. German Research Center for Artificial Intelligence GmbH
Talking about a location in a geometrically represented environment implies translating coordinates into suitable natural language expressions. Therefore, a reference semantics anchored in the given geometry has to be developed in order to be able to describe spatial relations between objects. Answering the question ``Where is object X?'' will lead to a large set of spatial propositions like ``X near Y'', ``X to the right of Z'' etc. Aiming to get a description of the specified location which is both easily understandable and highly informative for a human being the elements of this set must be analysed with respect to degree of applicability, compatibility, uniqueness and facility to be memorized. Furthermore, the actual context as well as presumptions of the user's knowledge must be taken into account.
As in, e.g., dialog situations it is a priori unknown how much time is given to generate an utterance we demonstrate a first approach of applying anytime algorithms which ensure the possibility of information retrieval at any moment with increasing quality according to the amount of time.