Generation of Humorous Situations in Cartoons through Plan-based Formalisations Marc Cavazza, Fred Charles and Steven J. Mead School of Computing and Mathematics, University of Teesside Middlesbrough TS1 3BA, United Kingdom {m.o.cavazza, f.charles, steven.j.mead}@tees.ac.uk Introduction Digital Storytelling aims at developing techniques for the automatic generation of narratives by computers, in the form of computer animations. In evaluating the quality of the stories produced, there has been some work dedicated to modelling suspense [Young, 1999] or the emotions of characters [Gratch and Mrasella, 2001]. However, virtually no work has addressed the characterisation of comic situations within these narratives and how they can be generated. The principal AI technique for digital storytelling is planning [Badler et al., 1996], as it describes the high-level behaviour of characters through action selection, and as a story can be described as a sequence of causally related actions. In this work, we attempt to characterise comic situations in digital storytelling through the planning formalisms used to generate the story. This formalisation also has the potential to gain a better understanding of some mechanisms of comedy in specific genres. In this research we use cartoons as an investigation field. Planning in Cartoons It is reasonable to think that formalisation would be initially easier to carry out in those narrative genres in which the plot itself makes explicit the plans constructed by the story’s characters to achieve their goals, and to some extent is actually based on these plans. This appears to be actually the case in several cartoons, for instance the Roadrunner™ series, and the popular Pink Panther™, which serves as a test application in this research. The description of stories as plans has been proposed in computational linguistics for story understanding [Schank and Ableson, 1977]. However, the analysis cartoons would reveal that many of them resort to dramatisation of problem solving, or the pursuit of a single goal (e.g., catching the “roadrunner”) as a main narrative mechanism. This is why, to support these experiments, we have opted for such a cartoon application, in the form of a short episode of the popular “Pink Panther” character. It appears that a typical Pink Panther cartoon involves problem-solving and goal-oriented activities, which tend to be contrasted by external factors (e.g. preventing a cake from being stolen, a room to be painted new, returning a cat to its owner, etc.). Many actual Pink Panther episodes can be analysed in terms of such planning aspects, i.e. achieving or preserving a given state. We have designed a simple scenario as a context, in which a single goal would provide a narrative drive but the intervening situations and their ordering are left open. This contrasts significantly from our previous work in interactive storytelling [Cavazza et al., 2002], in which the story would progress through pre-defined stages, story variability taking place within each stage. The scenario consists for the Pink Panther to get a quiet night’s sleep (the “sweet dreams” scenario). Many things can get in the way, and getting rid of these constitutes various planning sub-problems: in addition, there is much room for long-range dependencies between sub-tasks that are likely to generate action failure and comic situations. For instance, in our example, if we assume that the Pink Panther’s sleep goal can be contrasted by noise. The Pink Panther will have to identify various strategies for getting rid of noise, such as