Melissa Roemmele

Creative Help: A Story Writing Assistant

We present Creative Help, an application that helps writers by generating suggestions for the next sentence in a story as it being written. Users can modify or delete suggestions according to their own vision of the unfolding narrative. The application tracks users’ changes to suggestions in order to measure their perceived helpfulness to the story, with fewer edits indicating more helpful suggestions. We demonstrate how the edit distance between a suggestion and its resulting modification can be used to comparatively evaluate different models for generating suggestions. We describe a generation model that uses case-based reasoning to find relevant suggestions from a large corpus of stories. The application shows that this model generates suggestions that are more helpful than randomly selected suggestions at a level of marginal statistical significance. By giving users control over the generated content, Creative Help provides a new opportunity in open-domain interactive storytelling.

Triangle charades: a data-collection game for recognizing actions in motion trajectories

Humans have a remarkable tendency to anthropomorphize moving objects, ascribing to them intentions and emotions as if they were human. Early social psychology research demonstrated that animated film clips depicting the movements of simple geometric shapes could elicit rich interpretations of intentional behavior from viewers. In attempting to model this reasoning process in software, we first address the problem of automatically recognizing humanlike actions in the trajectories of moving shapes. There are two main difficulties. First, there is no defined vocabulary of actions that are recognizable to people from motion trajectories. Second, in order for an automated system to learn actions from motion trajectories using machine-learning techniques, a vast amount of these action- trajectory pairs is needed as training data. This paper describes an approach to data collection that resolves both of these problems. In a web-based game, called Triangle Charades, players create motion trajectories for actions by animating a triangle to depict those actions. Other players view these animations and guess the action they depict. An action is considered recognizable if players can correctly guess it from animations. To move towards defining a controlled vocabulary and collecting a large dataset, we conducted a pilot study in which 87 users played Triangle Charades. Based on this data, we computed a simple metric for action recognizability. Scores on this metric formed a gradual linear pattern, suggesting there is no clear cutoff for determining if an action is recognizable from motion data. These initial results demonstrate the advantages of using a game to collect data for this action recognition task.

An Authoring Tool for Movies in the Style of Heider and Simmel

Seventy years ago, psychologists Fritz Heider and Marianne Simmel described an influential study of the perception of intention, where a simple movie of animated geometric shapes evoked in their subjects rich narrative interpretations involving their psychology and social relationships. In this paper, we describe the Heider-Simmel Interactive Theater, a web application that allows authors to create their own movies in the style of Heider and Simmel’s original film, and associate with them a textual description of their narrative intentions. We describe an evaluation of our authoring tool in a classroom of 10th grade students, and an analysis of the movies and textual narratives that they created. Our results provide strong evidence that the authors of these films, as well as Heider and Simmel by extension, intended to convey narratives that are rich with social, cognitive, and emotional concerns.

Content-based similarity measures of weblog authors

With recent research interest in the confounding roles of homophily and contagion in studies of social influence, there is a strong need for reliable content-based measures of the similarity between people. In this paper, we investigate the use of text similarity measures as a way of predicting the similarity of prolific weblog authors. We describe a novel method of collecting human judgments of overall similarity between two authors, as well as demographic, political, cultural, religious, values, hobbies/interests, personality, and writing style similarity. We then apply a range of automated textual similarity measures based on word frequency counts, and calculate their statistical correlation with human judgments. Our findings indicate that commonly used text similarity measures do not correlate well with human judgments of author similarity. However, various measures that pay special attention to personal pronouns and their context correlate significantly with different facets of similarity.

Effective Scenario Designs for Free-Text Interactive Fiction

Free-text interactive fiction allows players to narrate the actions of protagonists via natural language input, which are automatically directed to appropriate storyline outcomes using natural language processing techniques. We describe an authoring platform called the Data-driven Interactive Narrative Engine (DINE), which supports free-text interactive fiction by connecting player input to authored outcomes using unsupervised text classification techniques based on text corpus statistics. We hypothesize that the coherence of the interaction, as judged by the players of a DINE scenario, is dependent on specific design choices made by the author. We describe three empirical experiments with crowdsourced subjects to investigate how authoring choices impacted the coherence of the interaction, finding that scenario design and writing style can predict significant differences.

Natural-language Interactive Narratives in Imaginal Exposure Therapy for Obsessive-Compulsive Disorder

Obsessive-compulsive disorder (OCD) is an anxiety-based disorder that affects around 2.5% of the population. A common treatment for OCD is exposure therapy, where the patient repeatedly confronts a feared experience, which has the long-term effect of decreasing their anxiety. Some exposures consist of reading and writing stories about an imagined anxiety-provoking scenario. In this paper, we present a technology that enables patients to interactively contribute to exposure stories by supplying natural language input (typed or spoken) that advances a scenario. This interactivity could potentially increase the patient’s sense of immersion in an exposure and contribute to its success. We introduce the NLP task behind processing inputs to predict new events in the scenario, and describe our initial approach. We then illustrate the future possibility of this work with an example of an exposure scenario authored with our application.