Computational Models of Narrative

Abstract: Narratives convey causal, temporal, and other kinds of relational knowledge, the sort of knowledge that comprises expertise. Comparing narratives is one of the quickest and most powerful ways to develop expert knowledge, as indicated by the analogy literature. Analogy provides a means for identifying commonalities, and in particular, structural commonalities, and so provides a basis for revealing narrative structure. A further insight is the pairing of repetition with contrast to form a plot structure that serves as a recipe for surprise. Folktales (e.g., the three billy goats gruff), jokes (e.g., three guys walk into a bar…), advertisements (e.g., MasterCard’s priceless campaign), and other types of narratives frequently make use of these “repetition-break” plots. The repetition in the narrative structure leads audiences to draw comparisons, perhaps learning something, and forming expectations that more similar items will follow. The break or contrast in the narrative structure surprises audiences by deviating from their expectations. Narratives with repetition-break plots are prevalent, often well-liked, often socially-selected both by popular attention and expert judges, and capable of influencing audiences towards adopting the narrative’s views. This is just one example of the potential for narrative structure to yield significant returns.

Abstract: Gerrig will outline a participatory perspective on readers’ experiences of narratives. He proposes that readers encode types of mental contents (called participatory responses) that fall beyond the ordinary scope of computational models of narrative. Readers, for example, encode responses toward characters actions and preferences for particular outcomes. Those participatory responses vary from reader to reader and structure their individual experiences toward narratives.

Abstract: Humans have an astonishing ability to infer different facets of narrative structure from a description of events. These facets include the representation of the temporal order of events as well as the motivation behind the actions and reactions of agents based on their goals and beliefs. Computational representations of narrative time and plot can be assessed in terms of their naturalness for humans to infer as well as their computability. Such an assessment suggests a number of psychological investigations that could help provide constraints on formal aspects of these representations.

Abstract: The former Google CEO’s Eric Schmidt’s dictum “In God we trust—all others bring data” epitomizes the neo-positivist underpinning of ‘big data’ approaches to complex phenomena. Search engine algorithms aim at representing such phenomena in terms of mathematical and statistical phenomena which by-pass human intelligence. Meister’s talk aims to reinterpret what looks like a purely methodological decision as an ethically problematic choice that is based on a reductionist concept of intelligent behavior. Meister will use examples from narrative to illustrate and emphasize his points.

Abstract: A range of methods for the automatic construction of narratives have been developed in the last 10 years, many of them based on automatic planning methods drawn from AI. While these techniques have benefited from extensions to the standard knowledge representations that target interesting aspects of narrative structure, they often leave out the role of narrative at the discourse level. These models speak to only part of the functional properties of narrative as a result. In this talk, Young will point to these limitations in his own work and describe new models being developed that focus on narrative as a means to prompt comprehension on the part of the reader/viewer/player. These new models attempt to produce narrative story and discourse in text, video and video games that create experiences for their users rather than focus solely on the creation of appropriate narrative structural elements.