Classical algorithms for query optimization presuppose the absence of inconsistencies or uncertainties in the database and exploit only valid semantic knowledge provided, e.g., by integrity constraints. Data inconsistency or uncertainty, however, is a widespread critical issue in ordinary databases: total integrity is often, in fact, an unrealistic assumption and violations to integrity constraints may be introduced in several ways. In this report we present an approach for semantic query optimization that, differently from the traditional ones, relies on not necessarily valid semantic knowledge, e.g., provided by violated or soft integrity constraints, or induced by applying data mining techniques. Query optimization that leverages invalid semantic knowledge cannot guarantee the semantic equivalence between the original user's query and its rewriting: thus a query optimized by our approach yields approximate answers that can be provided to the users whenever fast but possibly partial responses are required. Also, we evaluate the impact of use of invalid semantic knowledge in the rewriting of a query by computing a measure of the quality of the answer returned to the user, and we rely on the recent theory of Belief Logic Programming to deal with the presence of possible correlation in the semantic knowledge used in the rewriting.