Similarity-Connections between Natural Language and Spatial Situations

Abstract. Humans are able to describe specific spatial situations in various ways. The choice of particular descriptions is influenced by adequacy, general conventions, speaker preferences and alignments, dialogue strategies, and spa-tial circumstances. Such a description of a specific scene, however, can express different meanings in changing situations, i.e., is context-dependent. We are therefore aiming at a general framework that provides formal represen-tations of (i) the semantics of utterances, (ii) spatial situations, and (iii) their interrelationships determined by additional (contextual) information. In order to interpret natural language descriptions with respect to a specific context, this framework specifies ‘modules ’ for various aspects, and then defines similarity-based relations between them. Natural language describes spatial environments in a flexible way: within one de-scription, it changes fluently according to granularity, combinations of spatial relations, necessary information for specific purposes, discourse- and situation-dependent knowl-edge, functional aspects or attributes of spatial entities [Talmy, 2000; Herskovits, 1986