ANSWERING QUERIES FROM CONTEXT-SENSITIVE PROBABILISTIC KNOWLEDGE BASES

We define a language for representing context-sensitive probabilistic knowledge. A knowledge base consists of a set of universally quantifie d probability sentences that include context constraints, which allow inference to be focused on only the relevant portions of the probabili stic knowledge. We provide a declarative semantics for our language. W e present a query answering procedure that takes a query Q and a set o f evidence E and constructs a Bayesian network to compute P(Q/E). The posterior probability is then computed using any of a number of Bayesi an network inference algorithms. We use the declarative semantics to p rove the query procedure sound and complete. We use concepts from logi c programming to justify our approach.