This talk addresses three different issues that should be given more
attention when developing new systems for uncertainty management,
especially for the integration of information retrieval and database
systems. First, vague (or fuzzy) predicates abstract from specific
methods of text indexing, thus representing the appropriate logical
level for the IR component, and also can be extended to arbitrary data
types, especially multimedia. Second, probabilistic rules in
probabilistic Datalog extend reasoning over uncertain facts by
inferring uncertain statements from deterministic facts. Finally,
4-valued logic should be used when possibly contradicting facts from
different sources have to be integrated.