Mining the Web for Reciprocal Relationships

Proceedings of the Thirteenth Conference on Computational Natural Language Learning (CoNLL), pages 75–83, Boulder, Colorado, June 2009. c⃝2009 Association for Computational Linguistics Mining the Web for Reciprocal Relationships Michael Paul, Roxana Girju, and Chen Li Linguistics and Computer Science Departments and Beckman Institute, University of Illinois at Urbana-Champaign {mjpaul2, girju, chenli}@illinois.edu Abstract In this paper we address the problem of identifying reciprocal relationships in English. In particular we introduce an algorithm that semi-automatically discovers patterns encod- ing reciprocity based on a set of simple but effective pronoun templates. Using a set of most frequently occurring patterns, we extract pairs of reciprocal pattern instances by search- ing the web. Then we apply two unsuper- vised clustering procedures to form meaning- ful clusters of such reciprocal instances. The pattern discovery procedure yields an accu- racy of 97%, while the clustering procedures indicate accuracies of 91% and 82%. More- over, the resulting set of 10,882 reciprocal in- stances represent a broad-coverage resource. 1 Introduction Reciprocity is a pervasive concept which has been studied a lot in a wide variety of fields from ethics to game theory where it is analyzed as a highly ef- fective “tit for tat” strategy. The ethic of reciprocity (also known as the golden rule), for example, is a moral code born from social interaction: “Do onto others as you would wish them do onto you”. The golden rule appears in most religions and cultures as a standard used to resolve conflicts. According to sociologists and philosophers, the concept of reciprocity lies at the foundation of social organization. It strengthens and maintains social re- lations among people, beyond the basic exchange of useful goods. Thus, the way people conceptualize reciprocity and the way it is expressed in language play an important role in governing people’s behav- ior, judgments, and thus their social interactions. In this paper we present an analysis of the concept of reciprocity as expressed in English and present a way to model it. In particular we introduce an al- gorithm that semi-automatically discovers patterns encoding reciprocity based on a set of simple but ef- fective pronoun templates. We then rank the identi- fied patterns according to a scoring function and se- lect the most frequent ones. Using these patterns we query the web and run two unsupervised clustering procedures to form meaningful clusters of reciprocal pattern instances. The pattern discovery procedure yields an accuracy of 97%, while the clustering pro- cedures indicate accuracies of 91% and 82%. More- over, the resulting set of 10,882 reciprocal instances represent a broad-coverage resource. Next we define the concept of reciprocity as ex- pressed in English. Reciprocity in language The Oxford English Dictionary Online1 defines reciprocity as “a state or relationship in which there is mutual action, influence, giving and taking, cor- respondence, etc., between two parties”, while in WordNet the verb to reciprocate means “to act, feel, or give mutually or in return”. Reciprocity is defined as a relation between two eventualities eo (original eventuality) and er (recip- rocated …

Content truncated to 500 words