Predicting Visual Spatial Relations in the Maltese Language

Abstract

In this paper, the automatic detection of spatial prepositions between objects depicted in an image is studied. The ultimate aim is to incorporate the findings in a system that automatically describes images in a natural language. Whereas the explicit prediction of spatial relations in images has been previously studied in English and French, the work reported in this paper addresses relations in Maltese, an understudied language. A dataset consisting of images, spatial prepositions in Maltese, and a number of geometrical and language features is assembled from previous works. A number of predictive models are developed and the results are evaluated in terms of agreement with human-selected prepositions. The relative importance of the features in predicting the various relations is discussed and the paper concludes with a discussion on future work.