Interpreting Anaphoric Shell Nouns using Antecedents of Cataphoric Shell Nouns as Training Data

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 300–310, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Interpreting Anaphoric Shell Nouns using Antecedents of Cataphoric Shell Nouns as Training Data Varada Kolhatkar Department of Computer Science University of Toronto varada@cs.toronto.edu Heike Zinsmeister Institut f¨ur Germanistik Universit¨at Hamburg heike.zinsmeister@uni-hamburg.de Graeme Hirst Department of Computer Science University of Toronto gh@cs.toronto.edu Abstract Interpreting anaphoric shell nouns (ASNs) such as this issue and this fact is essential to understanding virtually any substantial natural language text. One obstacle in developing methods for automatically interpreting ASNs is the lack of annotated data. We tackle this challenge by exploiting cataphoric shell nouns (CSNs) whose construction makes them particularly easy to interpret (e.g., the fact that X). We propose an ap- proach that uses automatically extracted antecedents of CSNs as training data to interpret ASNs. We achieve precisions in the range of 0.35 (baseline = 0.21) to 0.72 (baseline = 0.44), depending upon the shell noun. 1 Introduction Anaphors such as this fact and this issue encapsulate complex abstract entities such as propositions, facts, and events. An example is shown below. (1) Here is another bit of advice: Environmental Defense, a national advocacy group, notes that “Mowing the lawn with a gas mower produces as much pollution in half an hour as driving a car 172 miles.” This fact may help to explain the recent surge in the sales of the good old- fashioned push mowers or the battery-powered mowers. Here, the anaphor this fact is interpreted with the help of the clausal antecedent marked in bold. The antecedent here is complex because it involves a number of entities and events (e.g., mowing the lawn, a gas mower) and relationships between them, and is abstract because the antecedent itself is not a purely physical entity. The distinguishing property of these anaphors is that they contain semantically rich abstract nouns (e.g., fact in (1)) which characterize and label their corresponding antecedents. Linguists and philoso- phers have studied such abstract nouns for decades (Vendler, 1968; Halliday and Hasan, 1976; Francis, 1986; Ivanic, 1991; Asher, 1993). Our work is in- spired by one such study, namely that of Schmid (2000). Following Schmid, we refer to these abstract nouns as shell nouns, as they serve as conceptual shells for complex chunks of information. Accord- ingly, we refer to the anaphoric occurrences of shell nouns (e.g., this fact in (1)) as anaphoric shell nouns (ASNs). An important reason for studying ASNs is their ubiquity in all kinds of text. Schmid (2000) ob- served that shell nouns such as fact, idea, point, and problem were among the 100 most frequently oc- curring nouns in a corpus of 225 million words of British English. Moreover, ASNs can play several roles in organizing a discourse such as encapsulation of complex information, cohesion, and topic bound- ary marking. So correct interpretation of ASNs can be an important step for correct interpretation of a discourse, and in a number of …

Content truncated to 500 words