Type Theory with Records : a General Framework for Modelling Spatial Language

Cross-disciplinary research has shown that spatial language is dependent on several contextual factors, for example geometrical arrangement of the scene (Regier and Carlson, 2001), the type of objects referred to and their interaction (Coventry et al., 2001) , and alignment in dialogue (Watson et al., 2004) among others. Although the contribution of these contextual factors has been well-studied, several questions relating to the modelling and representation in the domain of computational models for situated conversational agents still remain. Some basic requirements for such a representational system are: (i) formal accuracy; (ii) the ability to represent information from different modalities; (iii) bridging perceptual and conceptual domains; (iii) adaptability and learnability of representations; (iv) formal ability to capture the meaning relations typically found in human reasoning and language. In building situated conversational agents, several systems have been proposed but none of them capture all of these requirements. For example, semiotic schemas (Roy, 2005) account for the meaning of words that refer to entities and actions it is not straightforwardly evident how they relate to other linguistic representations. (Krujiff et al., 2007) adopt a layered model with distinct representations at each layer. For example, there is a feature map corresponding to features from sensory observations, a navigation graph containing way-points, topological map of areas, and a conceptual map of an ontology of objects. Although there exist mechanisms by which these representational levels interact, the kinds of representations at each level are quite distinct from each other and are shaped by different operations. The question we would like to address is whether such representational levels and operations can be generalised by taking inspiration from the way humans assign, learn and reason with meaning.