DeSRL: A Linear-Time Semantic Role Labeling System

This paper describes the DeSRL system, a joined effort of Yahoo! Research Barcelona and Università di Pisa for the CoNLL-2008 Shared Task (Surdeanu et al., 2008). The system is characterized by an efficient pipeline of linear complexity components, each carrying out a different sub-task. Classifier errors and ambiguities are addressed with several strategies: revision models, voting, and reranking. The system participated in the closed challenge ranking third in the complete problem evaluation with the following scores: 82.06 labeled macro F1 for the overall task, 86.6 labeled attachment for syntactic dependencies, and 77.5 labeled F1 for semantic dependencies. 1 System description DeSRL is implemented as a sequence of components of linear complexity relative to the sentence length. We decompose the problem into three subtasks: parsing, predicate identification and classification (PIC), and argument identification and classification (AIC). We address each of these subtasks with separate components without backward feedback between sub-tasks. However, the use of multiple parsers at the beginning of the process, and re-ranking at the end, contribute beneficial stochastic aspects to the system. Figure 1 summarizes the system architecture. We detail the parsing ∗All authors contributed equally to this work. ∗ c © 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. sub-task in Section 2 and the semantic sub-tasks (PIC and AIC) in Section 3.