Set, Triple, VarBinding, WorldState
general purpose ConceptualThing, Constraint, Synchronisation
temporal Precede, TemporalConstraint, TemporalEntity, TimeInterval, TimeLine, TimePoint
space Area, Elevation, Line, Point, SpatialConstraint, SpatialCoordinateSystem, SpatialEntity, SpatialIntersection, SpatialLocation, SpatialUnion
resources Resource, ResourceAllocated, ResourceCapability, ResourceConstraint, ResourceQuantity, ResourceSet
actions Activity, Effect, Precondition
collaborative problem solving
Choice Point, Collaboration, Commitment, Communication, ConstraintViolated, Decision, GoalSpecification, Influence, Issue, JointPersistentGoal, MutualGoal, Problem, Solution, Trust,
planning Allocation, Evaluation, EvaluationCriterion, InitialState, Plan, PlanTask, PlanTaskDescription, PlanTaskTemplate, PlanningProblem, PlanningProblemContext, ResourceCommitment, ResourceReq, TaskCommitment
A planning model should contain both the plan and the problem solving state
Statements the task 'Build Bridge' is achieved after the task 'Clear Road A‘ .the task ‘Build Bridge’ has 18 as the earliest start time.
Define Facts
Assumptions it is assumed by the agent A that the task ‘Build Bridge’ has 18 as earliest start time
Explore Hypotheses
Uncertainty It is true to degree A2 that the area a3 is a woodland terrain. Express Uncertainty
Logical Relations if ( the task T1 has the value X as earliest start time ) and( the task T1 is achieved after the task T )then( the task T has the value X as earliest completion time ) .
Capture logical connections between things, and use these to infer new information from existing data
Query for which task T is it true that the task T has the agent joe as executor Query for information
Rationale the task 'Clear Road A' has 18 as earliest completion time because the task 'Build Bridge' is achieved after the task 'Clear Road A' and the task 'Build Bridge' has 18 as earliest start time.
Explain reasoning, capture dependencies
New Concepts conceptualisethe task T ~ is achieved after ~ the task T1 .
conceptualisea ~ task ~ T that has the value V as ~ earliest start time ~ .
Create new models of things
Argumentation! in the argument arg1, by stating “she always lies” the agent fred disputes the claim that “helen told us that all feedback is good”
Analyse by challenging
ITA Controlled English
Controlled English is “curiously useful” for human and machine communication
Writing Controlled English
it is true that "the enemy is on the other side of the bridge".
there is an artillery unit named 'C Battery' that is a company and is a US unit and has friendly as affiliation .
there is a resource pool named 'C Bty Guns' that has '18' as quantity .
the unit 'FIRES' has OPCOM of the artillery unit 'C Battery' .
there is a plan named 'BDE Plan' that has the agent 'BDE' as executor and contains the objective 'Bridge Crossed' and contains the objective 'Deploy BG1' and contains the objective 'Enemy destroyed' and contains the task cross_bridge and contains the task destroy_enemy and contains the task move_to_oa .
the resource request rr0 is required by the task 'Advance to OA Rome' .
the task destroy_enemy occurs after the task cross_bridge .
the agent 'FIRES' states that the resource allocation constraint rac2 constrains the task fire_support and prohibits the resource 'C Bty Guns' because "C Bty out of range".
the agent 'BDE Cdr' states that the task destroy_enemy occurs after the task cross_bridge because the task cross_bridge realises the objective 'Bridge Crossed' and the objective 'Bridge Crossed' enables the task destroy_enemy.
the agent 'BDE Cdr' states that the task cross_bridge occurs simultaneously with the task fire_support because the task fire_support realises the objective 'Crossing Supported' and the objective 'Crossing Supported' supports the task cross_bridge.
the task destroy_enemy has 11 as latest completion time because the objective 'Enemy destroyed' has 11 as latest completion time and the task destroy_enemy realises the objective 'Enemy destroyed'.
the agent terrainRR states that the minimum path transit time 'mil:L118_LightGun_on_LONG4' has '3.08557' as minimum because the land route 'LONG' has unmetalled as classification and the maximum terrain speed ru3 has 10 as speed and the maximum terrain speed ru3 has unmetalled as terrain and the maximum terrain speed ru3 has 'mil:L118_LightGun' as resource and the land route 'LONG' has '30.8557' as length.
it is true that "the enemy is on the other side of the bridge".
Handwritten Domain Application
Editors
There are many ways to make writing CE easier, but CE should be readable by itself
the L118 Light Gun moves at 20 km on desert
Language “Extensions”
Hybrid Rationale
the agent FIRES states that "route SHORT is not available between 4-6" because "BG1 using SHORT between 0-12" and “C Bty and BG1 cannot use SHORT simultaneously".
[if ( the temporal entity T has the value X as earliest completion time ) and( the temporal entity T1 occurs after the temporal entity T ) then( the temporal entity T1 has the value X as earliest start time ).
Argumentation “Patterns”
Domain Application
Automated Reasoning
Handwritten User Rationale
Rationale must be integrated between human and machine to facilitate shared reasoning
RATIONALE
Logical Mappings between languages
Common Logic
ITA CE
RDF/S/OWL
RIF-FLD
Representations for different purposes must share a common semantics
MODELSConcepts
Logic RulesEvents
RationaleExplanation
DependenciesAssumptions
CollaborativeReasoning
ApplicationsHybrid user and machine
Domain specific
“Shared Understanding”Visualisation of Logic
Controlled English
The “CURIOUS” Reasoning Infrastructure
Integration of common concepts, CE, rationale
and logic will help facilitate shared understanding in
collaborative operations
BACKUP
Controlled Natural Language
A Controlled Natural Language is a human readable subset of English (or other natural language) that can also be machine parsed
understandable by machine and human Improves “impedance matching” between human and agent as both can
use the same language
Needs: A syntax (grammar) A lexicon (set of words and their grammatical roles) A semantics (things and relationships in the world) A mapping from syntax/lexicon to semantics (how does a word refer to a
thing?)
A CNL is easy to read, but harder to write
Different languages used by researchers: Rabbit, ACE, Controlled English
Controlled English Extensions But CE can be “stilted”, users want more natural expressivity We are exploring an extension mechanism
User-defined Linguistic transformation rule
“More Natural” CE
Basic CE .
the person fred attended the meeting finance1 with the person joe
the person fred attended the meeting finance1 and the person joe attended the meeting finance1
the Mk1Tank only fires the L15 round.
if ( the Mk1 Tank X fires the thing Y ) then ( the thing Y is an L15 round ) .
Examples
Definition of “only”
Anecdotal feedback on use of CE – Good Things• Non logical users can create models
Non-technical analyst SME could construct model on their own “As non formal logician, I can more easily construct models and instance data in CE”
• Improve Communication User requested a description of a planning scenario “in English”; the CE version satisfied their
request Use of text-based CE easily supported by Wikis, allowing easy communal sharing of CE models
and instances• Assists Design
“Concepts and rules are closer to my way of thinking and are easier to understand” Designing how to say something helped to clarify what the concepts really mean
• Common Language Rationale graph derived from human and agent reasoning can be seen as one due to use of
common language “Can combine queries of different information from totally different domains – “its all the same
language”
Anecdotal feedback on use of CE – need for improvement
Greater expressivity of syntax Multi-part relations
Greater expressivity of semantics Sets, embedded “Forall”
CE “intellisense” editors Context-sensitive words
“he”, “that”
Still experimental, BUT “Curiously useful”
ALL information must be represented in CEAny new CE syntax must make senseEven if not executing rules, still define the reasoning in CE
All information in one place in one formatDesigning syntax clarified understanding of semantics
Design Principles
Rationale may use structured or unstructured facts
• Rationale is defined in Controlled English– SENTENCE because SENTENCE– May contain structured facts and/or unstructured text
• Structured facts can match logical rules allowing further inferences– the person Fred is married to the person Jane because the person
Jane is married to the person Fred.• Unstructured text can represent information impossible to
capture in the model but cannot be used to match rules and generate new inferences– “I know Fred loves Jane” because “Jane told my brother”.
Why Rationale?
• Sharing of rationale enables team understanding of a solution (we hope)
• Human and machine reasoning may be integrated• Can be used to determine dependencies,
assumptions, knock on effects• Applications may generate rationale
automatically via the common conceptual model• BUT a standard to exchange for rationale is
required– The ITA “logic proposal” offers such a standard
Argumentation
Argumentation extends rationale to support informal reasoning
– Patterns of challenge and response• Why did you say that?• Your fact is wrong• Your reasoning is wrong
– Used to explore a problem when humans are uncertain – Can expose hidden assumptions and incorrect reasoning– Could be used to develop new concepts?
• Trying to argue may suggest missing properties or wrong conceptualisations
– Several Theories of argumentation
Argument
Claim A: we got good feedback
Response
Challenge
Justification QueryB: How do you know that?
Justification
Subargument
Claim A: Helen just said
all feedback was good
Subresponse
Challenge
Justification QueryB: You think that client
was nice to us?
Justification
Subargument
Claim A: If all feedback good then he didn’t write anything bad
Subresponse
Undercutting defeater
Subargument
Claim B: Maybe there
was NO feedback
Subresponse
Rebutting defeater
Subargument
Claim A: Helen couldnt have said
all feedback good
Subresponse
Rebutting defeater
Subargument
Claim B: No. The only situation she
couldn’t say it would be feedback that was bad
Feedback
Good
Rebutting defeater
Subargument
Claim A: Helen talking about “all
feedback received” implies its existence
Subresponse
AccepterA: OK
Subresponse
Rebutting defeater
Subargument
Claim B: Maybe she was being ironic
, “the best I can say is…
Subresponse
Rebutting defeater
Subargument
Claim A: No Helen is never ironic
Subresponse
AccepterB: OK well done
Subresponse
Using Lance J Rips Notation
Got good feedback
H says all feedback good
No bad feedback
Some feedback must existH is not ironic
NO feedback
No, according to logic all X is Y is true even if there is no X
H is ironic If you mention something
it must exist
Surely, if there is no X then you
cant say all X is Y
Incompatible
Incompatible
ARG2
Undercut (via alternative)
ARG3ARG1
ARG5 (logic)
ARG6 (linguistic)
ARG7
ARG8
Argument structures
rebut
rebut
ARG4
expand
Undercut (via alternative)
rebut
Undercut
Argumentation – Rebut Claim
• User clicks on rationale graph to add “Rebut Claim”• Argumentation CE generated in orange, and the corresponding rationale in blue
–Attempting to construct semantics of argumentation via:
Working with CUNY to
explore this idea
Argumentation CE Rationale CE
BDE Cdr rebuts claim
BDE Cdr reviews Argues against doctrine by tactical imperative Hidden assumption revealed