Unsupervised Semantic Role Labellin

Unsupervised Semantic Role Labelling Robert S. Swier and Suzanne Stevenson Department of Computer Science University of Toronto Toronto, Ontario, Canada M5S 3G4 � swier,suzanne � @cs.toronto.edu Abstract We present an unsupervised method for labelling the arguments of verbs with their semantic roles. Our bootstrapping algorithm makes initial unam- biguous role assignments, and then iteratively up- dates the probability model on which future assign- ments are based. A novel aspect of our approach is the use of verb, slot, and noun class informa- tion as the basis for backing off in our probability model. We achieve 50–65% reduction in the error rate over an informed baseline, indicating the po- tential of our approach for a task that has heretofore relied on large amounts of manually generated train- ing data. 1 Introduction Semantic annotation of text corpora is needed to support tasks such as information extraction and question-answering (e.g., Riloff and Schmelzen- bach, 1998; Niu and Hirst, 2004). In particular, la- belling the semantic roles of the arguments of a verb (or any predicate), as in (1) and (2), provides crucial information about the relations among event partic- ipants. 1. Kiva ���� �  � � � admires Mats ��� 2. Jo � �  ! returned to London ��"#� $ �� $ $%& Because of the importance of this task, a number of recent methods have been proposed for automatic semantic role labelling (e.g., Gildea and Jurafsky, 2002; Gildea and Palmer, 2002; Chen and Rambow, 2003; Fleischman et al., 2003; Hacioglu et al., 2003; Thompson et al., 2003). These supervised methods are limited by their reliance on the manually role- tagged corpora of FrameNet (Baker et al., 1998) or PropBank (Palmer et al., 2003) as training data, which are expensive to produce, are limited in size, and may not be representative. We have developed a novel method of unsuper- vised semantic role labelling that avoids the need for expensive manual labelling of text, and enables the use of a large, representative corpus. To achieve this, we take a “bootstrapping” approach (e.g., Hin- dle and Rooth, 1993; Yarowsky, 1995; Jones et al., 1999), which initially makes only the role assign- ments that are unambiguous according to a verb lexicon. We then iteratively: create a probability model based on the currently annotated semantic roles, use this probability model to assign roles that are deemed to have sufficient evidence, and add the newly labelled arguments to our annotated set. As we iterate, we gradually both grow the size of the annotated set, and relax the evidence thresholds for the probability model, until all arguments have been assigned roles. To our knowledge, this is the first unsupervised semantic role labelling system applied to general semantic roles in a domain-general corpus. In a similar vein of work, Riloff and colleagues (Riloff and Schmelzenbach, 1998; Jones et al., 1999) used bootstrapping to learn “case frames” for verbs, but their approach has been applied in very narrow topic domains with topic-specific roles. In other work, Gildea (2002) has …

Content truncated to 500 words