Automatic Generation of a Lexical Resource to support Semantic Role Labeling in Portuguese

Proceedings of the Fourth Joint Conference on Lexical and Computational Semantics (*SEM 2015), pages 216–221, Denver, Colorado, June 4–5, 2015. Automatic Generation of a Lexical Resource to support Semantic Role Labeling in Portuguese Magali Sanches Duran Sandra Aluísio Center for Computational Linguistics (NILC) São Paulo University (USP) São Carlos-SP, Brazil Center for Computational Linguistics (NILC) São Paulo University (USP) São Carlos-SP, Brazil magali.duran@uol.com.br sandra@icmc.usp.br Abstract This paper reports an approach to automatical- ly generate a lexical resource to support in- cremental semantic role labeling annotation in Portuguese. The data come from the corpus Propbank-Br (Propbank of Brazilian Portu- guese) and from the lexical resource of Eng- lish Propbank, as both share the same structure. In order to enable the strategy, we added extra annotation to Propbank-Br. This approach is part of a previous decision to in- vert the process of implementing a Propbank project, by first annotating a core corpus and only then generating a lexical resource to ena- ble further annotation tasks. The reasoning behind such inversion is to explore the task empirically before distributing the annotation task and to provide simultaneously: 1) a first training corpus for SRL in Brazilian Portu- guese and 2) annotated examples to compose a lexical resource to support SRL. The main contribution of this paper is to point out to what extent linguistic effort may be reduced, thereby speeding up the construction of a lexi- cal resource to support SRL for less resourced languages. The corpus Propbank-Br, with the extra annotation described herein, is publicly available. 1 Introduction The task of semantic role labeling (SRL) consists of identifying a predicate (a verb or a predicate noun) and its arguments, assigning to each argu- ment the semantic roles it play in the argumental structure (Palmer et al. 2010). For example, in the sentence “Parents complain to education depart- ment about schools constantly switching uni- forms”, there are two predicates: “complain” and “switching”. The argumental structure of “com- plain” is: “Parents” (agent), “to the education de- partment” (recipient), “about schools constantly switching uniforms” (theme). The argumental structure of “switching” is: “schools” (agent); “constantly” (time/frequency); “uniforms” (theme). There is no consensus regarding an ideal set of semantic role labels and, for this reason, the first difficult decision in a project of SRL is to choose which set to adopt. No matter which set is used, it is not always easy to decide which label to assign to each argument during the annotation task. In order to facilitate such decision, some projects of SRL developed lexical resources that predict the set of semantic roles required by each predicate. Some of such resources define semantic roles for verb classes, as Verbnet (Kipper et al. 2006); oth- ers for semantic frames, as Framenet (Baker et al. 1998); others define semantic roles for verb senses, as Propbank (Palmer et al. 2005) or for predicate nouns, as Nombank (Meyers et al, 2004). The more detailed and clear is the lexical re- source, the easier the decision about which role label to …

Content truncated to 500 words