Reid Swanson

Say Anything: A Massively Collaborative Open Domain Story Writing Companion

Interactive storytelling is an interesting cross-disciplinary area that has importance in research as well as entertainment. In this paper we explore a new area of interactive storytelling that blurs the line between traditional interactive fiction and collaborative writing. We present a system where the user and computer take turns in writing sentences of a fictional narrative. Sentences contributed by the computer are selected from a collection of millions of stories extracted from Internet weblogs. By leveraging the large amounts of personal narrative content available on the web, we show that even with a simple approach our system can produce compelling stories with our users.

Say Anything: Using Textual Case-Based Reasoning to Enable Open-Domain Interactive Storytelling

We describe Say Anything, a new interactive storytelling system that collaboratively writes textual narratives with human users. Unlike previous attempts, this interactive storytelling system places no restrictions on the content or direction of the user’s contribution to the emerging storyline. In response to these contributions, the computer continues the storyline with narration that is both coherent and entertaining. This capacity for open-domain interactive storytelling is enabled by an extremely large repository of nonfiction personal stories, which is used as a knowledge base in a case-based reasoning architecture. In this article, we describe the three main components of our case-based reasoning approach: a million-item corpus of personal stories mined from internet weblogs, a case retrieval strategy that is optimized for narrative coherence, and an adaptation strategy that ensures that repurposed sentences from the case base are appropriate for the user’s emerging fiction. We describe a series of evaluations of the system’s ability to produce coherent and entertaining stories, and we compare these narratives with single-author stories posted to internet weblogs.

Argument Mining: Extracting Arguments from Online Dialogue

Online forums are now one of the primary venues for public dialogue on current social and political issues. The related corpora are often huge, covering any topic imaginable. Our aim is to use these dialogue corpora to automatically discover the semantic aspects of arguments that conversants are making across multiple dialogues on a topic. We frame this goal as consisting of two tasks: argument extraction and argument facet similarity. We focus here on the argument extraction task, and show that we can train regressors to predict the quality of extracted arguments with RRSE values as low as .73 for some topics. A secondary goal is to develop regressors that are topic independent: we report results of cross-domain training and domain-adaptation with RRSE values for several topics as low as .72, when trained on topic independent features.

Automated story capture from internet weblogs

Among the most interesting ways that people share knowledge is through the telling of stories, i.e. first-person narratives about real-life experiences. Millions of these stories appear in Internet weblogs, offering a potentially valuable resource for future knowledge management and training applications. In this paper we describe efforts to automatically capture stories from Internet weblogs by extracting them using statistical text classification techniques. We evaluate the precision and recall performance of competing approaches. We describe the large-scale application of story extraction technology to Internet weblogs, producing a corpus of stories with over a billion words.

A Comparison of Alternative Parse Tree Paths for Labeling Semantic Roles

The integration of sophisticated inference-based techniques into natural language processing applications first requires a reliable method of encoding the predicate-argument structure of the propositional content of text. Recent statistical approaches to automated predicate-argument annotation have utilized parse tree paths as predictive features, which encode the path between a verb predicate and a node in the parse tree that governs its argument. In this paper, we explore a number of alternatives for how these parse tree paths are encoded, focusing on the difference between automatically generated constituency parses and dependency parses. After describing five alternatives for encoding parse tree paths, we investigate how well each can be aligned with the argument substrings in annotated text corpora, their relative precision and recall performance, and their comparative learning curves. Results indicate that constituency parsers produce parse tree paths that can more easily be aligned to argument substrings, perform better in precision and recall, and have more favorable learning curves than those produced by a dependency parser.

Identifying Narrative Clause Types in Personal Stories

This paper describes work on automatically identifying categories of narrative clauses in personal stories written by ordinary people about their daily lives and experiences. We base our approach on Labov & Waletzky’s theory of oral narrative which categorizes narrative clauses into subtypes, such as ORIENTATION, ACTION and EVALUATION. We describe an experiment where we annotate 50 personal narratives from weblogs and experiment with methods for achieving higher annotation reliability. We use the resulting annotated corpus to train a classifier to automatically identify narrative categories, achieving a best average F-score of .658, which rises to an F-score of .767 on the cases with the highest annotator agreement. We believe the identified narrative structure will enable new types of computational analysis of narrative discourse.

Learning visual composition preferences from an annotated corpus generated through gameplay

This paper describes a game called Panorama, designed to facilitate data collection to study visual composition preferences. Design considerations for Panorama, implementation of composition rules, and data collection for an experiment to learn individual and collective preferences is described. Images taken through gameplay in Panorama are automatically scored for composition quality and contribute to a corpus of domain-specific virtual photographs annotated by visual features and scores. Scores in Panorama represent rules of good composition from photography textbooks. In the current version, Panorama scores photographs along balance, thirds alignment, symmetry, and spacing dimensions. Pairwise preference rankings are collected on images from this corpus through crowd-sourcing. Results are presented from data on relative pairwise rankings on the images to learn individual as well as general composition preferences over features annotated in Panorama images. This work seeks to extend the ability of AI systems to learn and reason about high-level aesthetic features of photographs that could be utilized for various procedural camera control and aesthetic layout algorithms in video games.

A data-driven case-based reasoning approach to interactive storytelling

In this paper we describe a data-driven interactive storytelling system similar to previous work by Gordon & Swanson. We addresses some of the problems of their system, by combining information retrieval, machine learning and natural language processing. To evaluate our system, we leverage emerging crowd-sourcing communities to collect orders of magnitude more data and show statistical improvement over their system. The end result is a computer agent capable of contributing to stories that are nearly indistinguishable form entirely human written ones to outside observers.

Rich computational model of conflict for virtual characters

Rich interactions with virtual characters in narrative-based environments can be enabled by providing characters with representation of parameters for reasoning about various types of conflict. This paper proposes a model of conflict that includes mechanics, context, and dynamics of conflict scenarios. This model extends and reconciles prior work on conflict management from various disciplines. This model complements task-oriented conflicts that are implemented in current agent architectures and seeks to motivate exploration to a new design space of possible conflict situations. This work is based on initial analysis of a corpus of conflict scenarios annotated with personality profiles and resolution strategies. This annotated corpus is made available to the community for further research on conflict.

Remember That Time? Telling Interesting Stories from Past Interactions

Sociability is a human trait that plays a central part in relationships over time. Today, humans are increasingly in long-term interactions with intelligent agents, which have proven most useful when they are sociable. Such sociability requires the agent to remember and appropriately refer to past interactions. A common way in which humans refer to their past interactions and collaborations is through storytelling. Such stories, often abbreviated, include a small set of interesting and extraordinary events. We propose the design, development and preliminary evaluation of a generic computational architecture for finding and retelling such interesting event sequences. Our system mines interesting interaction episodes in a corpus of prior interactions. Initial evaluation of interactions selected by the system for retelling are encouraging. A future goal of the research is to support collaborative composition of stories about prior interactions between humans and agents in a mixed-initiative framework to produce interesting retellings.