Our research in the DAML program is focused on the development of concepts and tools for representing temporal information. Temporal information is ubiquitous in real world situations. It includes dates of events (“January 3, 2002,” “next Wednesday”), durations of activities (“drive for twenty minutes,” “wait about an hour”), ordering between events (“wait for Fred then drive to Rochester”), and constraints between events (“don’t touch the button while the switch is on,” “only one flight can use the runway at a time”), among other things. Our objectives in this project are (1) the development of a practical ontology and markup language suitable for representing temporal information in realistic situations and (2) the development of tools and algorithms to assist in the coding and use of temporal information.
Our approach is to build on formally well-defined and extensively-studied model of temporal representation, Interval Temporal Logic (the so-called “Allen algebra” of time). This and other work on formal temporal ontologies provides the foundation for development of the more practical ontology. This will be specifically designed to support the representation of certain types of temporal information that occur regularly and are of value to end users. The goal is a markup language that allows users to naturally indicate temporal information without being ontology or temporal reasoning experts.
We have three primary technical goals for the upcoming year. First, we intend to continue working with the DAML-Time group to refine the core mathematical ontology time presented at the Portland meeting. Second, we intend to continue the development of the practical temporal ontology, including producing “version 0” of a user’s guide to practical temporal markup and performing an evaluation of user markup of temporal information. And finally, we will begin investigating the development of automated validation tools for temporally marked up information. We will elaborate on each of these below.
We have been involved from the outset with DAML-Time group coordinated by Jerry Hobbs. In addition to the work on the axiomatization of interval time, we also produced the DAML/RDF encoding of the axioms and host the DAML-Time website. We believe that this core ontology serves two important purposes. First, it clarifies the meaning of the terms and concepts being used, providing a common language with which to discuss ideas about temporal representation. Second, it can serve as an interlingua between theories and markup languages, if these are defined in terms of extensions to the common ontology. In fact, we defined an extension to the ontology that allows it to describe the commonly used and much-studied Interval Temporal Logic (the so-called “Allen algebra”), which forms the basis of our practical temporal ontology. Refinement of the core temporal ontology is ongoing and we are coordinating changes on the website.
Based on the mathematical temporal ontology, we are developing an ontology and markup language suitable for representing temporal information in realistic situations. Our goal is to strike a balance between technical purity and practical applicability. Our initial focus is clearly on the practical side: we are developing an ontology and corresponding DAML markup elements that allows us to represent ordering and metric constraints between temporally-situated events. The focus is on the temporal aspects, not on the semantics of the events themselves. Our objective is to support markup of such things as flight schedules, logistics plans, or reports from operations in progress. Our prior experience in planning and scheduling serves as our guide for what information is practically available, as well as what information is useful as a product of the markup.
Ontologies and markup languages are only useful insofar as they allow people (and, hopefully, machines) to represent useful information in meaningful ways. We therefore intend this year to evaluate how well users can use the ontology and markup to represent the content of information from realistic sources. Users (starting with ourselves) will be given “version 0” of a user’s guide to practical temporal markup, and asked to annotate the temporal information in a set of documents such as new reports. We can quantitatively evaluate whether different users consistently represent the same information using our practical ontology using standard measures such as inter-annotator reliability (kappa). If the reliability is low, this means either than the conceptual vocabulary provided by the ontology and markup is not suitable for that task (in which case we refine the ontology/markup), or the users need more guidance in marking up the information (in which case we refine our “user’s guide” and look at different training possibilities).
One of the main reasons to use a markup language is to allow automated processing of the resulting information (e.g., for visual display, or by software agents making autonomous decisions, etc.). In the coming year, we hope to begin the development of reasoning tools that can extract or infer additional information from the marked up input. Because our approach is based on a well-understood formal model of temporal reasoning, we should be able to develop a range of tools that would be useful in practice in dealing with temporal information. As an initial example, we will try to develop a “validator” for temporally marked-up documents based on the consistency-checking procedures of the underlying temporal logic. We could use then use this to validate reports before adding them to a “situation awareness” database. Our sense is that it will be automated procedures such as these that will convince data providers to use DAML/OWL in general and our practical temporal ontology in particular.
We believe that, ultimately, users (Subject Matter Experts) should be able to use our temporal ontology and markup language to capture and share information about the DAML Experiment scenario. This information might include such things as courses of action, logistics requirements and schedules, observation and surveillance data, and so on. It is probably premature to consider doing this until we have done a few iterations of the evaluate-refine cycle described previously.
The next logical steps for our project are:
1. To continue the specification, evaluation, and refinement of both the core mathematical ontology (in conjunction with the DAML-Time group) and of our own practical temporal ontology and markup language.
2. To continue the development of useful reasoning techniques and implementations thereof to support the use of temporally marked up information. An obvious place to work on would be reasoning with metric time constraints, as well as the qualitative interval relations.
3. To begin to investigate automatic extraction of temporal information from unstructured or semi-structured sources, including natural language. This would vastly increase the amount of useful information available to the users of the markup, would make it much more timely, and would lessen the need for personnel trained to do the markup.