DAML JBI TIE

Mike Dean
Principal Investigator, DAML Integration and Transition
mdean@bbn.com

AFRL/IF
Rome, NY
19 June 2003

http://www.daml.org/2003/06/daml-jbi/Overview.html

$Id: all.htm,v 1.16 2003/06/24 08:34:03 mdean Exp $

Outline

• Objectives
• Previous Work
• Recent Work and Experimental Results
• Status and Plans
• Conclusions

Primary Objective

• assess the value added by DAML+OIL and the Semantic Web to the Joint Battlespace Infosphere

Secondary Objective

• demonstrate benefits in the context of the Semantic Operational Network Assessment Tool (SONAT) in the DAML Experiment

Tertiary Objectives

• support the use of contextual information (e.g. reference Order of Battle to support queries like "all the units that report to me")
• move toward statement-level publish/subscribe (e.g. deliver only the portions of an Air Tasking Order that pertain to my unit)

Previous Work

• kickoff briefing
• JBI 0.8 used for Battle Damage Assessment (BDA) agent in DAML Experiment (SONAT)
  • included in plenary demo at October 2002 DAML PI Meeting
  • standard XML, no DAML
  • BDA reports not a very rich application domain
• Minnowbrook update
  • DAML+OIL ontology for Information Object definitions
  • DAML specialization of JBI 0.8 API using CoABS Grid and JINI
    • DAML metadata converted into JBI 0.8-compliant XML representation
  • focus on use of Information Object hierarchies to aggregate pub/sub registrations
    • registrations internally expanded based on Info Object hierarchy
  • test harness showing multiple publishers and subscribers

Recent Work

• back off from JBI implementation to assess general applicability of Semantic Web technology
  • integrate when proven
• identify a richer application
• migrate from DAML+OIL to OWL
  • the OWL Web Ontology Language is the revision of DAML+OIL from the W3C Web Ontology Working Group

A Semantic Web-based JBI

• my observation from Minnowbrook: Semantic Web technologies could meet the key requirements of JBI (pub/sub)
  • Information Object definitions (OWL)
  • Metadata Repository (WWW: URI naming, HTTP, searchable cache like DAML Ontology Library)
  • metadata (RDF)
  • predicate language (DQL and/or other query)
  • Information Object Repository (DAML DB)
• alternative JBI approaches
  • XML (current)
  • middle ground (e.g. OWL-specific metadata and queries in CAPI)
  • Semantic Web
• Semantic Web is probably also highly applicable to fuselets and force templates (not addressed here)

An OWL Ontology for Information Objects

• http://www.daml.org/2002/11/jbi/jbi-ont (dumpont)
  • reflects discussion in Information Object subgroup at Minnowbrook (e.g. relations can be separate InfoObjects or embedded in InfoObjects)

Notices to Airmen (NOTAMs)

• effort under AFRL In-Flight Management (IFM) PRDA
  • Ed DePalma is COTR; Rusty Bobrow of BBN is PI
  • apply natural language technology to generate OWL from "pidgin" text used in NOTAMs
  • example (much more structured than most):
    

      
    A0126/03 NOTAMN A) KSFO B) 0304010630 C) 0304041500 D) 0630-1500 DLY E) QMRLC 01R/19L
              

    
    

    <NOTAM rdf:about="uri:uuid:c2f41010-65b3-11d1-a29f-00aa0c14882">
      <qcode>QMRLC</qcode>
      <icao rdf:resource="http://www.daml.org/cgi-bin/airport?KSFO"/>
      <text>A0126/03 NOTAMN A) KSFO B) 0304010630 C) 0304041500 D) 0630-1500 DLY E) QMRLC 01R/19L</text>
      <id>A0126/03</NS0:id>
      <effective>200304010630<effective>
      <about>
        <RunwayStatusReport>
          <runwayOf>
            <Runway/>
          </runwayOf>
          <statusOf>
            <RunwayClosed/>
          </statusOf>
          <intervalOf>
            <IntervalDescription>
              <daysOf>
                <DayOfWeek/>
              </daysOf>
              <low>0630</low>
              <high>1500</high>
            </IntervalDescription>
          </intervalOf>
        </RunwayStatusReport>
      </about>
    </NOTAM>
    

  • over 1 million NOTAMs processed
    • we're using a snapshot of 24,374 NOTAMs active on March 30
  • caveats
    • "bottom-up ontology" from natural language parser
      • could canonicalize via post-processing
    • currently handle 50-75% of the text
• NOTAMs are internally typed using QCodes <= look here
  • OWL representation (TBD) (dumpont)
  • caveat: not required for civilian NOTAMs
    • could be synthesized from other features
  • real-life example of multiple inheritance

Subscriptions as Class Definitions

• JBI currently uses XML templates or XPath expressions in subscription requests
  • we can instead express JBI subscriptions as OWL class definitions
• owl:hasValue can handle exact value matches, e.g.
  

   
  <owl:Class rdf:ID="RunwayClosure">
    <sameClassAs>
      <owl:Restriction>
        <owl:onProperty rdf:resource="#qcode"/>
        <owl:hasValue>QMRLC</owl:hasValue>
      </owl:Restriction>
    </sameClassAs>
  </owl:Class>
        

• we can extend this to value ranges using XML Schema datatype reasoning, e.g. lat/lon box
  • Jeff Pan's dissertation work at Manchester
• simplifies API
  • moves the information from the client to a (private) ontology definition
• allows use of Description Logic classifiers
  • Description Logic Implementors Group (DIG) protocol supports
    • FaCT (open source)
    • RACER
    • Cerebra (a scalable high-performance commercial reimplementation of FaCT)
• possible future inference option: Jena 2 (inference layer under development)

DAML Query Language (DQL)

• we've discussed using the DAML Query Language (DQL) as a predicate language for JBI
• DQL example (N3 syntax): Runway Closures at San Francisco International Airport
  

  ?notam rdf:type notam:NOTAM
  ?notam notam:airport <http://www.daml.org/cgi-bin/airport?KSFO>
  ?notam notam:qcode "QMRLC"
        

• RDF (Semantic Web) Query Working Group is likely next step in W3C Semantic Web Activity
  • probably 6 months off
  • 10+ groups currently working on some aspect of Semantic Web Query
    • DQL is at one extreme
  • renewed interest in applying XQuery to Semantic Web (RDF-based) content
    • RDF Query BOF at WWW 2003 last month

Java Expert System Shell (JESS)

• Java Expert System Shell (JESS)
  • OPS 5/CLIPS family production rule system
  • implemented in and accessible from Java
  • widely used
  • developed by Sandia National Labs
  • some license issues
    • need to fully investigate
• our usage
  • provided rules to implement the DAML+OIL language features we used in class expressions: hasValue, subClassOf, sameClassAs, intersectionOf, hasClass
  • employed a retraction mechanism like that used for DQL premises
  • integrated using the JESS Java API
• UMBC has developed a DQL implementation using JESS
  • need to investigate

Information Repository

• DAML DB
  • memory-mapped persistent linked storage optimized for RDF/DAML+OIL/OWL
  • intended as persistence layer beneath higher-level query and inference tools
  • Semantic Web Database Scalability demonstration
    • shown at DAML PI Meeting and SWMU
    • 100 million+ statements on 64-bit architecture
• we currently use DAML DB to store NOTAMs being published
  • same representation as Info Object Repository usage
• loaded 24,374 NOTAMs in 5 minutes

Experiment Setup

• architecture
  
• subscriptions
  1. RunwayClosures at San Francisco International Airport
  2. RunwayClosures at any U.S. airport
• simplifications
  • used subset knowledge base for testing
  • used 7164 NOTAMs with qcodes and US ICAOs

Experiment Results

• milliseconds to match Information Object against 2 subscriptions on my 850 MHz laptop:
  
• JESS provided reasonable performance
• we're very close to having results for RACER and FaCT
  • expected to fall between JESS and DQL

Limitations of Class Expressions

• see Venn Diagram: Expressive Overlaps among KR's (page 6)
• DQL covers DL and Horn Logic Programs because it's a rule language that also uses OWL types
• an example of a DQL query that can't be expressed as a DL class expression: domestic flights
  

  ?flight rdf:type flight:Flight
  ?flight flight:origin ?origin
  ?flight flight:destination ?dest
  ?origin airport:country ?country
  ?dest airport:country ?country
        

• probably not a requirement to support such queries (at least initially)

Translating Queries/Subscriptions

• OWL class expressions aren't natural to most people
• we should be able to translate a query from a more familiar language (possibly restricted DQL) into class expressions
  • similar to current translation from XPath to JMS SQL
  • also allows sharing of common Restrictions

Financial Status

• provided offline

Possible Next Steps

1. provide feedback to DQL and DIG developers
2. scalability testing
   • full KB
     • all 178 x 88 = 15664 QCode combinations
     • associate countries with all 9177 airports
   • 100s of subscriptions
3. integrate KSL DQL and DAML DB
   • implementation of Info Object Repository
   • persistent forward chaining for KB
4. embed into JMS JBI platform
   • separate servlet
5. additional applications

Conclusions

• OWL is good for specifying Info Object types and metadata
• most subscriptions can be cast as OWL class expressions
  • more powerful than existing JBI subscription requests
  • allows effective classification of InfoObjects across all subscriptions
  • same mechanisms can be used to query Info Object Repository
  • performance seems acceptable
• (interfaces to) reasoners are still temperamental

Acknowledgements

• Troy Self
  • embedded DQL
  • JESS integration
• Dave Rager
  • DIG, RACER, and FaCT integration
• Jeremy Lerner
  • high-resolution timer