A mediator is a domain-specific tool to support uniform access to multiple heterogeneous information sources and to abstract and combine data from different but related databases to gain new information. This middleware product is urgently needed for these frequently occurring tasks in a decision support environment. In order to provide a front end, a mediator usually defines a new language. If an application or a user submits a question to the mediator, it has to be decomposed into several queries to the underlying information sources. Since these sources can only be accessed using their own query language, a query translator is needed. This paper presents a new approach for implementing query translators. It supports conjunctive queries as well as negation. Care is taken to enable information sources of which processing capabilities do not allow conjunctive queries in general. Rapid implementation is guided by reusing previously prepared code. The specification of the translator is done declaratively and domain-independently.

We present an approach towards a unified modeling and query processing tool for mediatory information systems. Based upon coloured Petri nets we are able to model the integration of parametric data (external, uncertain and temporal information) and to visualize the dataflow in mediatory information systems.

Integrating data and knowledge from multiple heterogeneous sources -- like databases, knowledge bases or specific software packages -- is often required for answering certain queries. Recently, a powerful framework for defining mediated views spanning multiple knowledge bases by a set of constrained rules was proposed [24, 4, 16]. We investigate the materialization of these views by unfolding the view definition and the efficient maintenance of the resulting materialized mediated view in case of updates. Thereby, we consider two kinds of updates: updates to the view and updates to the underlying sources. For each of these two cases several efficient algorithms maintaining materialized mediated views are given. We improve on previous algorithms like the DRed algorithm [12] and introduce a new fixpoint operator WP which -- opposed to the standard fixpoint operator TP [9] -- allows us to correctly capture the update's semantics without any recomputation of the materialized view.