Lisp and the Semantic Web: It aint all hype.

1
Lisp and the Semantic WebIt aint all hype.

• Yarden Katz
• University of Maryland
• College Park

2
A Hype-Free Approach

• Legitimate frustration with XML and its vision
• ..Especially in Lisp community
• Present The Semantic Web differently not as a
  new phenomenon, but as an application of existing
  research (in the AI and KR worlds) to the Web
  that leads to new possibilities.

3
What is the Semantic Web?

• "The Semantic Web is an extension of the current
  web in which information is given well-defined
  meaning, better enabling computers and people to
  work in cooperation." (W3C)
• An attempt to represent and intelligently link
  data on the web using ontologies.
• A Lisper's Definition The Semantic Web is an
  attempt to apply AI to the web.

4
What can the Semantic Web do?

• "My agent will talk to your agent" replacing
  mundane human tasks with automated agents.
  Example scheduling an appointment at the
  doctor's office (more on this later.)
• Search engines on crack better, more precise
  searches, better languages support,

5
How does it work? Ontologies and W3C Standards

• Resource Description Framework (RDF) make
  statements about "resources" (anything that has a
  URI)
• RDF Schema (RDFS) Class hierarchy for RDF (Not
  important for today)
• OWL (Web Ontology Language) for writing
  ontologies definitions of concepts that we
  refer to when marking up our data.
• OWL-S (DAML-S, for now) Web Services in OWL

6
Making statements in RDF

• Statement John Steinbeck is the author of "The
  Grapes of Wrath", which is identified by
  http//www.grapes-of-wrath.com and was published
  in November 1939.

7
RDF/XML Serialization

• ltrdfRDF xmlnsrdf"http//www.w3.org/1999/02/22-r
  df-syntax-ns" xmlnsdchttp//purl.org/d
  c/elements/1.1/
• ltrdfDescription rdfabout"http//www.grapes-of
  -wrath.com"gt
• ltdctitlegtThe Grapes of Wrathlt/dctitlegt
• ltdccreatorgtJohn Steinbecklt/dccreatorgt
• ltdcdategtNovember 1939lt/dcdategt
• lt/rdfDescriptiongt
• lt/rdfRDFgt

8
Making Models with OWL - 1

• A basic class
• ltowlClass rdfID"Wine"gt
• ltrdfssubClassOf rdfresource"foodPotableLiqui
  d"/gt
• ltrdfslabel xmllang"en"gtwinelt/rdfslabelgt
• ltrdfslabel xmllang"fr"gtvinlt/rdfslabelgt
• lt/owlClassgt

9
Making Models with OWL - 2

• Class restrictions
• ltowlClass rdfID"Wine"gt
• ltrdfssubClassOf rdfresource"foodPotableLiquid
  "/gt
• ltrdfssubClassOfgt
• ltowlRestrictiongt
• ltowlonProperty rdfresource"madeFro
  mGrape"/gt
• ltowlminCardinalitygt1lt/owlminCardinalitygt
• lt/owlRestrictiongt
• lt/rdfssubClassOfgt
• lt/owlClassgt
• ..examples from McGuiness et al., OWL Guide

10
The Semantic Layer Cake

• Tim Berners-Lees Visualization

11
This stuff has some basis!

• OWL is a Description Logic.
• OWL Lite
• OWL DL
• and OWL Full, which is not a DL (more on this
  later.)
• OWL has a formalized foundation

12
We have DLs, thus we have reasoning

• Ian Horrocks Example
• Smoking causes Cancer Cancer causes death
• inference Smoking causes death.
• Reasoning in calendars Ability to make specific
  queries "Find all of Joe's afternoon flights in
  the next three days", etc.

13
Going back to the hype..

• A seemingly-hyped up example of Semantic Web
  Application The Scienific American article
  scenario by Tim Berners-Lee, James Hendler and
  Ora Lassila.
• Scenario follows..

14
The SciAm Scenario

• Lucy and Pete must equally partition the task of
  chauffeuring their ill mother to several doctor's
  appointments.
• Lucy's personal Semantic Web agent, built into
  her PDA, will communicate with Pete's to find a
  solution.
• Using the siblings' respective schedule, the
  agents together will not only divide the task
  between the two (keeping in mind the docs
  schedule), but also locate all relevant services
  needed to execute it (more on this later.)
• WWW03 Panel said No way

15
What information do we need?

• The siblings' schedule, in marked up form.
• The doctor's schedule, with her available
  appointment times
• Service discovery the pharmacy's service for
  ordering the drugs, the medical insurance's
  service for locating a doctor that is covered
  under insurance plan, etc.

16
What processing capabilities are needed?

• Ability to interpret web services descriptions
• The ability to determine best meeting times for
  three parties (doctor, Pete and Lucy) as well as
  divide the task equally
• Ability to invoke web services

17
This sounds a lot like an AI planning problem..

• It is!
• Use SHOP (Simple Hierarchical Ordered Planner), a
  Lisp planner (operates in Ordered Task
  Decomposition) to convert the DAML-S descriptions
  to problem descriptions.
• Planner returns a set of web services to execute,
  in order, which are called by a Java web service
  interface.
• Problem solved.

18
A SHOP Problem Description
(defdomain PrescribedTreatmentSchedule
((method (ns_TreatmentProcess) ((SHOPTotal
?SHOPTotal) (assign ?SHOPTotal_1 (
?SHOPTotal 1)) (assign ?SHOPTotal_2 (
?SHOPTotal 2)) (assign ?SHOPTotal_3 (
?SHOPTotal 3))) ((!!UpdateInstance 3)
(!!ns_getPrescription ?SHOPTotal_1)
(!!ns_getTreatments ?SHOPTotal_2)
(TreatmentProcess_ns_FillPrescription
?Prescription ?SHOPTotal_1)
(!!ns_GetDrivers ?SHOPTotal_3)
(TreatmentProcess_ns_TwoTreatmentTwoDriverSchedule
?Treatment1 ?SHOPTotal_2 ?Treatment2
?SHOPTotal_2 ?DriverList1 ?SHOPTotal_3)
(TreatmentProcess_FollowUpDoctor ?TreatmentD1
?SHOPTotal_2 ?TreatmentD2 ?SHOPTotal_2
?DriverListD ?SHOPTotal_3)))
19
SHOPs Final Plan
(((!NS_ORDERPRESCRIPTIONCVS PRESCRIPTION1)
(!MAKEAPPOINTMENTEMG1 APPOINTMENT78)
(!UPDATESCHEDULELUCY APPOINTMENT78)
(!MAKEAPPOINTMENTH1 APPOINTMENT182)
(!UPDATESCHEDULELUCY APPOINTMENT182)
(!MAKEAPPOINTMENTD APPOINTMENT203)
(!UPDATESCHEDULEPETE APPOINTMENT203)))
20
Another Semantic Web Application

• Carnegie Mellon Universitys myCampus
• Uses rules (in Jess expert system)
• Example I am a professor. If a student is trying
  to reach me, only return my status, i.e. if I am
  on or off campus. If faculty member is looking
  for me, return my location on campus. If the
  president of the university is looking for me,
  return the building I'm in, the room number,
  coordinates of where I am standing, and telephone
  number of the room.

21
The Semantic Web really is AI

• The above solutions are a result of well-founded,
  academic research into the areas of Knowledge
  Representation (KR) and Description Logics (DL).
• Semantic Web utilizes years of DL and KR research
  rather than throwing it away academia is put
  into practice.

22
Problems in Paradise

• OWL Full v. OWL DL
• Web Explanations? What is implied and what's
  explicit?
• Can anybody say anything on anything?
• What sources do you trust? CYC ontology example.
• Databases are private, but markup cant be for
  the Semantic Web to work

23
The Lisp Developer's point of view

• To handle Semantic data programmatically, we must
  be able to
• Retrieve access all elements of RDF. Wilbur, an
  RDF toolkit, handles this part.
• Reason make inferences about ontologies.
  OWLLisaKB is our solution.
• All of the above software is written in Lisp.

24
Wilbur RDF Toolkit

• Includes RDF parser, DAML parser, and now an
  add-on OWL parser.
• Node-centric uses Ivanoe, a frame-based path
  language to query RDF.
• Example find all labels of subClassOf relations
  in a graph starting from startNode.
• (all-values !startNode '(rep !rdfssubClassOf))

25
MINDSWAPs Lisp Development

• OWLLisaKB
• LISA a forward-chaning rule system, able to
  reason with arbitrary CLOS objects
• Wilbur encodes triples from RDF graphs as CLOS
  objects.
• OWLLisaKB a combination of the two to create a
  rule-based OWL reasoner.

26
OWLLisaKB How it works
27
OWLLisaKB A Simple Rule
(defrule subclass-instances () (triple
(subject ?C) (predicate "http//www.w3.org/2000
/01/rdf-schemasubClassOf") (object ?P))
(triple (subject ?I) (predicate
"http//www.w3.org/1999/02/22-rdf-syntax-nstype")
(object ?C)) (not (triple (subject
?I) (predicate "http//www.w3.org/1999/02/22-r
df-syntax-nstype") (object ?P))) gt
(assert ?I "http//www.w3.org/1999/02/22-rdf-sy
ntax-nstype" ?P))
28
MINDSWAP Search Page

• A pure Lisp SW application
• Uses OWLLisaKB to reason about over entire
  owl.mindswap.org database in approx. two-minutes
  (ask me about a demo later!)
• Interfaces the results to the browser via
  AllegroServe (Go Franz!)
• Ability to cross boundaries of ABox and TBox in
  searches (instances and taxonomies)

29
MINDSWAP Search Page
30
Conclusion
Lisp may never become more popular than Java/Perl
for currently common web tasks (CGIs, servlets,
etc.) ..But it has a significant advantage over
those languages in areas that are becoming common
on the (Semantic) Web, such as reasoning and
planning. Lisp could benefit from contributing
to the basis of the Semantic Web, and we could
benefit from you. LEARN MORE http//owl.mindswap
.org
31
Questions?
Credits The work presented here has been
the result of close collaboration with MINDSWAP,
namely Bijan Parsia and Jim Hendler.