Reasoning Breakout Session

ReasoningReasoningBreakout SessionBreakout Session

7/20/01

Richard Fikes Mike DeanRichard Fikes Mike Dean

DAML PI MeetingDAML PI MeetingNashua, New HampshireNashua, New Hampshire

July 18-20, 2001July 18-20, 2001

Knowledge Systems Laboratory, Stanford University

2

Roles For Reasoning In The Semantic WebRoles For Reasoning In The Semantic Web

For ontology builders Classification

Inconsistency detection

For Web site builders Recognition

Inconsistency detection

For users seeking information Query answering

For users seeking services to take actions Planning


3

Why Reasoning Services?Why Reasoning Services? Ontology design

Check class consistency and (unexpected) implied relationships Particularly important with large ontologies/multiple authors

Ontology integration Assert inter-ontology relationships Reasoner computes integrated class hierarchy/consistency

Ontology deployment Determine if set of facts are consistent w.r.t. ontology Determine if individuals are instances of ontology classes No point in having a semantics unless exploited by “agents”

“The Semantic Web needs a logic on top”– Henry Thompson


4

A Reasoner For Every Web SiteA Reasoner For Every Web Site

Associate an information services agent with each Web siteAssociate an information services agent with each Web site

An expert on the information contained in that siteAn expert on the information contained in that site

Agent provides information services based on that expertise

> Query answering using the markup on the site as its knowledge base

> Some or all of the content of that knowledge base in various forms

E.g., RDF statements, KIF logical theory, HTML document, …

Agent is a “knowledge server” for the site

Perhaps Perhaps the *only* thing one encounters at a Web site is an agent(!)the *only* thing one encounters at a Web site is an agent(!)

Where one of the agent’s services is to provide the site’s pages


5

Query Answering ExamplesQuery Answering Examples Declaring an inverse of a propertyDeclaring an inverse of a property

<Property ID = "childchild">

<inverseOf resource = "#parent” />

</Property>

Bill is a child of Joe.Bill is a child of Joe.Is Joe a parent of Bill?Is Joe a parent of Bill?Yes.Yes.

Declaring a property to be a subproperty with a rangeDeclaring a property to be a subproperty with a range<Property ID = "fatherfather">

<subProperty resource = "#parent” />

<range resource = "#Man” /></Property>

John is a father of Joe.John is a father of Joe.Is John a parent of Joe?Is John a parent of Joe?Yes.Yes.What is John? What is John? A Man.A Man.


6

Query Answering ExamplesQuery Answering Examples Inferences using toClass and hasValue

ContentA Seafood-Course is a Meal-Course.A Seafood-Course is a Meal-Course.

Every drink of a Seafood-Course has white as a color.Every drink of a Seafood-Course has white as a color.

New-Course is a Seafood-Course.New-Course is a Seafood-Course.

W1 is a drink of New-Course.W1 is a drink of New-Course.

Key forward chaining rules> toClass rule 1

> (=> (and (PropertyValue onProperty ?r ?p) (=> (and (PropertyValue onProperty ?r ?p) (PropertyValue hasValue ?r ?v) (PropertyValue hasValue ?r ?v) (Type ?i ?r)) (Type ?i ?r)) (PropertyValue ?p ?i ?v))(PropertyValue ?p ?i ?v))

Query What is a color of W1? What is a color of W1? WhiteWhite


7

Inconsistency Detection ExampleInconsistency Detection Example Incorrect portion of translated Wines KB

(Type color Property) (PropertyValue maxCardinality color 1)

Assumed assertions (PropertyValue subClassOf Restriction Class) (PropertyValue subClassOf Class rdfs:Class) (PropertyValue disjointWith Property rdfs:Class) (PropertyValue domain maxCardinality Restriction)

Key forward chaining rules (=> (and (PropertyValue domain ?prop ?dm) (PropertyValue ?prop ?fr ?val)) (Type ?fr ?dm)) (=> (and (PropertyValue disjointWith ?c1 ?c2) (Type ?i ?c1) (Type ?i ?c2))

false) Conclusion

InconsistentInconsistent


8

Reasoning Work By DAML ContractorsReasoning Work By DAML Contractors Cycorp

Cycorp's OpenCyc for DAML ontologies will provide taxonomic inferences as described at http://opencyc.sourceforge.net/daml/daml-taxonomic-inferences.html. Cycorp has provided java bindings for its ontology navigation api that will soon be published at http://www.opencyc.org and http://www.sourceforge.net/projects/opencyc. Stephen Reed

Lockheed Martin, VIS, KestrelWithin the UBOT project (Lockheed Martin, VIS and Kestrel) we are working onconsistency checking of DAML ontologies. We have developed a program calledConsVISor which checks whether all axioms of DAML are satisfied by aparticular ontology or annotation. Additionally, we have translated DAML KIFaxioms into Slang. This allowed us to perform "deeper" consistency checkingof both the DAML axiomatization and of DAML ontologies and annotations. Oncean ontology is translated to Slang, we can not only check its consistency,but also perform reasoning (theorem proving). More information on our efforts can be found at http://vis.home.mindspring.com. Mitch Kokar


9

Reasoning Work By DAML ContractorsReasoning Work By DAML Contractors

Stanford KSLKSL is developing technology for reasoning with knowledge expressed in DAML on distributed Web sites. We are addressing both the standard issues about how to reason effectively with knowledge expressed in an object-oriented language augmented with rules and the issues raised by the knowledge using ontologies resident on (perhaps multiple) other Web sites. The technology includes a DAML reasoner called JTP implemented in JAVA that contains a general-purpose theorem prover integrated with a collection of special-purpose reasoners designed specifically for DAML+OIL and specific task domains. http://www.ksl.stanford.edu/projects/DAML/

Richard Fikes


10


TeknowledgeWhile it's more the focus of other projects rather than our DAML effort,we are doing some work with extending Mark Stickel's PTTP theorem prover tosupport our inference needs. We can read and do first order logicinference on a version of KIF. Since we can translate KIF to DAML andback, we expect that this software will be useful for our DAML efforts inthe coming year, especially as we develop more sophisticated ontologytranslation methods. Adam Pease


11


UMBCUMBC has developed an environment for reasoning with information expressed in DAML to support agents which do intelligent filtering of talk announcements as part of the ITTALKS application. The ITTALKS agent sends a user's agent ACL messages notifying them of new talks or changes to earlier talks using DAML as the "content language". The user's agent reasons about the new talk to decide (1) how well it matches the user's interests, (2) if it is feasible for him to attend based on his expected location and (3) it it fits his current schedule. If the outcome is positive, the agent places an item for the talk on the user's schedule. We are also doing a more limited range of reasoning with DAML using XSB in support of service description and discovery for bluetooth agents. Our current environment uses XSB as the inference engine, YAJXB as the bridge between XSB and Java, and the RDF API as a DAML parser. More information can be found at http://daml.umbc.edu/papers/, http://daml.umbc.edu/reasoning/ and by contacting Youyong Zou (mailto:[email protected]). Tim Finin


12


University of ManchesterThe FaCT system (http://www.cs.man.ac.uk/FaCT) provides reasoning services for the SHIQ description logic via a CORBA client-server interface. A simple translation of DAML+OIL into SHIQ allows FaCT to be used as a reasoning service for DAML+OIL (a direct DAML+OIL interface is under development). By using a highly optimized implementation of a sound and complete tableaux algorithm, FaCT is able to provide reasoning services that are both efficient and effective. FaCT is used by both the OilEd(http://img.cs.man.ac.uk/oil) and Protégé ontology editors to provide subsumption and consistency checking support for ontology design. Ian Horrocks


13


Yale/BBN/Kestrel

The Yale/BBN/Kestrel group is working on the problem of taking a service description from a web-based agent, and using it as the basis for planning. This raises several issues, all of which involve reasoning: (1) If the service description is in an unexpected vocabulary, how do you translate? (2) What must a service description look like in order for a planner to use it? (3) What changes to existing planners must be made for them to use these descriptions? (4) In the end, the only primitive actions you can take on the web are sending and receiving messages. How are these messages constructed and deconstructed?

Drew McDermott


14

A DAML+OIL Reasoning Working GroupA DAML+OIL Reasoning Working Group

Promote interaction and collaboration among DAML contractors working on reasoning E-mail distribution list Web site

Design consensus DAML query language?

Design consensus DAML justification language?


15

DAML Query LanguageDAML Query Language

Issues Relationship to DAML rules Is it confined to what is expressible in DAML+OIL? Expressive as SQL?

> Would then be problematic to represent in DAML+OILBecause of “operators” like max, min, average, ascending


16

Action ItemAction Item

Design query language for DAML+OIL Straw man proposal

Premises> Conjunction of RDF statements containing premise variables> Premise variables treated as existential

E.g., (and (Type ?sc Seafood-Course) (PropertyValue drink ?sc ?d))

Query pattern> Conjunction of RDF statements containing premise and query

variables> Each answer is a binding of the query variables for which the

query pattern is trueE.g., (PropertyValue color ?d ?c)


17


Promote interaction and collaboration among DAML contractors working on reasoning E-mail distribution list Web site

Design consensus DAML query language? Design a consensus language for querying DAML+OIL Design a consensus language for querying DAML+OIL

knowledge basesknowledge bases

Design consensus DAML justification language?


18

DAML Justification LanguageDAML Justification Language

Support both backward and forward inferencing

Conclusions, e.g., answer to query, includes link to general info about

the reasoner and its service

Augment query to indicate what kind/depth of justifications are desired Permit iteration to “investigate”

Minimal sufficient justification: {premises including sources} + {indication of KR entailment semantics}

Crucial: attribute the multiple sources Related: delegation, trust policies dealing with conflict

Typical rule/fact has source info for it

Issue: pass-thru of justifications from sources


19


Promote interaction and collaboration among DAML

contractors working on reasoning E-mail distribution list

Web site

Design consensus DAML query language? Design a consensus language for querying DAML+OIL knowledge Design a consensus language for querying DAML+OIL knowledge

basesbases

Design consensus DAML justification language? Do this in conjunction with and Do this in conjunction with and afterafter design of DAML rule language. design of DAML rule language.

Documents

Reasoning Breakout Session