Abstraction and Generalisation in Semantic Role Labels: PropBank, VerbNet or both?

Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, pages 288–296, Suntec, Singapore, 2-7 August 2009. c⃝2009 ACL and AFNLP Abstraction and Generalisation in Semantic Role Labels: PropBank, VerbNet or both? Paola Merlo Linguistics Department University of Geneva 5 Rue de Candolle, 1204 Geneva Switzerland Paola.Merlo@unige.ch Lonneke Van Der Plas Linguistics Department University of Geneva 5 Rue de Candolle, 1204 Geneva Switzerland Lonneke.VanDerPlas@unige.ch Abstract Semantic role labels are the representa- tion of the grammatically relevant aspects of a sentence meaning. Capturing the nature and the number of semantic roles in a sentence is therefore fundamental to correctly describing the interface between grammar and meaning. In this paper, we compare two annotation schemes, Prop- Bank and VerbNet, in a task-independent, general way, analysing how well they fare in capturing the linguistic generalisations that are known to hold for semantic role labels, and consequently how well they grammaticalise aspects of meaning. We show that VerbNet is more verb-specific and better able to generalise to new seman- tic role instances, while PropBank better captures some of the structural constraints among roles. We conclude that these two resources should be used together, as they are complementary. 1 Introduction Most current approaches to language analysis as- sume that the structure of a sentence depends on the lexical semantics of the verb and of other pred- icates in the sentence. It is also assumed that only certain aspects of a sentence meaning are gram- maticalised. Semantic role labels are the represen- tation of the grammatically relevant aspects of a sentence meaning. Capturing the nature and the number of seman- tic roles in a sentence is therefore fundamental to correctly describe the interface between gram- mar and meaning, and it is of paramount impor- tance for all natural language processing (NLP) applications that attempt to extract meaning rep- resentations from analysed text, such as question- answering systems or even machine translation. The role of theories of semantic role lists is to obtain a set of semantic roles that can apply to any argument of any verb, to provide an unam- biguous identifier of the grammatical roles of the participants in the event described by the sentence (Dowty, 1991). Starting from the first proposals (Gruber, 1965; Fillmore, 1968; Jackendoff, 1972), several approaches have been put forth, ranging from a combination of very few roles to lists of very fine-grained specificity. (See Levin and Rap- paport Hovav (2005) for an exhaustive review). In NLP, several proposals have been put forth in recent years and adopted in the annotation of large samples of text (Baker et al., 1998; Palmer et al., 2005; Kipper, 2005; Loper et al., 2007). The an- notated PropBank corpus, and therefore implicitly its role labels inventory, has been largely adopted in NLP because of its exhaustiveness and because it is coupled with syntactic annotation, properties that make it very attractive for the automatic learn- ing of these roles and their further applications to NLP tasks. However, the labelling choices …

Content truncated to 500 words