SRI DAML Project Intent of Work for 2002
Jerry R. Hobbs,
Grit Denker, Patrick Lincoln, David Martin,
Srini Narayanan, and Richard Waldinger
Artificial Intelligence Center
SRI International
Menlo Park, California
In our work on the DAML
project in 2002, we plan to continue the development of several key ontologies,
as well as techniques for doing
inference with ontologies, especially within the context of the DAML
experiment. We are collaborating in the development of DAML-S, a language and
ontology for the specification of services on the Web. We will continue this development, especially in the
areas of ontologies for time and security.
We will continue our work on a tool that we have developed for defining
and instantiating ontologies in DAML+OIL.
Ontology Development and Inference
Web pages have content,
capabilities, and modes of access. Therefore, key to characterizing web pages
is a development of ontologies of the structure of documents, of processes and
services, and of security, privacy and trust. One of the key elements required
in describing services is an ontology of time, and geographic ontologies and
inference are required for many applications, especially military ones. We will continue our work in all five of
these areas.
1.
The Structure of Documents: We have
sketched out a core theory of the structure of information and the structure of
documents, explicating for example various ways different parts of a document
can be related to other parts and to external entities and information. This,
for example, could make searches for photographs and maps much more
precise. We did not achieve as much as
we had hoped in 2001 in this area, since our DAML-S efforts were more urgent. In 2002, we will examine a number of web
pages and other documents within this framework and develop the ontology on
this basis. This ontology will generalize and extend the ontology we have
already implemented for publications.
2. Processes and Services: Work on DAML for Services (DAML-S), in
collaboration with other members of the "DAML-S Coalition" (DAML
researchers from BBN, CMU, Nokia, SRI, Stanford, and Yale), began with
discussions led by SRI at last February's PI meeting, and has been productive
and fruitful. In the coming year, we plan to continue with our contributions to
its evolution in at least the following ways:
· Organize regular DAML-S Coalition
telecons, maintain the DAML-S section of daml.org, and coordinate the
activities required for releasing new versions.
· Contribute to the technical
development of the profile, process, and grounding ontologies, and the supporting ontologies for time and resources. We expect to be
especially active in the work on process modeling, service grounding, time, and
resources, as we have been in 2001. In
particular, we expect to see the completion of current work on the grounding
specification, and will work toward the resolution of important questions about
how to meet the expressiveness requirements of the process model, and how to
provide a developer-friendly surface syntax for process modeling.
· Monitor and participate in W3C Web
services activities (in particular, the Web Services Architecture Working Group
and the Web Services Description Working Group), and make sure the Architecture
Working Group is aware of the potential contributions of DAML-S. (The Architecture working group charter
already calls for a liaison with our DAML-S Coalition.)
· Participate in the DAML Experiment,
by providing markup and online implementations of particular services that
support the selected scenarios.
· Work with interested third parties to sustain
interest in DAML-S, and get it into use.
The
metric we will use to measure our progress in this area is simply the number of
users and sites that adopt DAML-S for describing their services.
3. Time: An ontology of time
is essential for the description of most services and nearly all devices. In the DAML experiment, it is crucial for
stating the constraints in the Foreign Clearance Guide, where, for example,
permissions must be requested fifteen work days before the flight, and one must
know about the holidays of the country.
We have sketched out an ontology of time that we believe is
adequate. In the coming year, we would
like to set up a coalition, like the DAML-S coalition, of people building
temporal ontologies and come to an agreement about what should be in the
ontology and how it should be expressed, so that as wide as possible a user
community can be developed. We would
like to get this process started at the February PI meeting, just as we got the
DAML-S effort started last year at the February PI meeting. We would also like to link up with the
community that is coding natural language texts for temporal expressions, so
that our efforts will be mutually beneficial.
4.
Geographical Space Ontology and Inference: We will look at the problem of inference from DAML-enabled
geographical sources. We have developed
an ontology for geographical space capable of coordinating many agents to cooperate
in a common task, even though they have not been designed to work together and
may adopt different naming schemes, conventions, notations, and
representations.
The onotology is capable of dealing
with the problem that one name can describe many places on Earth. For example,
(feature populated-region palo-alto
(subdivision
california united-states))
is a logical term that describes the
town of Palo Alto, California, and distinguishes it from the sixty-odd other
places in the world called Palo Alto, including the high school.
The ontology also deals with multiple
coordinate systems and notations, such as the various ways of representing
latitude and longitude; thus
(lat-long-compass-string "22.6N" "58.6E")
and
(lat-long-real 2.26e+01 5.86e+01)
are
two names for the same point on Earth.
We have linked this theory with the
Alexandria Digital Library Gazetteer, which knows about more than four million
places on Earth. The linkage is through SNARK's procedural-attachment mechanism and the OAA agent library. Capabilities of the gazetteer agent are described
by axioms in the SNARK geographical theory; when the axiom is used in the proof
attempt, the agent is invoked, and the information it supplies can be used in
the proof. In this way, SNARK behaves
as if all the facts in the gazetteer were present as axioms in the theory.
We are also in the process of linking
this theory to the CIA World Factbook, which has recently been encoded in DAML
by Mike Dean of BBN; an OAA agent can query DAML documents. The theory is also suitable for linking with
the Foreign Clearance Guide, NASA satellite image data sites, and the Web map
servers of the OpenGIS consortium, using similar mechanisms. The combined theory is capable of finding
answers that must be inferred from more than one of these sources because no
one source has the entire answer.
Once SNARK has been made a DAML-enabled
OAA agent, it will be possible to use the theory to answer DAML-encoded
queries.
Work
in this area will have synergistic benefits with SRI's ARDA-sponsored AQUAINT
project and the NASA Intelligent Systems project and with SRI's Terravision
system. We will be able to use the
results of this research in the DAML experiment in connection with applications
involving the Foreign Clearance Guide.
The geographical reasoning will contribute to the Joint Intelligence,
Surveillance and Reconnaissance (JISR) part of RDO and the IXO mission.
5.
Security, Privacy, and Trust: Given
the increased importance of the World Wide Web for the military, business, industry, finance, education, government and
other sectors, security will play a vital role in the success of the Semantic
Web. It is essential that we have tools
and techniques in place that will enable us to store, manipulate, and process
the information on the Semantic Web in ways that meet security requirements
such as authentication, authorization, and data integrity, among others. We have proposed a core security ontology
that enables us to mark-up access control restrictions and data integrity of
web pages. We will apply this ontology to describe the security measures of
well-established military and e-business sites, access-restricted web pages,
and web content that is protected against malicious alterations.
Security
mark-up is not meaningful by itself. It is motivated by web applications that
implement the various security techniques to protect data that is exchanged in
transactions. A user can make the decision as to whether the application or web
service meets his or her security requirements based on the security mark-up. Similarly, software agents that are equipped
with a user policy can select web services on the basis of their security
annotation. For this reason, we will extend our core security ontology in two ways: constructs to express basic
user security policies and a basic trust logic that will enable reasoning about
trust among agents. We will implement the trust logic in the SNARK theorem
prover. This will be the basis for a software module that enables us to use the
security information from a web page, import it into SNARK together with a user
policy, and reason about the appropriateness of the security measures used for
the web content with respect to the given user policy. Part of the reasoning
will consist of performing cryptographic operations on web content. For this purpose,
we intend to provide an interface from SNARK to a crypto library using SNARK’s
procedural calls. This work will be
coordinated with the work on DAML-S, and we will follow the DAML-S model in
forming a coalition to get broad support and use for the security ontology and
trust logic.
Another
line of the security work at SRI will be concerned with capturing security
policies of the Joint Battlespace Infosphere in an appropriate DAML+OIL
security ontology. This work will depend on the availability of such policies
to our DAML team. We would like to provide DAML+OIL ontologies that express the
security requirements of JBI and other military users. Based on such semantic annotations, we aim
at a software module that is capable of using web content and decide its
appropriateness and trustworthiness for a given situation. This is an area where it would help us to
team with a military partner for instantiating the ontology for particular
sites, because of security issues.
6. Translation
between DAML Language and Logic: During the next year, we will continue to
work on the representation of logical formulas within the DAML syntax and the
translation between DAML and the language of the SRI theorem prover SNARK. This will enable us to translate existing ontologies
and logical theories from the SNARK language into DAML and to use SNARK to
answer queries based on DAML ontologies.
Some of this work will be assisted by a Kestrel Institute effort to
translate DAML into Slang, the language of their software development
environment Specware, because an interface already exists that translates Slang
into SNARK's notation.
DAML Plug-in for Protégé-2000
We
have implemented an initial version of a DAML plug-in for Protégé-2000. We chose Protégé-2000 for its user-friendly
and adaptable interface, its open-source license and good developer support,
and its wide acceptance among knowledge engineers in research and practice. The
DAML plug-in parses DAML+OIL specifications using the Jena API parser and
transforms the triple model into Protégé-2000 frames. Using user-definable
Protégé-2000 forms, we designed a GUI to appropriately represent DAML+OIL
specific constructs such as restrictions and logical definitions (sameClassAs,
intersectionOf, and the like). This way all editing features of Protégé-2000,
such as creating and manipulating ontologies and their instances, are made
available to DAML+OIL. Our DAML plug-in also supports a DAML+OIL export function from Protégé-2000.
However, the tool is not complete yet. The export function is lacking the
capabilities to handle instances adequately, and some of the more sophisticated
DAML+OIL features that make use of the DAML:collection parsetype are not
handled yet. Moreover, a treatment of
user-defined datatypes as well as a comprehensive treatment of XML Schema
datatypes is missing. We intend to complete our DAML plug-in with respect to
these features. We plan to make the
editor available as a stand-alone software packet as well as in the form of a service used via a web interface.
Once
the tool has full functionality with respect to DAML+OIL syntax, we will use it
to create core ontologies for the DAML Experiment. In particular, we are
interested in advancing our research
and supporting other groups in the tasks of creating core ontologies and
populating them with instances in the application areas of the Foreign
Clearance Guide, DAML services, and the Joint Battle Infosphere.
New
versions of DAML+OIL that extend the syntax (for example with a rule language)
will require us to update our DAML plug-in for Protégé-2000. We expect that
some of the other Protégé-2000 plug-in tools that provide consistency checks
and other semantic checks will come in handy when DAML+OIL is enriched in its
logical expressiveness.
We
also plan to provide an interface to SNARK, a theorem prover developed at the
Artificial Intelligence Center of SRI. An automatic translation from DAML+OIL
into SNARK will make formal verification and reasoning about ontologies much
more convenient than is now the case.
Metrics
For
a project like DAML, whose primary mission is to have an impact on the world,
the most appropriate measure is the number of users of the ontologies and tools
we develop. Good results on controlled
experiments mean nothing if no one is using DAML. Our goal will be to make the DAML+OIL Protégé plug-in widely
available and to support its use, in an effort to maximize its use in the DAML
community. Similarly, the number of
users of the DAML-S, temporal, geographical space, and security ontologies is
the most appropriate metric of their success.
An intermediate metric for ontologies is the number of teams that are
involved in their development, since wide acceptance and use is promoted by
early buy-in. An intermediate metric
for the tools we develop is how well they integrate with other tools and
contribute to an overall task, for example, in the DAML experiment, since easy
integration promotes wide use.