In the CASAM deliverable D3.3 an agent was presented that builds interpretations upon multimedia annotations by incrementally consuming analysis results as well as input from a human annotator. These interpretations are based on background knowledge of a specific domain, e.g. an environmental domain as it was exemplarily chosen for the CASAM project. As a result of the interpretation process, multiple interpretation alternatives are possible. A preference measure for the alternatives is realised by a probabilistic scoring function. As an extension of the interpretation agent, a mechanism for meta-level reasoning is presented with the aim to disambiguate interpretation alternatives. This is achieved by generating queries from a set of interpretation alternatives and stating them to the human annotator. Queries themselves are ranked by an importance value, representing the benefit of an answer to a query for the disambiguation process.

After a revision of the interpretation process the query generation mechanism is explained, followed by a detailed description of different query types, together with the format they are communicated in. Furthermore, the processing of responses to queries is addressed.