Building a Hebrew Semantic Role Labeling Lexical Resource from Parallel Movie Subtitles

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5934–5942 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5934 Building a Hebrew Semantic Role Labeling Lexical Resource from Parallel Movie Subtitles Ben Eyal, Michael Elhadad Dept. of Computer Science, Ben-Gurion University of the Negev Beer Sheva, Israel bene@post.bgu.ac.il, elhadad@cs.bgu.ac.il Abstract We present a semantic role labeling resource for Hebrew built semi-automatically through annotation projection from English. This corpus is derived from the multilingual OpenSubtitles dataset and includes short informal sentences, for which reliable linguistic annotations have been computed. We provide a fully annotated version of the data including morphological analysis, dependency syntax and semantic role labeling in both FrameNet and PropBank styles. Sentences are aligned between English and Hebrew, both sides include full annotations and the explicit mapping from the English arguments to the Hebrew ones. We train a neural SRL model on this Hebrew resource exploiting the pre-trained multilingual BERT transformer model, and provide the first available baseline model for Hebrew SRL as a reference point. The code we provide is generic and can be adapted to other languages to bootstrap SRL resources. Keywords: SRL, FrameNet, PropBank, Hebrew, Cross-lingual Linguistic Annotations 1. Introduction Semantic role labeling (SRL) is the task that consists of an- notating sentences with labels that answer questions such as “Who did what to whom, when, and where?”, the answers to these questions are called “roles.” Two major SRL annotation schemes have been used in re- cent years, FrameNet (Baker et al., 1998) and PropBank (Palmer et al., 2005). Research in SRL has generated fully- annotated corpora in various languages, several CoNLL shared tasks, and many automatic SRL systems. Besides its important theoretical role, SRL has been found useful for information extraction (Christensen et al., 2010; Stanovsky et al., 2018) and question answering (FitzGerald et al., 2018). In this work, we set out to contribute a new Hebrew cor- pus and lexical resources to further the advancement of the Hebrew FrameNet Project and to provide the first SRL re- source covering both FrameNet and PropBank annotations (Hayoun and Elhadad, 2016). Our approach bootstraps He- brew SRL annotations by projecting predicted annotations from English to Hebrew aligned sentences. To ensure that the resulting Hebrew annotations are reliable, we design confidence metrics on both the original English annotations and the projection method itself. We describe previous work in multilingual FrameNet and cross-lingual SRL. We then describe the large unannotated dataset we selected, OpenSubtitles, and the method we use to align documents and sentences between English (EN) and Hebrew (HE), and to project SRL annotations from EN to HE. We present the dataset that is obtained as a result of this procedure, which includes aligned annotated sentences in EN and HE, with morphological analysis, dependency syntax and SRL in FrameNet and in PropBank formats. Fi- nally, we present baseline automatic SRL systems in HE trained on the generated data. Our main contribution in the paper is the description of an …

Content truncated to 500 words