Our goal within the DAML program is to develop language specifications, tools, and applications that support the creation, location and interoperation of services both across the World Wide Web, and within multi-agent system communities. Our research aims include developing and evaluating methodologies for:

• semantic service matching, and to contrast these with current industrial developments;
• semantic interoperation between agents through middle agent services;
• agent-based planning/scheduling/execution for service decomposition, access and composition.

We have started working with and will continue working with Sheila McIlraith of Stanford KSL, Jerry Hobs (et.al.) of SRI, Mark Burstein of BBN, and Ora Lassila of Nokia on the development of a DAML based service process control language (DAML-S). This language describes: (a) the service profile/capabilities of the service, i.e. an advertisement describing what the service does, what are the input/output parameters, pre conditions and effects on resources etc, (b) the service model; this is the process model of how the service works, its composition in terms of lower level services, etc, and (c) the service grounding which describes how agents may access the service (in terms of communications protocols, service parameters etc).

While working with the rest of the group in the development of DAML-S, CMU is taking the lead in the development of the service profile part of DAML-S, and in the definition of an ontology for domain level resources (such as money, commmunication bandwidth etc) to be included in DAML-S. In addition, CMU will focus on the development of tools and applications to support the construction of service profiles, and access to high-level services, and resources such as ontologies and services that can be used to evaluate the languages/tools etc.

A more detailed description of the CMU research activities follows:

1. Processes and Services: In collaboration with SRI, BBN, Nokia & Stanford KSL, CMU has defined an initial minimal language of processes and services, DAML-S. We expect to release it for feedback in the next few weeks to the part of the DAML community interested in services for comment and feedback. Later on, the DAML-S will be released to the whole DAML research community to be used by them for describing and accessing services. The ontology consists of simple events, means for composing events into more complex events, and a simple theory of time including instants and intervals. Future layers of the ontology will address the execution status of processes, ontologies of domain-specific resources, temporal measurement, the clock and calendar, and the structure of particularly important subclasses of events. As we develop this ontology, we will use it in characterizing various web sites that we believe will push its limits. The language and the associated ontologies are going to be minimal, and can be expanded as additional requirements arise. The language and ontologies will be used in the future by web designers to annnotate web based services so that they can be automatically accessed by agents.
2. Matching Algorithms for Middle agents: CMU is taking the lead in developing matching algorithms in order to facilitate discovery and semantic interoperation of services described in the services process language. In particular, we will develop matching algorithms that can compare service requests with advertised service descriptions. The matching algorithms will:

1. Compare and reason about concepts defined as URIs using DAML. This may be as simple as determining that two concepts have corresponding equivalentTo or sameClassAs expressions, but will also identify conceptual distance based on unweighted arcs between concepts (e.g. in the case of subsumed concepts).
2. Utilize a layered approach to matching, to contrast the utility of performing broad matches (such as comparing inputs and outputs) to focused matches (checking matched pre-conditions, valid effects etc) in terms of accuracy and speed.

We will embed these algorithms within two types of middle-agents: a yellow-pages based matchmaker, and a facilitator.

3. Service Composition / Decomposition: High level service requests can be decomposed into a sequence of lower level service requests, such as decomposing the travel domain goal into booking flights, hotel rooms etc. CMU will develop workflow / planning algorithms (starting with utilizing Hierarchical Task Network planning representations) that can decompose high-level goals into workflows that can be executed by available services, and demonstrate how this workflow can change given changes in the services available.
4. Development of sample services and ontologies: As part of the evolution and evaluation of the process & services language and associated tools, CMU will develop sample services (both web-based and agent based; both domain independent and domain-dependent), such as airline services, entertainment services, hotel booking services, etc. These ontologies will include:

• Ontologies for describing different domain independent middle agent services. Such an ontology will allow agents to recognize middle agents and contact them for appropriate domain independent services. In particular, we will provide an ontology to describe the agents in the taxonomy of middle agents that we have identified and for any additional middle agents our research may identify.
• Sample ontologies to describe different domain specific agent services / capabilities, such as scheduling air missions, finding routes for military vehicles, finding weather information, making hotel reservations etc.
• Initial ontologies for the description of service parameters and request preferences, such as network latency, service availability, cost, etc. that may accompany service profiles/agent capability advertisements and requests.
• Ontologies for the description of security parameters and resources that are associated with acessing services.

Additional Collaborative Opportunities: The travel domain will provide a good opportunity for additional collaboration and evaluation with other researchers, including SRI, BBN, Nokia & Stanford KSL. We have not yet collectively decided upon a set of scenarios, but it is likely to involve the automatic composition of travel-related services, using the tools and the DAML and DAML-S encoding of web resources that various research groups provide. In particular, we envision that each of our collaborators (KSL, SRI, BBN, Nokia) will provide DAML-S-based annotations for a subset of web recources (e.g. in the travel domain). These annnotations could be used by agents developed independently by other members of our collaborators (not the annotators) to access the annotated web resources and use them to compose, for example a travel plan. We hope that, after initial experiments with our collaborators during the first year, such annotated services could be more widely used by other members of the DAML research community (not just our collaborating group).

We will release emerging ontologies and the process & services language to the DAML community as they are developed, and make available the sample services. We expect the DAML research community to be the primary audience for this work (for the first year), and expect that this community will use them to advertise and describe their services (e.g. ITTalks etc.). In addition we will deploy the middle-agent services with a basic web-based interface to allow end users to browse and monitor what services have been advertised, and generate human mediated queries. Basic web pages will also be provided for the sample services to facilitate and contrast DAML supported service composition with current, user oriented web forms. Initially, we expect the DAML community to be the audience for these tools and services but, as they become robustified, based on feedback from their use within the DAML community, we expect to put them on the web for general use.

In developing this work, we expect to be drawing upon the developments of the DAML language, and utilize DAML-based applications and tools, such as:

• DAML editors/validators for developing future ontologies;

• The DAML crawler, which we hope to modify to locate and register service descriptions with middle agents;
• DAML parsers/inferencing tools to be used by middle agents;
• DAML ontology translators;
• DAML based ontologies and ontology agents for service matching.

Towards the end of Contract Year 2, we expect to start development of authoring tools for web designers to create and validate service descriptions, and for debugging service requests. In addition, we will continue development of matching techniques, develop additional ontologies to support matching, and investigate methodologies for identifying and resolving semantic interoperation at increasing levels of task complexity.