Extracting Networks of People and Places from Literary Texts

Copyright 2012 by John Lee and Chak Yan Yeung 26th Pacific Asia Conference on Language,Information and Computation pages 209–218 Extracting Networks of People and Places from Literary Texts John Lee and Chak Yan Yeung Halliday Centre for Intelligent Applications of Language Studies Department of Chinese, Translation and Linguistics City University of Hong Kong {jsylee,chayeung}@cityu.edu.hk Abstract We describe a method to automatically extract social networks from literary texts. Similar to those in prior research, nodes represent characters found in the texts; edges connect them to other characters with whom they interact, and also display sentences describing their interactions. Furthermore, other nodes encode places and are connected to characters who were active there. Thus, these networks present an overview of the “who”, “what”, and “where” in large text corpora, visualizing associations between people and places. 1 Introduction To fully understand a matter, one must be able to answer, as it were, the “Five W” questions: who, what, where, when, and why. In Humanities research, scholars comb texts to answer similar questions --- who the principal figures were, with whom they interacted, what they did, where and when they lived, and why they made an impact. The vast amount of texts available in digital libraries has, on the one hand, enlarged the breadth on which scholars can perform textual research (Crane, 2006); on the other hand, the sheer volume overwhelms an individual’s ability to read the texts in depth to answer these questions. Overviews — information abstracted from a collection of texts — can help a reader rapidly grasp the scope and nature of the collection in question (Greene et al., 2000), thereby supporting “distant reading” of large text corpora (Moretti, 1999). Ideally, they should also serve as gateways to the primary source by helping the reader locate points of interest for closer reading. Manually written overviews tend to be centered on one of the W’s. For example, biographies summarize the “who” in a text; a plot précis explains the “what” of a novel; and a gazetteer gives a list of locations. Most approaches in computational linguistics also focused on each of the W’s in isolation. Named entity recognition systems retrieve lists of personal entities, organizations, geographical names, and the like (Chinchor et al., 1999); temporal resolution systems detect temporal expressions (Mani and Wilson, 2000); discourse parsers can help answer why questions (Marcu, 1998). In more recent work, there has been much effort to synthesize two or more of the W’s, for example, detecting co-occurrences of dates and place names (Smith, 2002); linking time to events (Pustejovsky et al., 2005); connecting people to the events in which they interact with others (Doddington et al., 2004; Agarwal et al., 2010); as well as “nexus points” of groups of people at particular locations (Bingenheimer et al., 2009). This paper contributes another step in this direction, reporting the first attempt to automatically construct social networks from literary texts integrating who, what, and where. The rest of the paper is organized as follows. The next section reviews …

Content truncated to 500 words