Automatically Producing Plot Unit Representations for Narrative Text

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 77–86, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Automatically Producing Plot Unit Representations for Narrative Text Amit Goyal Dept. of Computer Science University of Maryland College Park, MD 20742 amit@umiacs.umd.edu Ellen Riloff School of Computing University of Utah Salt Lake City, UT 84112 riloff@cs.utah.edu Hal Daum´e III Dept. of Computer Science University of Maryland College Park, MD 20742 hal@umiacs.umd.edu Abstract In the 1980s, plot units were proposed as a conceptual knowledge structure for represent- ing and summarizing narrative stories. Our research explores whether current NLP tech- nology can be used to automatically produce plot unit representations for narrative text. We create a system called AESOP that exploits a variety of existing resources to identify af- fect states and applies “projection rules” to map the affect states onto the characters in a story. We also use corpus-based techniques to generate a new type of affect knowledge base: verbs that impart positive or negative states onto their patients (e.g., being eaten is an undesirable state, but being fed is a desir- able state). We harvest these “patient polar- ity verbs” from a Web corpus using two tech- niques: co-occurrence with Evil/Kind Agent patterns, and bootstrapping over conjunctions of verbs. We evaluate the plot unit representa- tions produced by our system on a small col- lection of Aesop’s 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 affect states of characters and the tensions between them as the driving force behind interesting and co- hesive stories. Plot units were used in narrative sum- marization studies, both in computer science and psychology (Lehnert et al., 1981), but previous com- putational models of plot units relied on tremendous amounts of manual knowledge engineering. The last few decades have seen tremendous ad- vances in NLP and the emergence of many resources that could be useful for plot unit analysis. So we em- barked on a project to see whether plot unit repre- sentations can be generated automatically using cur- rent NLP technology. We created a system called AESOP that uses a variety of resources to iden- tify words that correspond to positive, negative, and mental affect states. AESOP uses affect projection rules to map the affect states onto the characters in the story based on verb argument structure. Addi- tionally, affect states are inferred based on syntactic properties, and causal and cross-character links are created using simple heuristics. Affect states often arise from actions that produce good or bad states for the character that is acted upon. For example, “the cat ate the mouse” pro- duces a negative state for the mouse because being eaten is bad. Similarly, “the man fed the dog” pro- duces a positive state for the dog because being fed is …

Content truncated to 500 words