Semi-automatically Annotated Learner Corpus for Russian

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 832–839 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 832 Semi-automatically Annotated Learner Corpus for Russian Anisia Katinskaia1,2, Maria Lebedeva3, Jue Hou1,2, Roman Yangarber2 1 Department of Computer Science, University of Helsinki, Finland 2 Department of Digital Humanities, University of Helsinki, Finland 3 Language and Cognition Laboratory, Pushkin State Russian Language Institute, Russia first.last@helsinki.fi, m.u.lebedeva@gmail.com Abstract We present ReLCo— the Revita Learner Corpus—a new semi-automatically annotated learner corpus for Russian. The corpus was collected while several hundreds L2 learners were performing exercises using the Revita language-learning system. All errors were detected automatically by the system and annotated by type. Part of the corpus was annotated manually—this part was created for further experiments on automatic assessment of grammatical correctness. The Learner Corpus provides valuable data for studying patterns of grammatical errors, experimenting with grammatical error detection and grammatical error correction, and developing new exercises for language learners. Automating the collection and annotation makes the process of building the learner corpus much cheaper and faster, in contrast to the traditional approach of building learner corpora. We make the data publicly available. Keywords: Learner corpus, L2, ICALL, language learning, grammatical error detection 1. Introduction This paper presents a new version of Revita Learner Corpus (ReLCo), automatically and manually anno- tated learner corpus for Russian1 (Katinskaia et al., 2020). It is created to facilitate research in second language (L2) acquisition and foreign language teach- ing (Granger, 1996), as well as in various NLP tasks, e.g., grammatical error correction (GEC) and grammat- ical error detection (GED). ReLCo is built based on the Revita language-learning platform (Katinskaia et al., 2018; Katinskaia and Yan- garber, 2018).2 It is collected continually and annotated automatically, while students perform a variety of exer- cises in the platform. All collected learner data is used by Revita for updating student models of language pro- ficiency, which in turn are used for generating new per- sonalised exercises. This process creates the “learning feedback loop” helping students to improve their lan- guage skills more effectively. Manual annotation was added to the part of released data for further experi- ments with improving automatic annotation, i.e., detec- tion of grammatical errors in learner answers and tag- ging of errors by their type. Most corpora available today contain English as the tar- get learning language (L2). Of the 174 learner corpora in the list prepared by the Centre for English Corpus Linguistics at the Université Catholique de Louvain,3 93 are English learner corpora—over 53%. Only a few of the corpora in the list are available for download. This paper partly address both of these problems by fo- cusing on Russian and making our data freely available. 1https://github.com/Askinkaty/Russian_learner_corpora 2revita.cs.helsinki.fi 3https://uclouvain.be/en/research- institutes/ilc/cecl/learner-corpora-around-the-world.html ReLCo is not a “conventional” learner corpus—it does not include sentences that were fully written by L2 stu- dents. However, we claim that this is a valuable re- source of learner data, which can be used for improving teaching …

Content truncated to 500 words