Reasoning about scene descriptions

When a scene is described by means of natural language sentences, many details are usually omitted, because they are not in the focus of the conversation. Moreover, natural language is not the best tool to define precisely positions and spatial relationships. The process of interpreting ambiguous statements and inferring missing details involves many types of knowledge, from linguistics to physics. This paper is mainly concerned with the problem of modeling the process of understanding descriptions of static scenes. The specific topics covered by this work are the analysis of the meaning of spatial prepositions, the problem of the reference system and dimensionality, the activation of expectations about unmentioned objects, the role of default knowledge about object positions and its integration with contextual information sources, and the problem of space representation. The issue of understanding dynamic scenes descriptions is briefly approached in the last section.