The Hebrew FrameNet Project

The Hebrew FrameNet Project Avi Hayoun, Michael Elhadad Dept. of Computer Science Ben-Gurion University Beer Sheva, Israel {hayounav,elhadad}@cs.bgu.ac.il Abstract We present the Hebrew FrameNet project, describe the development and annotation processes and enumerate the challenges we faced along the way. We have developed semi-automatic tools to help speed the annotation and data collection process. The resource currently covers 167 frames, 3,000 lexical units and about 500 fully annotated sentences. We have started training and testing automatic SRL tools on the seed data. Keywords: FrameNet, Hebrew, frame semantics, semantic resources 1. Introduction Recent years have seen growing interest in the task of Semantic Role Labeling (SRL) of natural language text (sometimes called “shallow semantic parsing”). The task is usually described as the act of identify- ing the semantic roles, which are the set of semantic properties and relationships defined over constituents of a sentence, given a semantic context. The creation of resources that document the realiza- tion of semantic roles in natural language texts, such as FrameNet (Fillmore and Baker, 2010; Ruppenhofer et al., 2010) and PropBank (Kingsbury and Palmer, 2002) have advanced the field of semantic analysis no end and have allowed the development of learning algo- rithms for automatically analyzing the semantic struc- ture of text. Shallow semantic analysis has been shown to con- tribute to the advancement of a wide spectrum of nat- ural language processing tasks, ranging from informa- tion extraction (Surdeanu et al., 2003) and question answering (Shen and Lapata, 2007), to machine trans- lation (Wu and Fung, 2009) and abstractive summa- rization (Melli et al., 2005). 1.1. FrameNet FrameNet (Fillmore and Baker, 2010; Ruppenhofer et al., 2010) is a human-annotated linguistic resource with rich semantic content based on the linguistic the- ory of Frame Semantics proposed by Fillmore (1982). FrameNet defines a formal structure for semantic frames, and various relationships between and within them. Each frame contains a list of frame-evoking words which also serve as the predicates of events described by the frames. These words are called Frame Evoking Elements (FEEs) or Lexical Units (LUs). Addition- ally, each frame defines a list of event-participants and a list of constraints on and relationships between these participants. The participants are called Frame Elements (or FEs). Finally1, and perhaps most importantly, FrameNet contains human-annotated examples of realizations of frames and their structures in natural language. 1.2. FrameNet in Other Languages The original FrameNet project has been adapted and ported to multiple languages. The most active interna- tional FrameNet teams include the Swedish FrameNet (SweFN) covering close to 1,200 frames with 34K LUs (Ahlberg et al., 2014); the Japanese FrameNet (JFN) with 565 frames, 8,500 LUs and 60K annotated ex- ample sentences (Ohara, 2013); and FrameNet Brazil (FN-Br) covering 179 frames, 196 LUs and 12K anno- tated sentences (Torrent and Ellsworth, 2013). Inspired by the ideas developed by the Swedish FrameNet++ project (Friberg Heppin and Voionmaa, 2012) and by Petruck (Petruck, 2005; Petruck, 2009) and Boas (Boas, 2011), we have started the develop- ment of a Hebrew FrameNet, …

Content truncated to 500 words