Brutus: A Semantic Role Labeling System Incorporating CCG, CFG, and Dependency Features

Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, pages 37–45, Suntec, Singapore, 2-7 August 2009. c⃝2009 ACL and AFNLP Brutus: A Semantic Role Labeling System Incorporating CCG, CFG, and Dependency Features Stephen A. Boxwell, Dennis Mehay, and Chris Brew Department of Linguistics The Ohio State University {boxwe11,mehay,cbrew}@1ing.ohio-state.edu Abstract We describe a semantic role labeling system that makes primary use of CCG-based fea- tures. Most previously developed systems are CFG-based and make extensive use of a treepath feature, which suffers from data spar- sity due to its use of explicit tree configura- tions. CCG affords ways to augment treepath- based features to overcome these data sparsity issues. By adding features over CCG word- word dependencies and lexicalized verbal sub- categorization frames (“supertags”), we can obtain an F-score that is substantially better than a previous CCG-based SRL system and competitive with the current state of the art. A manual error analysis reveals that parser errors account for many of the errors of our system. This analysis also suggests that simultaneous incremental parsing and semantic role labeling may lead to performance gains in both tasks. 1 Introduction Semantic Role Labeling (SRL) is the process of assign- ing semantic roles to strings of words in a sentence ac- cording to their relationship to the semantic predicates expressed in the sentence. The task is difficult because the relationship between syntactic relations like “sub- ject” and “object” do not always correspond to seman- tic relations like “agent” and “patient”. An effective semantic role labeling system must recognize the dif- ferences between different configurations: (a) [The man]Arg0 opened [the door]Arg1 [for him]Arg3 [today]ArgM−T MP . (b) [The door]Arg1 opened. (c) [The door]Arg1 was opened by [a man]Arg0. We use Propbank (Palmer et al., 2005), a corpus of newswire text annotated with verb predicate semantic role information that is widely used in the SRL litera- ture (M`arquez et al., 2008). Rather than describe se- mantic roles in terms of “agent” or “patient”, Propbank defines semantic roles on a verb-by-verb basis. For ex- ample, the verb open encodes the OPENER as Arg0, the OPENEE as Arg1, and the beneficiary of the OPENING action as Arg3. Propbank also defines a set of adjunct roles, denoted by the letter M instead of a number. For example, ArgM-TMP denotes a temporal role, like “to- day”. By using verb-specific roles, Propbank avoids specific claims about parallels between the roles of dif- ferent verbs. We follow the approach in (Punyakanok et al., 2008) in framing the SRL problem as a two-stage pipeline: identification followed by labeling. During identifica- tion, every word in the sentence is labeled either as bearing some (as yet undetermined) semantic role or not . This is done for each verb. Next, during label- ing, the precise verb-specific roles for each word are determined. In contrast to the approach in (Punyakanok et al., 2008), which tags constituents directly, we tag headwords and then associate them with a constituent, as in a previous CCG-based …

Content truncated to 500 words