June 18, 2005
I spent Thursday at the ACH/ALLC conference in Victoria. I was invited to participate in the panel Story Generation: Models and Approaches for the Generation of Literary Artifacts, organized by Jan Christoph Meister and Birte Loenneker. The panel consisted of three presentations: Chris and Birte with “Dream On: Designing the Ideal Story Generator Algorithm”, Federico Peinado (whom I met at TIDSE last summer) with “A Generative and Case-Based Implementation of Proppian Morphology”, and myself with “Beyond Story Graphs: Story Management in Game Worlds”. Chris and Birte define paper-and-pencil story generation architectures with the aim of pushing on structuralist narratology. The goal of the work is to integrate various narratological theories, reveal where these theories are underspecified (their architectures are much more detailed than narratological theories expressed in natural language), and push narratology in new directions. Reminds me of some of Marie Laure-Ryan’s work, particularly in Possible Words, Artificial Intelligence and Narrative Theory. Birte coined the term “computational narratology” (has a nice ring to it) to describe this work. Fernando, a Ph.D. student working with Pablos Gervas (who has himself done work in poetry generation), described a case-based story generator based on Propp’s story functions. Given an initial user query specifying the story functions that should appear in the story, the system recalls the most similar story from its case base and performs generate-and-test on the retrieved case. This consists of randomly tweaking the story (performing story function substitutions) many times, stopping when a story is found that both includes the functions requested by the user and satisfies constraints captured by the ontology. He is starting a project with Birte to implement within his system the architectural theory she and Chris have developed for discourse-level manipulation (e.g. flashbacks, flash forward). Finally, I talked about what happens when generation is combined with real-time interactivity, presented story management as a far more scaleable and robust alternative to story graphs, described the author-centric viewpoint that infuses my approach to Expressive AI (I don’t care about automation for automation’s sake, but about building architectures with powerful authorial affordances), and gave an overview and comparison of both the beat-based drama manager used in Facade and the search-based drama manager proposed and Bates and Weyhrauch and recently revived in my own work (more on this in a later post).
I spent the rest of the day talking with Chris, Birte and Federico; it was great to spend the day talking about story generation and drama management, and plotting possible future collaborations. They had a chance to play Facade; Chris is very interested in the narratological implications (as the readers of this blog are painfully aware, narratology does not traditionally discuss interaction).
I’m now at DIGRA (Friday through Sunday) and so missed the rest of ACH/ALLC. Since I didn’t attend any sessions (well, part of one session on the social and archival issues involved in creating successful, free, online humanities journals) and am not a member of the humanities computing community (at least, not the definition of humanities computing that seems to be held by ACH/ALLC), I didn’t come away with a feel for the conference. There were some sessions on text analysis (e.g. author attribution, stylistics), some sessions on humanities computing education (doesn’t seem to include procedural literacy; very tool-centric), and a ton of work on meta-data (XML, etc.). Our story-generation session was definitely an outlier. There are a couple of sessions on natural language processing and computational linguistics that I’m sorry to miss.
I ran into Noah’s friends David and Elli, who graciously hosted our GTxA confab last November in Providence. David is presenting a paper he and Noah wrote on Cardplay: A New Textual Instrument. Sorry to miss this session as well.