Automatic Generation of Typicality Measures for Spatial Language in Grounded Settings

In cognitive accounts of concept learning and representation three modelling approaches provide methods for assessing typicality: rulebased, prototype and exemplar models. The prototype and exemplar models both rely on calculating a weighted semantic distance to some central instance or instances. However, it is not often discussed how the central instance(s) or weights should be determined in practice. In this paper we explore how to automatically generate prototypes and typicality measures of concepts from data, introducing a prototype model and discussing and testing against various cognitive models. Following a previous pilot study, we build on the data collection methodology and have conducted a new experiment which provides a case study of spatial language for the current proposal. After providing a brief overview of cognitive accounts and computational models of spatial language, we introduce our data collection environment and study. Following this, we then introduce various models of typicality as well as our prototype model, before comparing them using the collected data and discussing the results. We conclude that our model provides significant improvement over the other given models and also discuss the improvements given by a novel inclusion of functional features in our model.