Inspired by my colleague Neil Cohn's review of his publications in 2016, I thought it nice to reflect on what research I have published alongside other colleagues over the past year. I'm happy this year was very productive in terms of publications, and happier that it will culminate in my PhD and will continue in my future academic position (did I mention that I am on the job market?) These papers are ordered by publication date:
- Question Answering in the Context of Stories Generated by Computers - This paper culminates a long research effort designed to establish a link between Art Graesser's Question answering-based Model of Story Comprehension and automated planning data structures. Previous research found that the same automated planning knowledge representation that can be used to procedurally generate stories can be mapped onto a directed graph (called a QUEST graph), which itself can be used to predict how people answer 5W/1H questions about the story. Thomas Price, David R. Winer, R. Michael Young and I found that different mappings contribute different prediction accuracies. For instance, certain mappings privilege "Why" questions; others privilege "How". What's exciting about establishing this link between planning and cognition is that we can potentially traverse it the other way around: we can use a desired cognitive state as the target of the planning algorithm!
- The Mimesis Effect: The Effect of Roles on Player Choice in Interactive Narrative Role-Playing Games - What originally started as a Research Experience for Undergraduates project within Intelligent and Interactive Media, flourished into this really interesting study on how narrative structure (in this case, narrative roles) affects decision-making within a single-player interactive narrative role-playing game. Ignacio X. Domínguez, James K. Vance, David L. Roberts, and I found evidence for a "Mimesis Effect". This effect has two primary manifestations: (1) if you are given a role, you adhere to it (which shouldn't be too surprising in light of social desirability and consistency behavior), and (2) (and perhaps more interestingly) if you are aware of available roles, you select one and play to it without existing prompting or coaxing. It illustrates how we as players use game affordances in a way that shapes our behavior, and in particular the strength of narrative structure in predicting it. This line of inquiry opens up several more questions: which roles do we naturally gravitate to and why? how are roles distributed/negotiated in the presence of more than one player? I hope to continue this line of work, but it has already shaped how I think about player cognition and inference-making. I was super happy with this work, and even happier that it got an honorable mention for best paper and also got picked up by Vice: Motherboard!
- PLOTSHOT: Generating Discourse-Constrained Stories Around Photos - This paper reports on work carried out while I was visiting Disney Research in Pittsburgh, PA, USA. Historically, story generation systems first simulate a story world and then generate the discourse for that story, which is what we as audience members consume (for example: words in a book, shots in a film). In this work, Albert Li and I present a way to introduce discourse constraints during story generation. Think of a discourse constraint as a kind of "window" into the narrative universe; by constraining the story with (for example) a picture, you are constraining the underlying story universe to exist as depicted at the moment you show it. Leveraging recent advances in knowledge representation within computer vision, we transform photos into logical constraints that a story generation system takes into account during story creation.
- Generating Abstract Comics - While in Pittsburgh, I collaborated with (recently-minted professor) Chris Martens on a project in the spirit of discourse-driven narrative generation. Our paper presents a system that Chris built, encoding a theory of comic generation she first wrote about on her blog. In trying to operationalize (i.e. computationally encode) Scott McCloud's analytical theory of comic constituents, Chris discovered that McCloud's theory is ambiguous. In attempting to derive a more syntactic generative theory, she found the output of the generator to be a bit nonsensical. Chris approached me with this, and together we tried to bring in theories of cognition that might help guide the generative process. We found and decided to use awesome work by Neil Cohn and his colleagues on a visual narrative grammar. This grammar deserves its own post, but the gist of its relevance to us is that it helped us structure the generator's output to create more coherent narrative structure. In essence, the grammar's top-down constraints are important for story structure, whereas McCloud's bottom-up reasoning helps guide the selection of subsequent discourse content.
I'm excited for what the next year will bring! All of my papers (including ones that will be officially published in 2017) are available on my publications page.