Acquiring Ontologies and Domain Theories

START's annotations are as powerful as the ontologies and domain theories which stand behind them. The core of these, for START, is lexical knowledge: knowledge about words and their relationships. As we indicated earlier, linguistic phenomena provide strong evidence about the correct taxonomic relationships in any representation. However, these alone do not provide all the power desired. Significant knowledge about the world is needed to help retrieve assets which are useful in solving problems. We have every intention of employing ontologies and partial domain theories produced by other HPKB participants; but our goals include supporting problem solving efforts such as intelligence analysis which involve broad (but not necessarily very deep) knowledge of the world. It will be HAWK's task to help elicit such knowledge about the world from its users. We see two main avenues to guide such acquisition:

• Volunteered information: Users will often provide not only annotations of documents but also general information about the domain discussed in the document. Such information is provided as Natural Language sentences. Taxonomic relationships, for example, can be provided by sentences like: ``Nuclear weapons are weapons of mass destruction''; other standard relationships such as part-whole, quantity restrictions etc. all have standard natural language renditions.

• Information solicited in response to breakdowns: When a user submits an annotation there is the possibility that the sentence can't be parsed; the most common cause of this in START today is missing vocabulary. When such a failure occurs, HAWK will collect whatever syntactic constraints were inferred from the failed parse and use these in a session in which it tries to obtain from the user both the lexical knowledge and the semantic knowledge needed to understand the failed sentence.