A Supervised Approach for Enriching the Relational Structure of Frame Semantics in FrameNet

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 3542–3552, Osaka, Japan, December 11-17 2016. A Supervised Approach for Enriching the Relational Structure of Frame Semantics in FrameNet Shafqat Mumtaz Virk and Philippe Muller and Juliette Conrath IRIT, Toulouse University 118 Route de Narbonne 31062 Toulouse, France Abstract Frame semantics is a theory of linguistic meanings, and is considered to be a useful framework for shallow semantic analysis of natural language. FrameNet, which is based on frame semantics, is a popular lexical semantic resource. In addition to providing a set of core semantic frames and their frame elements, FrameNet also provides relations between those frames (hence providing a network of frames i.e. FrameNet). We address here the limited coverage of the network of conceptual relations between frames in FrameNet, which has previously been pointed out by others. We present a supervised model using rich features from three different sources: structural features from the existing FrameNet network, information from the WordNet relations between synsets projected into semantic frames, and corpus-collected lexical associations. We show large improvements over baselines consisting of each of the three groups of features in isolation. We then use this model to select frame pairs as candidate relations, and perform evaluation on a sample with good precision. 1 Introduction In the area of formal linguistics, frame semantics is a theory of meanings, which was introduced by Charles J. Fillmore and his colleagues back in the early 1980’s. Frame semantic analysis (Das et al., 2014), which is based on frame semantics, itself is a type of shallow semantic analysis in which the focus is on predicates and their arguments. Frame semantic analysis is abstracting away from single verbal predicates, used in other semantic analysis approaches such as Propbank-based work (Kingsbury and Palmer, 2012), to ”semantic frames” (Fillmore, 1982). The idea in frame semantics is to group predicates referring to similar events, processes, and/or object types under the umbrella term ’a semantic frame’, which can be expressed with different parts of speech and with arguments that can have various syntactic realizations. For instance, the sentence in example (1-a) could be seen as introducing a commerce frame, indicated by the lexical unit buy. This frame has a set of necessary arguments, a buyer, some goods, and a seller, which are realized by role fillers Lester, a car and Jimmy in the example. Frames can be realized with different lexical units and syntactic constructions, so that sentences (1-b)-(1-c) would have a similar meaning with respect to the frame in question. (1) a. [Lester]buyer [bought]commerce [a car]goods [from Jimmy]seller. b. [Jimmy]seller [sold]commerce [a car]goods [to Lester]buyer. c. [Lester]buyer’s [purchase]commerce of [Jimmy]seller’s [car]goods (...) Most recent NLP work on such semantic frames are based on FrameNet (henceforth FN) (Baker et al., 1998), a resource listing frames, how they can be evoked, and their argument types. FrameNet also lists how the frames are organized with respect to each other, with a set of frame-to-frame relations. If FrameNet based analysis has …

Content truncated to 500 words