Toward Plot Units: Automatic Affect State Analysis

Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, pages 17–25, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Toward Plot Units: Automatic Affect State Analysis Amit Goyal and Ellen Riloff and Hal Daume III and Nathan Gilbert School of Computing University of Utah Salt Lake City, UT 84112 {amitg,riloff,hal,ngilbert}@cs.utah.edu Abstract We present a system called AESOP that au- tomatically produces affect states associated with characters in a story. This research repre- sents a first step toward the automatic genera- tion of plot unit structures from text. AESOP incorporates several existing sentiment analy- sis tools and lexicons to evaluate the effective- ness of current sentiment technology on this task. AESOP also includes two novel compo- nents: a method for acquiring patient polar- ity verbs, which impart negative affect on their patients, and affect projection rules to propa- gate affect tags from surrounding words onto the characters in the story. We evaluate AE- SOP on a small collection of fables. 1 Introduction In the 1980s, plot units (Lehnert, 1981) were pro- posed as a knowledge structure for representing nar- rative stories and generating summaries. Plot units are fundamentally different from the story represen- tations that preceded them because they focus on the emotional states and tensions between characters as the driving force behind interesting plots and cohe- sive stories. Plot units were used in narrative sum- marization studies, both in computer science and psychology (Lehnert et al., 1981), but the compu- tational models of plot units relied on tremendous amounts of manual knowledge engineering. Given the recent swell of activity in automated methods for sentiment analysis, we embarked on a project to see whether current techniques could auto- matically detect the affect states needed for plot unit analysis. Plot units are complex structures that in- clude affect states, causal links, and cross-character links, and generating complete plot unit structures is beyond the scope of this work. As an initial step to- ward the long-term goal of automatically generating plot units, we began by creating a system to automat- ically identify the affect states associated with char- acters. An affect state represents the emotional state of a character, based on their perspective of events in the story. Plots units include three types of af- fect states: positive (+) states, negative (-) states, and mental (M) states that have neutral emotion (these are often associated with plans and goals). Our system, called AESOP, pulls together a va- riety of existing technologies in sentiment analy- sis to automatically identify words and phrases that have positive/negative polarity or that correspond to speech acts (for mental states). However, we needed to develop a method to automatically map these affect tags onto characters in the story.1 To address this issue, we created affect projection rules that propagate affect tags from words and phrases to characters in the story via syntactic relations. During the course of our research, we came to ap- preciate that affect states, of …

Content truncated to 500 words