We live in an intricate three-dimensional environment in which we must navigate, perceive our surroundings, and establish our position and the position of other objects around us. Vision plays a significant role in these tasks by providing detailed information about the spatial layout of our environment. We employ linguistic expressions to communicate the spatial knowledge we acquire from the environment to others. These linguistic expressions enable us to effectively communicate the spatial relationships of objects within the visual scene. Human language is strongly colored by our sensory-motor experiences. This interplay between visual perception and linguistic communication forms a crucial aspect of human cognition and serves as the focal point of exploration in this research endeavor.



Spatial semantics is the study of the meaning of spatial language. Spatial prepositions constitute a lexical category that encodes some of the fundamental spatial properties of the physical world. Compared to other lexical items in English, spatial prepositions are limited in number and form a closed class. Studying these words will provide insights into our internal representations of space and the interplay of space, cognition, and language.



‘Far’ and ‘near’ are two spatial prepositions in English that encode the distance along all three spatial dimensions and explicitly convey distance. Spatial prepositions are advantageous because they inherently connect to the visual scenes being depicted, allowing for more precise semantic definitions compared to numerous other linguistic expressions.



The approach to this problem is to use Deep Neural Networks as a computational metaphor for cognition and study the environmental conditions that influence the cognition of ‘far’ and ‘near’ in visual perception. We present a computational study that explores how the different characteristics of the visual world impact the learning of spatial prepositions, such as ‘far’ and ‘near’.