A semiotic approach for the generation of themed photo narratives

Abstract

A wide variety of systems could be considered 'narrative systems', either directly working towards generating rich narratives or, more frequently, because they present or handle information in a narrative context. These narratives, generated or otherwise handled, may contain themes; an essential part of the subtext of narrative communicating important concepts outside the capabilities of the literal meaning of the content and forming the thematic cohesion that aids the flow of the presented narrative. However despite this very little work has been undertaken to understand of take advantage of these themes, particularly in narrative generation where the presence of well defined themes may improve the richness of those generated narratives. In this paper we evaluate the performance of a system utilising a thematic model in order to generate simple narratives in the form of photo montages compared to a keyword based system that does not. The experiment demonstrates that the system utilising the thematic model is capable of successfully connoting themes within these narratives. It also shows that the relevance of the resulting narratives to the titles used to generate them is higher in the thematic system than those generated by the other system.