Measuring Narrative Cohesion: A Five Variables Approach

In this paper we present a five variable approach to measuring narrative cohesion. Increasingly narratives are dynamically adapted for presentation to enhance personalisation or fit different presentational objectives. Narrative generation systems seek to either automatically generate stories from scratch or, create stories from predefined conditions. With the rise of machines as co-authors it is important to consider what the affect is upon the cohesion of the narratives represented or created in this way. Typically, in existing work, this is limited to an analysis of the use of textual language within the narrative to communicate a coherent message. However we find that beyond linguistic connections narrative elements are coherently bound together through other concepts and structures such as themes, genre, narrator, and style. We present these variables, and features that may be used to identify their presence, as an alternative approach to measuring narrative cohesion and demonstrate their application on two generated narratives.