The Russian PropBank

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5995–6002 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5995 Russian PropBank Sarah Moeller, Irina Wagner, Martha Palmer, Kathryn Conger, Skatje Myers University of Colorado Boulder {first.last}@colorado.edu Abstract This paper presents a proposition bank for Russian (RuPB), a resource for semantic role labeling (SRL). The motivating goal for this resource is to automatically project semantic role labels from English to Russian. This paper describes frame creation strategies, coverage, and the process of sense disambiguation. It discusses language-specific issues that complicated the process of building the PropBank and how these challenges were exploited as language-internal guidance for consistency and coherence. Keywords: lexicon, lexical database, corpus (creation, annotation, etc.), semantics 1. Introduction This paper presents a proposition bank for Russian (RuPB) that balances parallelism with the English PropBank against guidance from linguistic properties specific to Rus- sian. A proposition bank, or PropBank, is a lexical resource that follows the PropBank scheme (Palmer et al., 2005) to provide consistent labeling of semantic roles for large corpora. Semantic Role Labeling (SRL) provides consis- tent semantic information for natural language processing at a level appropriate for statistical machine learning of semantic relations. Data annotated with proposition bank labels supports the training of automatic SRL which im- proves question answering (Zapirain et al., 2013), infor- mation extraction (Moreda et al., 2005), and textual en- tailments (Sammons et al., 2010), and statistical machine translation (Bazrafshan and Gildea, 2013). Semantic roles communicate “Who does What to Whom and How and When and Where?” They may appear in various positions in the sentence, as in the examples below where “John” is always the semantic agent, but appears in (1) as the subject noun phrase, in (2) as the direct object in a passive transi- tive clause, and in (3) as the object of the final prepositional phrase of a passive ditransitive clause. Despite these syn- tactic alternations, in all three sentences the semantic role of “John” as the Agent of the hitting event never changes. (1) John hit the nail with the hammer. (2) The nail was hit by John. (3) The nail was hit with a hammer by John. It is important that natural language processing models works with multiple languages and language structures. Systems that have been evaluated on multiple languages are more likely to generalize well to new languages (Co- hen, 2020). The initial goal for developing a proposi- tion bank for Russian was to support alignment of En- glish and Russian predicates and automatic projection of English semantic roles to Russian texts. This is a com- plicated task because Russian is more complex morpho- logically and exhibits more flexible word order than En- glish or other Indo-European languages currently modeled by proposition banks. This project tests the portability of the PropBank scheme. During development of the the Rus- sian PropBank (RuPB), language-specific issues presented unique challenges. Yet, an in-depth linguistic understand- ing of these issues also provided …

Content truncated to 500 words