Semantic Agent Programming Language (S-APL): A Middleware Platform for the Semantic Web

1
Semantic Agent Programming Language (S-APL)A
Middleware Platform for the Semantic Web

• Artem Katasonov and Vagan Terziyan
• University of Jyväskylä, Finland
• MSW at ICSC, Santa Clara
• August 4, 2008

2
Ontological coordination problem

• Two robots meet on Mars. If there is a need for
  cooperation or at least coordination
• Assume both follow FIPAs Agent Communication
  Language (ACL) can exchange messages, understand
  INFORM, REQUEST, REJECT, etc.
• Assume both follow Beliefs-Desires-Intentions
  (BDI) architecture share meaning of I intend,
  My goal is.
• Both designed for Mars hopefully have similar
  domain ontology concepts surface, rock
  (labels may be different though).

• Assume the US robot has a jet pack and thus can
  fly, while the EU robot has no idea about
  possibility of moving in third dimension.
• How the US robot can communicate its movement
  intentions, so that two can coordinate?

3
Ontologies that need to be shared
To have effective coordination, agents have to
share

• ExtOnt(A) - properties of the external world
• MentOnt(A) - internal mental properties of the
  agent
• BodyOnt(A) - properties of the agent's (physical)
  body
• InOnt(A)
• - properties of the sensory input
• - properties of the communication input
• OutOnt(A)
• - properties of the action output
• - properties of the communication output

• Domain ontology
• E.g. BDI model
• Available sensors and actuators
• Sensing vocabulary
• E.g. FIPAs ACL
• Acting vocabulary
• E.g. FIPAs ACL

So that jet pack and fly would have
accessible and interpretable descriptions
Are not treated
Impossible to standardize, so should enable
ontological modelling
4
UBIWARE project

• UBIWARE (2007 - 2010) Smart Semantic Middleware
  for Ubiquitous Computing.
• Work towards the Semantic Web of Things
  interconnecting Web Services, humans, and
  physical entities.

• Each resource has a proxy (representative) a
  software agent
• To adapt between the data formats and protocols.
• To enable new interactions (business processes).
• Agents are controlled (programmed) in semantic
  RDF-based language - S-APL.

• A

Example industrial maintenance case
5
Semantic Agent Programming Language (S-APL)

• Motivation Need to be able to ontologically
  describe not only the domain, but the agents
  themselves, to enable interaction without
  pre-programming each agent about all possible
  capabilities and behaviors of other agents.
• Similar to the challenge of semantic description
  of Web Services, but going beyond usage
  interfaces.
• Our approach RDF-based (more precisely, Notation
  3) agent programming language.
• In a nutshell
• Lets start treating agents programs as data.
• If it is data, lets use the semantic data model
  (advantages are well-known).
• Use Notation 3 (extension of RDF by Tim
  Berners-Lee) to overcome limitations of basic RDF
  and make code human-readable/writable.

6
S-APL

• It is a hybrid of semantic reasoners like
  Berners-Lees CWM and agent programming languages
  (APL) like AgentSpeak, AFAPL.
• From APLs point of view S-APL is a language that
  provides all the features (and more) as normal
  APLs, while being RDF based and thus providing
  advantages of semantic data model and reasoning.
• From semantic reasoning point of view S-APL is
  CWM extended with common APL features such as
• Beliefs-Desires-Intentions (BDI) architecture,
  i.e. ability to describe goals and commitments -
  data items presence of which leads to some
  executable behavior, and
• Ability to link to sensors and actuators
  implemented in a procedural language, namely
  Java.

7
S-APL Agent Architecture
Behavior Engine
Beliefs storage
Data
Rules
RAB
RAB
RAB
RAB
Blackboard
RAB Reusable Atomic Behavior
8
Levels of semantics in S-APL use

• The developers of a specific agent can decide how
  semantic they want it to be
• Semantic Reasoning. S-APL rules operating on N3
  semantic data.
• Semantic Data. RABs (i.e. Java components)
  operating on N3 semantic data.
• Workflow management. RABs operating on Java
  blackboard objects, with S-APL used only as
  workflow management tool, specifying what RABs
  are engaged and when.
• Any combination of above.

9
Examples of S-APL syntax

• Unconditional commitment to an action
• saplI sapldo javaubiware.shared.MessageSenderB
  ehavior
• saplconfiguredAs
• preceiver saplis John.
• pcontent saplis get temperatureIn
  room1.
• saplSuccess sapladd John informs
  temperature
• Rules
• ?room hasTemperature ?temp measuredBy
  ?sensor.
• ?temp gt 30
• gt ...

10
Non-semantic data transformation

• First, the original data from the resource is
  transformed into a Notation 3 representation
  based on the ontology of the data model of the
  original data (e.g. table-row-column
  ontology for tables). This transformation is
  performed by a Reusable Atomic Behavior (RAB).
• Second, the data from the data-model-ontology N3
  is transformed into the final domain-ontology N3
  using S-APL own means, i.e. rules.
• Our platform includes a set of standard RABs such
  as TextTableReader, SQLReader, XMLReader,
  XMLWriter, etc.

11
Example application architecture
RAB TextTableReader
SCADA agent
Operator agent
SCADA events in CSV
Rule Send every new event to the Operator agent
RAB XMLWriter
RAB HTTPResponseSender
12
S-APL as communication content language

• We use S-APL also as inter-agent communication
  content language
• One agent sends a query as
• saplI saplwant saplYou saplanswer
  ..query.. , the other answers with the
  matching part of its beliefs.
• One agent requests another agent to perform an
  action plan by sending the plan with
• saplI saplwant saplYou sapldo ..plan..
  
• Such interaction is easily programmed with S-APL
  itself. Our platform includes a set of standard
  S-APL models Listener, Believer, Informer and
  Follower.

13
Example application architecture 2
Note data processing is performed by DB agent,
but the code for that is sent to it by User agent
14
Policies in S-APL (example)

• ActionCommitment saplis
• saplI sapldo ?behavior saplconfiguredAs
  ?parameters
• .
• Action rdfssubClassOf ActionCommitment
• sbacrestriction saplI owlsameAs ?subject.
  saplI owlsameAs ?object.
• Print rdfssubClassOf Action
• sbacrestriction ?behavior Print.
• ?parameters saplhasMember pprint
  saplis ?print.
• Swear rdfssubClassOf Print
• sbacrestriction BadWord saplis ?word.
  sapltrue "contains(?print,?word)".
• Then, can specify a policy as e.g.
• fgEmployee Swear fgEmployee saplis
  sbacProhibition
• The platform includes the standard S-APL model
  SBACReasoner that enforces such access control
  policies.
• Example of use define an ontology of
  communicative acts and define a policy so that
  e.g. a Follower agent obeys orders of only a
  certain class of agents.

15
Conclusions

• It is not only that the Semantic Web needs agents
  as active software components to interconnect
  heterogeneous Web resources, but also agents need
  the Semantic Web technologies to enable open
  flexible coordination among them.
• By no means, we do not claim that our work has
  solved either of these challenges.
• We believe, however, that the S-APL language and
  other our related solutions provide a proper
  foundation for building middleware platforms for
  the Semantic Web.

16
Backup slides
17
UBIWARE project

• Current project of Industrial Ontologies Group
  (IOG) lead by Vagan Terziyan (Professor in
  Distributed Systems at MIT department).
• IOG now Vagan 2 PhDs 2 PhD students 1 MSc
  student 1 BSc student.

• UBIWARE (2007 - 2010) Smart Semantic Middleware
  for Ubiquitous Computing. Previous project
  SmartResource (2004-2007).
• Runs in Agora Center. Funded 80 by Tekes
  (Finnish National Agency for Technology and
  Innovation) and 20 by a set of companies
• Metso (paper industry), ABB and Fingrid (power
  networks), Inno-W (software), Nokia
  (telecommunications).
• In SmartResource participated TeliaSonera
  (telecommunications) and Tietoenator (software).

18
S-APL data model

• RDF compliant. Statements are reified and linked
  to resources of type ContextContainer through
  hasMember predicate (and inverse memberOf).
• Possibility of efficient object-oriented
  implementation inside the agents behavior
  engine
• HashMapltString, SemanticStatementgt and
  HashMapltString, ContextContainergt
• ContextContainer objects hold links to their
  member SemanticStatements

19
S-APL Rules

• Conditional commitment
• John loves ?x gt ...
• gt is the shorthand for saplimplies. If the left
  side of gt is truth (beliefs are connected with
  AND), the right side is copied into G. The
  commitment is removed after that (as fulfilled).
  Otherwise, it stays in beliefs and checked again
  and again until becomes true.
• Persistent behavior rule (not removed upon
  execution)
• ... gt ... saplis saplRule
• Conditional action
• ... -gt ... saplelse
• -gt is the shorthand for saplimpliesNow. Such a
  condition will be checked only once and will be
  removed even if the condition was false. Adding
  gbelse block defines what should be added to
  beliefs if the condition was false.

20
S-APL Rules (2)

• Exclusive condition
• John Loves ?x .
• saplI sapldoNotBelieve ?x .. ..
• gt ...
• If the object of sapldoNotBelieve matches G, the
  left side of gt is false. If there are several
  sapldoNotBelieve, they are connected with OR
• Commitment with a guard condition
• ... gt ... saplis sapltrue
• saplexistsWhile ...
• If the object of saplexistsWhile is false, its
  subject is removed in this case, the commitment
  is dropped.

21
S-APL Action commitments

• Unconditional commitment to an action
• saplI sapldo javaubiware.shared.MessageSe
  nderBehavior
• saplconfiguredAs
• preceiver saplis John .
• pcontent saplis bla bla .
• saplSuccess sapladd John was
  notified
• Special statements to add or remove beliefs the
  subject can be saplStart, saplEnd,
  saplSuccess, and saplFail. The predicate is
  either sapladd or saplremove.
• Sequential plan
• saplI sapldo ... saplconfiguredAs
• ... saplSuccess sapladd saplI sapldo
  ...

22
S-APL Querying constructs

• ?man loves ?girl. ?girl age ?age saplAll
  ?girl.
• ?person age ?age. ?age lt 25 saplAll
  ?person.
• ?person age ?age. ?person loves ?x saplis
  saplOptional saplAll ?person.
• John loves ?girl saplor Bill loves
  ?girl saplAll ?girl.
• ?person age ?age saplOrderBy ?age.
• saplOrderByDescending, saplLimit and
  saplOffset are also available. As in SPARQL

23
S-APL statements and containers

• After query ?x accordingTo Bill, can do
• ... ?x saplis sapltrue accordingTo John
  gt ...
• ?x saplis sapltrue
• saplI sapladd ?x
• saplI saplremove ?x (used as pattern for
  removal)
• saplI saplerase ?x (removed itself)
• ?x saplhasMember room1 hasTemperature 25
  (new member added)
• Can get ID of a statement with (then, can do most
  of above)
• hasTemperature 25 saplID ?x
  accordingTo Bill
• ?x rdfpredicate hasTemperature accordingTo
  Bill
• ?x saplis sapltrue accordingTo Bill
• ?c accordingTo Bill. ?c saplhasMember ?x
• Can do query like ?x saplis sapltrue
  according to Bill. ?x saplis sapltrue
  accordingTo John gt ...