Showing papers by "Andrew S. Gordon published in 2015"

Creative Help: A Story Writing Assistant

Abstract: 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.

One Hundred Challenge Problems for Logical Formalizations of Commonsense Psychology.

Abstract: We present a new set of challenge problems for the logical formalization of commonsense knowledge, called TriangleCOPA. This set of one hundred problems is smaller than other recent commonsense reasoning question sets, but is unique in that it is specifically designed to support the development of logic-based commonsense theories, via two means. First, questions and potential answers are encoded in logical form using a fixed vocabulary of predicates, eliminating the need for sophisticated natural language processing pipelines. Second, the domain of the questions is tightly constrained so as to focus formalization efforts on one area of inference, namely the commonsense reasoning that people do about human psychology. We describe the authoring methodology used to create this problem set, and our analysis of the scope of requisite commonsense knowledge. We then show an example of how problems can be solved using an implementation of weighted abduction.

Tackling Benchmark Problems of Commonsense Reasoning.

Abstract: There is increasing interest in the field of automated commonsense reasoning to find real world benchmarks to challenge and to further develop reasoning systems. One interesting example is the Triangle Choice of Plausible Alternatives (Triangle-COPA), which is a set of problems presented in first-order logic. The setting of these problems stems from the famous Heider-Simmel film used in early experiments in social psychology. This paper illustrates with two logical approaches— abductive logic programming and deonitc logic—how these problems can be solved. Furthermore, we propose an idea of how to use background knowledge to support the reasoning process.

Insights on Privacy and Ethics from the Web's Most Prolific Storytellers

Abstract: An analysis of narratives in English-language weblogs reveals a unique population of individuals who post personal stories with extraordinarily high frequency over extremely long periods of time. This population includes people who have posted personal narratives everyday for more than eight years. In this paper we describe our investigation of this interesting subset of web users, where we conducted ethnographic, face-to-face interviews with a sample of these bloggers (n = 11). Our findings shed light on a culture of public documentation of private life, and provide insight into these bloggers' motivations, interactions with their readers, honesty, and thoughts on research that utilizes their data. We discuss the ethical implications for researchers working with web data, and speak to the relationship between large social media datasets and the real people behind them.

An Integrated Evaluation of Perception, Interpretation, and Narration

Abstract: In this paper, we describe our efforts to create an evaluation tool to aid in the development of artificial intelligence systems that integrate perception, reasoning, and language abilities. Based on an early and influential study by social psychologists Fritz Heider and Marianne Simmel, we created 100 short movies depicting the motions of two triangles and a circle around a box with a hinged opening. For each movie, we provide quantitative information about each object's trajectory, a formal description of the actions that can be perceived in each object's behavior, a formal interpretation of the social situation that is depicted, and a short English narration of the interpreted events. Perception, Interpretation, and Narration In an early and influential study, social psychologists Fritz Heider and Marianne Simmel (1944) presented subjects with a short animated film depicting the motions of two triangles and a circle around a box with a hinged opening. Asked to describe what they saw, subjects responded with coherent narratives of the social interactions among three characters, with interpretations of the motivations, intentions, and emotions of the anthropomorphized shapes. From an artificial intelligence perspective, these subjects engaged in a particularly interesting suite of cognitive tasks, effortless for people but challenging for today's technologies. What would it take to build an artificial intelligence system that could watch a movie in the style of Heider and Simmel, and generate a textual narrative indistinguishable from one written by a human? We imagine that successfully building such a system would require the integration of three emerging technologies from three different areas of artificial intelligence research. First is the problem of perception. The system must recognize the sequence of actions intended by the animator in the trajectories of abstract shapes. In the original HeiderSimmel film, the larger of the two triangles jabs at the Copyright © 2015, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. smaller one, pushing it up against the side of the box. The circle and the smaller triangle kiss and dance after being separated for some time. Procedurally, the task is one of segmentation and labeling of time-series data, and could be approached in much the same manner as seen in pen-based handwriting recognition systems and video-based gesture recognition systems. Second is the problem of interpretation. The system must explain the observed actions in anthropomorphic terms, ascribing motivations, intentions, and emotions to each of the characters. In the original Heider-Simmel film, the circle cowers in fear as the larger triangle corners it inside the box. The larger triangle becomes furious when the two other character escape, and destroys the box in a fit of rage. Procedurally, we see this interpretation task as one of logical abduction, where a knowledgebase of human sociology and psychology is used to find a parsimonious explanation of the observed actions. Third is the problem of narration. The system must generate a coherent discourse to communicate its interpretation of the perceived actions in a natural, humanlike style. By Heider and Simmel's report (1944), subjects authored narratives rich with mentalistic phrases: the girl hesitates, she doesn't want to be with the first man, the girl gets worried, is still weak from his efforts to open the door, they finally elude him and get away, he is blinded by rage and frustration. Procedurally, the task is that of discourse generation, where the relational structure of the interpretation informs the discourse structure, and the semantics of the interpretation inform the lexical choices. An Integrated Evaluation Individually, each of the tasks of perception, interpretation, and narration pose difficult artificial intelligence challenges, although none seem insurmountable given recent progress in their respective research areas. However, the integrated task presents new challenges and new questions. Recent work in segmentation and labeling of time-series data has focused on supervised approaches that output probability distributions. How are these distributions to be successfully integrated with the formalisms used in logical abduction? Moreover, how can the results of the interpretation process be used to adjust these probability distributions, such that the interpretation influences what is perceived? Recent work on discourse generation in computational linguistics has pursued a strongly statistical approach, realizing the most probable surface text given the underlying representation. How is this process changed when the underlying representations are structural interpretations, containing both informative and commonsense assertions? To tackle both the individual and integrated research challenges of perception, interpretation, and narration, we developed a new evaluation tool based on the original Heider-Simmel film. This evaluation tool consists of 100 short movies in the style of Heider and Simmel. To support the development of an integrated processing pipeline, this collection includes gold-standard annotations for each of the intermediate processes needed to translate these movies into coherent textual narratives, described below.