Future Work

  The one important aspect of functionality within Haystack that we have failed to address is the actual mechanism by which queries are made. Although there are independent methods to query each information system there is currently no uniform method to simultaneously query all of them.

Ideally, a natural language solution will one day exist that will be able to dispatch different aspects of the query to different information systems. For example, the question, ``What documents about apples that I read yesterday?'' would be split into two different questions: ``find me all documents about apple'' and ``find all documents read yesterday.'' The first question is obviously based on some semantic considerations and so it will be dispatched to an IR system. On the other hand, the other question is inherently a database type question and can be dispatched to Lore. Result combination is also an open issue. Ideally, the user should be able to scale the results of the different information systems in some way.

The second alternative for querying utilizes the Lore DataGuide model. By taking a snapshot of the current database schema and representing it graphically a user's can simply fill in different parts of the graph and the appropriate query will be generated. A special box can hold free text queries (those intended for an IR system).

Regardless of the exact query mechanism, the hybrid-search system proposed here has the potential to solve a much greater variety of problems than any independent information system.