The Role of Attention in Understanding Spatial Expressions under the Distractor Condition

Abstract

To develop a computational model of understanding spatial expressions, various factors should be taken into account. We have been exploring the relations between the goodness-of-fit of spatial terms and various geometric factors such as the object's size, the distance between objects and the observers' viewpoint. Although the dual-object relation between the located and reference objects can be handled with relatively simple models, introducing a distractor object requires a model considering further factors to explain relations, such as attention to the objects. Based on our experiment using Japanese topological and projective terms, this paper proposes a computational model to estimate the goodness-of-fit of spatial terms which incorporates an attention model for a distractor object. The proposed model was evaluated by using our experimental data.