Ontology-Based Computing

Kenneth Baclawski Northeastern University and Jarg

The Onslaught
Increasingly large amounts of information is becoming accessible electronically. The information sources are increasingly complicated. The diversity of types of information source is also increasing.
Technologies are emerging to cope with this onslaught: ontology-based computing.

Ontologies
Shared understanding within a community of people Declarative specification of entities and their relationships with each other Constraints and rules that permit reasoning within the ontology Behavior associated with stated or inferred facts

Relational Database Schemas

Well established technique for specifying the structure of shared data, not for communication between people or agents Declarative specification but of tables, not of entities and relationships Some constraints are expressible but no significant rules (such as inheritance) No explicit behavior Standard language is SQL.

Object-Oriented Schemas
Emerging technology for communication between software components Declarative specifications Constraints and some rules Several ways to specify behavior The Unified Modeling Language (UML) is the standard OO modeling language.

Pathw ay name : string 1..1

consists of input

1..* 1..*

0..* 2..* Reaction 0..* description : string 1..1

Chemical name : string formula : string w eight : number

output

catalyzed by

0..1

Enzyme sequence : string

Logic
Very expressive but very difficult to use. Not designed for communication. Most logical languages are not based on entities and relationships. Very powerful inferencing capabilities. Do not usually have any associated behavior. Many examples: Prolog, KIF, Slang, ...

XML DTDs and XML Schema

Defines a hierarchical document type. XML Schema defines data types. Designed for communication over the Web. Good support for entities and hierarchical relationships; awkward for others. Constraints can be imposed on the hierarchical structure and on data types. Behavior can be specified procedurally.

Knowledge Representations
Very well developed branch of AI. Many tools, but mostly academic. Not yet used for communication over the Web. Powerful language for specifying entities and their relationships. Most are linked with inference engines. Behavior is typically handled in an ad hoc manner.

RDF and DAML

Resource Description Framework (RDF) is a knowledge representation language represented in XML. It is a WWW Consortium Recommendation. The DARPA Agent Markup Language (DAML) is an extension of RDF to serve as the basis for ontology-based computing over the Web: the Semantic Web.

Ontological Reasoning in RDF

Class
type Person type Wendy type owns type domain Wanda type Fish range owns

Property
type

Type constraint violation: The range of owns is Fish. OR There is no inconsistency: Wanda is a fish! Mermaid?

DAML
type

Class
type Student

Property
type domain

College
type type

range majors subClassOf onProperty maxCardinality

type Restriction 1

Engineering
equivalentTo majors type George Arts & Sciences majors

Cardinality constraint violation: George cant have two majors OR There is no inconsistency: Engineering = Arts & Sciences

Representing information
Relational database: records OO database: objects and links Logic: facts XML: documents Knowledge Representations: annotations

All of these are graph structures: entities related to other entities by relationships.

Where is the meaning?

Databases: select-project-join queries Logic: rules determined by unification XML: XSLT patterns Knowledge Representations: templates

All of these are forms of graph matching. The units of meaning are small connected subgraphs that I call motifs.

Ontology Infrastructure
Simply introducing a language is not enough. There must be an infrastructure to support ontology-based computing, including: Ontology development tools Content creation systems Storage and retrieval systems Ontology reasoning, mediation, ... Integration with applications

Ontology Development
Ontologies can be developed using graphical tools specifically for ontologies or by adapting existing tools such as CASE tools. Testing ontologies is not easy because they include constraints and inference rules. Ontology testing is analogous to type checking in programming languages.

Content Creation
Databases: Data warehousing technology Text: Natural Language Processing (NLP) Image processing Direct creation of content

No matter how the content is created it must be tested using consistency checking.

Storage and Retrieval

Scaling up will require high-performance, distributed storage and indexing technology. The natural units for indexing are the motifs (precomputed joins), but the number of motifs is large. Jarg Corporation has developed a scalable, high-performance indexing technology for ontology-based knowledge representations.

Jarg Architecture
Document NLP Knowledge Representation fragmentation Knowledge Fragments Distributed Index Engine Knowledge Motifs fragmentation Knowledge Representation Matching Documents

Query

NLP

Conclusion
Ontology-based computing is emerging as a natural evolution of existing technologies to cope with the information onslaught. Ontology-based technology must be scalable if it is to contribute to the solution rather than add to the problem. Consistency checking is important for the development of ontologies and content.

Bibliography
Semantic Web: www.w3.org/2001/sw Ontologies: www.ontology.org Unified Modeling Language: www.omg.org/uml Knowledge Interchange Format: logic.stanford.edu/kif Specware and Slang: www.kestrel.edu XML and XML Schema: www.w3.org/xml RDF and RDFS: www.w3.org/rdf DAML: www.daml.org Notation 3: www.w3.org/DesignIssues/Notation3.html Consistency checking: vis.home.mindspring.com Jarg Knowledge Engine: www.jarg.com