RRGparbank: A Parallel Role and Reference Grammar Treebank

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4833–4841 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4833 RRGparbank: A Parallel Role and Reference Grammar Treebank Tatiana Bladier, Kilian Evang, Valeria Generalova, Zahra Ghane, Laura Kallmeyer, Robin M¨ollemann, Natalia Moors, Rainer Osswald, Simon Petitjean Heinrich Heine University D¨usseldorf Universit¨atsstr. 1, 40225 D¨usseldorf, Germany first.last@hhu.de Abstract This paper describes the first release of RRGparbank, a multilingual parallel treebank for Role and Reference Grammar (RRG) that contains annotations of George Orwell’s novel 1984 and its translations. The release comprises the entire novel for English and a constructionally diverse, parallel “seed” sample for German, French, Russian, and Farsi. The paper gives an overview of the annotation decisions taken and describes the adopted treebanking methodology. As a possible application, a multilingual parser is trained on the treebank data. RRGparbank is one of the first resources for which RRG has been applied to large amounts of real-world data. It enables comparative and typological corpus studies in RRG and creates new possibilities of data-driven NLP applications based on RRG. Keywords: Syntax, Treebank, Parallel Corpus, Role and Reference Grammar, English, German, French, Russian, Farsi 1. Introduction Role and Reference Grammar (RRG) (Van Valin and LaPolla, 1997; Van Valin, 2005; Van Valin, 2010) has been proposed as a theory of grammar with an empha- sis on typological adequacy. More recently, RRG has also been studied from the perspective of formal and computational linguistics: A formalization of RRG has been proposed in (Kallmeyer et al., 2013; Osswald and Kallmeyer, 2018), based on which a symbolic parser for precision grammars has been developed (Arps et al., 2019). Moreover, there have been recent initiatives for creating treebanks for RRG (Bladier et al., 2018; Chiarcos and F¨ath, 2019). In this paper, we present the first release of RRG- parbank, a multilingual parallel treebank for RRG, based on George Orwell’s novel 1984 and transla- tions thereof.1 RRGparbank is the first effort to apply RRG to a parallel large-scale corpus, making RRG us- able as a framework for data-driven NLP and corpus- linguistic research. We expect the parallel nature of the treebank to make it especially useful for compara- tive and typological studies, for which RRG has been designed. Applying RRG to large amounts of real data has raised (and already helped answer) a number of questions about the details of RRG analyses which were previously undefined. We have used several inno- vative techniques to make treebanking efficient, includ- ing rule-based conversion from Universal Dependen- cies to RRG (Evang et al., 2021) and statistical parsing as starting points for annotation, as well as incremen- tal improvements of these starting points with human annotators in the loop. We make the resulting treebank available with various download and search options. In this paper, we give a brief introduction to RRG (Section 2), describe our treebanking methodology and 1RRGparbank is available at: https://rrgparbank.phil.hhu.de highlight important aspects of the annotation guide- lines we developed (Section 3), describe the …

Content truncated to 500 words