The Intelligence Community, among others, is increasingly using document metadata to improve document search and discovery on intranets and extranets. Document markup is still often incomplete, inconsistent, incorrect, and limited to keywords via HTML and XML tags. OWL promises to bring semantics to this markup to improve its machine understandability. A usable markup tool is becoming a barrier to the more widespread use of OWL markup in operational settings. This paper describes some of our attempts at building markup tools, lessons learned, and our latest markup tool, the Semantic Markup Tool (SMT). SMT uses automatic text extractors and templates to hide ontological complexity from end users and helps them quickly specify events and relationships of interest in the document. SMT automatically generates correct and consistent OWL markup. This comes at a cost to expressivity. We are evaluating SMT on several pilot semantic web efforts.1 This paper uses the terms "web" to refer to unclassified and classified intranets and extranets (as well as the World Wide Web). 2 Google and similar search engines are still the predominant tools in operational use although extensions to these are the subject of advanced technology pilot projects.

I n case-based planning (CBP), previously generated plans (cases) are stored in memory and can be reused to solve similar planning problems in the future. CBP can save considerable time over planning f i o m scratch (generative planning). C B P thus offers a potential (heuristic) mechanism for handling intractable problems. One drawback of CBP systems has been the need f o r a highly stmctured memory that requires significant domain engineering and complex memory indexing schemes to enable eficient case retrieval. I n contrast, our CBP system, CaPER, uses a massively parallel frame-based A I language ( P A R K A ) andcan do extremely fast retrieval of complex cases from a large, unindexed memory. T h e ability to do fast, frequent retrievals has many advantages: indexing is unnecessary; very large casebases can be used; and memory can be probed in numerous alternate ways, allowing more specific retrieval of stored plans that better fit the target problem u i t h less adaptation.

No abstract

No abstract

goals of the target problem. (Some systems can retrieve pieces of cases.2) Such systems can retrieve cases more efficiently by using indexing to restrict the kinds of features in a retrieval probe (thereby restricting the search to a subset of the case base), but this approach also raises several problems (see the first sidebar).We are developing the Caper case-based planner to address some of the problems of serial retrieval on an indexed case base. Caper uses the massive parallelism of the Connection Machine to quickly retrieve cases and plans from a large, unindexed memory. The system can retrieve cases and plans based on any feature of the target problem, including abstractions of target features. By controlling which features are part of the retrieval probe and their level of abstraction, a wide range of queries can be