CMILLS: Adapting Semantic Role Labeling Features to Dependency Parsing

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 433–437, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics CMILLS: Adapting Semantic Role Labeling Features to Dependency Parsing Chad Mills Gina-Anne Levow University of Washington University of Washington Guggenheim Hall, 4th Floor Guggenheim Hall, 4th Floor Seattle, WA 98195, USA Seattle, WA 98195, USA chills@uw.edu levow@uw.edu Abstract We describe a system for semantic role label- ing adapted to a dependency parsing frame- work. Verb arguments are predicted over nodes in a dependency parse tree instead of nodes in a phrase-structure parse tree. Our system par- ticipated in SemEval-2015 shared Task 15, Subtask 1: CPA parsing and achieved an F- score of 0.516. We adapted features from prior semantic role labeling work to the dependency parsing paradigm, using a series of supervised classifiers to identify arguments of a verb and then assigning syntactic and semantic labels. We found that careful feature selection had a major impact on system performance. How- ever, sparse training data still led rule-based systems like the baseline to be more effective than learning-based approaches. 1 Introduction We describe our submission to the SemEval-2015 Task 15, Subtask 1 on Corpus Pattern Analysis (Baisa et al. 2015). This task is similar to semantic role labeling but with arguments based on nodes in dependency parses instead of a syntactic parse tree. The verb’s arguments are identified and labeled with both their syntactic and semantic roles. For example, consider the sentence “But he said Labour did not agree that Britain could or should abandon development, either for itself or for the developing world.” This subtask involves taking that sentence and making the following determina-  tions relative to the given verb “abandon”: “Britain” is the syntactic subject of “abandon” and falls under the “Institution” semantic type  “development” is the syntactic object of “aban- don” and is of semantic type “Activity” We organize the remainder of our paper as fol- lows: Section 2 describes our system, Section 3 presents experiments, and Section 4 concludes. 2 System Description Our system consists of a pipelined five-component system plus source data and resources. A system di- agram is shown in Figure 1. A cascading series of MaxEnt classifiers are used to identify arguments, their syntactic labels, and then their semantic labels. Each token in an input sentence was a training ex- ample. Sketch Engine (Kilgarriff 2014) was used to help with featurization. All sentences in the training data were parsed and POS tagged using the Stanford CoreNLP tools (Manning et al. 2014). This data was used to generate features which are then supplied to an Argument Identification Classifier (AIC) that identifies whether or not a particular token is one of the relevant verb’s arguments. For the tokens identified as arguments to the verb, a Syntax Classifier identifies the syntactic role of the token. This is done using a multi-class MaxEnt model with the same features as the AIC plus fea- tures derived from the AIC’s predictions. A similar Semantics Classifier …

Content truncated to 500 words