Unsupervised Semantic Role Induction with Graph Partitioning

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1320–1331, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Unsupervised Semantic Role Induction with Graph Partitioning Joel Lang and Mirella Lapata Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh 10 Crichton Street, Edinburgh EH8 9AB, UK J.Lang-3@sms.ed.ac.uk, mlap@inf.ed.ac.uk Abstract In this paper we present a method for unsuper- vised semantic role induction which we for- malize as a graph partitioning problem. Ar- gument instances of a verb are represented as vertices in a graph whose edge weights quan- tify their role-semantic similarity. Graph par- titioning is realized with an algorithm that it- eratively assigns vertices to clusters based on the cluster assignments of neighboring ver- tices. Our method is algorithmically and con- ceptually simple, especially with respect to how problem-specific knowledge is incorpo- rated into the model. Experimental results on the CoNLL 2008 benchmark dataset demon- strate that our model is competitive with other unsupervised approaches in terms of F1 whilst attaining significantly higher cluster purity. 1 Introduction Recent years have seen increased interest in the shal- low semantic analysis of natural language text. The term is most commonly used to describe the au- tomatic identification and labeling of the seman- tic roles conveyed by sentential constituents (Gildea and Jurafsky, 2002). Semantic roles describe the se- mantic relations that hold between a predicate and its arguments (e.g., “who” did “what” to “whom”, “when”, “where”, and “how”) abstracting over sur- face syntactic configurations. In the example sentences below, window occu- pies different syntactic positions — it is the object of broke in sentences (1a,b), and the subject in (1c) — while bearing the same semantic role, i.e., the phys- ical object affected by the breaking event. Analo- gously, ball is the instrument of break both when realized as a prepositional phrase in (1a) and as a subject in (1b). (1) a. [Jim]A0 broke the [window]A1 with a [ball]A2. b. The [ball]A2 broke the [window]A1. c. The [window]A1 broke [last night]TMP. The semantic roles in the examples are labeled in the style of PropBank (Palmer et al., 2005), a broad- coverage human-annotated corpus of semantic roles and their syntactic realizations. Under the Prop- Bank annotation framework (which we will assume throughout this paper) each predicate is associated with a set of core roles (named A0, A1, A2, and so on) whose interpretations are specific to that predi- cate1 and a set of adjunct roles such as location or time whose interpretation is common across predi- cates (e.g., last night in sentence (1c)). The availability of PropBank and related re- sources (e.g., FrameNet; Ruppenhofer et al. (2006)) has sparked the development of great many seman- tic role labeling systems most of which conceptu- alize the task as a supervised learning problem and rely on role-annotated data for model training. Most of these systems implement a two-stage architec- ture consisting of argument identification (determin- ing the arguments of the verbal predicate) and ar- gument …

Content truncated to 500 words