Cycorp Intent of Work:

Rapid Knowledge Formation (RKF) Technology for the DARPA Agent Markup Language (DAML)

 

Prepared by Stephen Reed

January 31, 2001

 

Goals

In the remainder of this fiscal year, Cycorp will provide three tools packaged as Semantic Web services integrated with other DAML contributors in the DAML Integrated Demonstration & Experiment, May and September releases:

1.      Ontology Interpretation Tool: interprets a DAML ontology by analyzing the lexical and semantic content of the ontology and of the web pages marked-up by that ontology, in order to derive mappings from the local ontology to semantic concepts in Cyc.  We will import into Cyc all the ONA experimental ontologies and as much of the instance content as the experiments require.  Each ONA XML namespace will be modeled in a separate microtheory and the DAML statements will be modeled using Cyc’s functional notation to associate a resource (Cyc constant) with a namespace (Cyc microtheory).  Imported DAML resources will be named according to a convention that will clearly separate different meaning senses of the same named resource.  The suite of Cyc NL parsers will produce a list of candidate matching concepts for each resource based upon a parse of the resource tag name and taxonomic context.  Once a set of initial Cyc mappings is identified, the Interpretation Tool will perform a structural comparison and correlation of the two ontological structures (the source DAML ontology and target Cyc concepts) to discover additional mappings between uncorrelated terms in the source ontology and additional target concepts in Cyc.

2.       Ontology Elaboration Tool: uses Cyc's world knowledge and common sense inference capabilities to derive additional semantic concepts closely related to concepts represented in the DAML ontology and then suggest additional terms for incorporation into the ontology. Once a concept has been isolated, the Elaboration Tool will search for semantic connections in all directions in the Cyc KB.  This will include lateral connections derived from terms sharing the same parents and/or the same children in an ontological hierarchical graph. Elaborations will also be made by commonsense inference.  The Cyc KB contains a great deal of general knowledge about human activities, including business, military, educational, and recreational activities, which it can use to infer related concepts.  These related concepts could then be suggested as additional elaborations of the original source ontology.

 

3.       Ontology Translation Tool: uses mappings between DAML ontologies and Cyc knowledge base as a reference framework for deriving a translation between the two DAML ontologies. The Ontology Translation Tool will produce a translation between the two ontologies by deriving a translation function.  This translation function will be the composition of: the mappings from the original ontology to a set of equivalent Cyc concepts (or compound expressions); the mappings from the translation target to a set of equivalent Cyc concepts (or compound expressions); and the mappings from the two sets of Cyc concepts (or expressions) to each other.  The final product will be a mapping from the source DAML ontology to the target DAML ontology that does not refer to the intermediate reference ontology, Cyc.

In addition we will release an executable version of the Cyc Upper Ontology, which will be available for downloading from www.OpenCyc.org. Cycorp will, for the first time, provide the Cyc Inference Engine and a suite of tools for creating knowledge-based applications. OpenCyc 1.0 will be released and by year-end, this software will be integrated with the Semantic Web for DAML ontology import and deductive queries.

 

DAML Integrated Demonstration & Experiment

 

Separately developed ontologies will describe an Operational Net Assessment scenario in Afghanistan.  Cycorp will help develop mappings for these user-developed, overlapping ontologies into the Cyc reference ontology.  Once tied to the Cyc, the Elaboration Tool provides additional terms for use with the source ontologies.  For example, once the user’s term for kind of vehicle is mapped to Cyc’s term #$TransportationDevice, then a wealth of more specific vehicle terms can be exported for use by other DAML project participants.  Regarding the Extended Capabilities phase of the experiment, Cyc’s Ontology Translation tool will provide a translation service for users and other DAML project participants.

 

Metrics

 

The Cyc KB and associated applications are already well instrumented for RKF user interactions.  The DAML tools will be instrumented in a similar fashion, measuring the performance of the ontology interpretation, elaboration and translation processes.  Cyc will conform to any end to end metrics gathering standards mandated by the DAML experiment manager.

 

DAML Life Cycle

 

Once end-users are empowered by DAML to create their own ontologies, there will be an urgent need to interrelate those ontologies in a useful way. Cycorp's services are targeted at the situation where the ontologies to be translated are not richly specified, where a novice has quickly created a “light weight” ontology, just to get started.  In that case, it will be necessary to interpret, elaborate, and fill-in intended meanings of the terms, and then map those meanings to better-defined semantic structures.  End-users want their locally created semantics to be interpreted globally on the Semantic Web, and Cycorp’s DAML services provide that capability.  End-users will appreciate the Cycorp’s lexical tools that employ the Cyc KB to provide intelligent parsing and interpretation of noun phrases in DAML tags; KB-based tools for automatically constituting complex concepts that are implied by the DAML tags, but are not explicitly present.

 

Ontology interpretation, translation, and elaboration services are widely useful as middleware in an application, or as a member of an agent community, which is performing tasks on behalf of end-users.  Many such tasks involve information stated in one ontology, but also involve information stated in other ontologies – for example a Semantic Web query stated in a manufacturing industry ontology, but whose answer set contains information translated from a retailing industry ontology.  Semantic Web search engines such as Haircut, are highly motivated to employ Cycorp DAML services to expand the set of search terms via translation and elaboration.  Ontology translation services can delegate to one another, portions of a request for which they may not have the entire answer.

 

The ontology discovery ability of the Cycorp DAML services is important to users because ontologies will be continually evolving and new ones coming into existence.  When a request to interpret a novel or updated ontology is received by Cycorp DAML services, the ontology (or only the relevant portions – if large) will be dynamically accessed and mapped to the Cyc reference ontology.  Then the request will be processed.  Subsequently, in an offline mode the novel ontology will be entirely mapped using time consuming methods if necessary.  Then future requests involving this ontology will be processed without the delay of dynamic interpretation.  End-users will require dynamic ontology discovery somewhere in the process chain or else the Semantic Web will grow stale and break.