Fine-grained Morphosyntactic Analysis and Generation Tools for More Than One Thousand Languages

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 386–391 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 386 Toward a Paradigm Shift in Collection of Learner Corpora Anisia Katinskaia,1,2 Sardana Ivanova,1,2 Roman Yangarber2 1 Department of Computer Science, University of Helsinki, Finland 2 Department of Digital Humanities, University of Helsinki, Finland first.last@helsinki.fi Abstract We present the first version of the longitudinal Revita Learner Corpus (ReLCo), for Russian. In contrast to traditional learner corpora, ReLCo is collected and annotated fully automatically, while students perform exercises using the Revita language-learning platform. The corpus currently contains 8 422 sentences exhibiting several types of errors—grammatical, lexical, orthographic, etc.—which were committed by learners during practice and were automatically annotated by Revita. The corpus provides valuable information about patterns of learner errors and can be used as a language resource for a number of research tasks, while its creation is much cheaper and faster than for traditional learner corpora. A crucial advantage of ReLCo that it grows continually while learners practice with Revita, which opens the possibility of creating an unlimited learner resource with longitudinal data collected over time. We make the pilot version of the Russian ReLCo publicly available. Keywords: L2 learner corpus, grammatical error detection, grammatical error correction, CALL, ICALL, Revita, low-resource languages 1. Introduction The most widely used definition of a learner corpus was provided by (Granger, 2002): “computer learner corpora are electronic collections of authentic textual data col- lected according to explicit design criteria for a particu- lar SLA/FLT1 purpose in a standardized format.” Granger (2002) notes that in the context of foreign/second language learning the notion of authenticity is problematic: language is learnt in classrooms and learner data is rarely fully nat- ural. Even when learners are free to write what they like, they are usually constrained by topic and time limits. Language teachers and researchers in language teaching and learning have been collecting learner corpora for decades. Most of the corpora available today contain En- glish as the learning language (L2). Of the 174 learner cor- pora in the list prepared by the Centre for English Corpus Linguistics at the Universit´e Catholique de Louvain,2 93 are English learner corpora (over 53%). We can observe that most languages, except English, Spanish, French, and German, are low-resourced in relation to available learner data. For example, only two Russian leaner corpora appear in the list.3 Creation of learner corpora is an extremely labor-intensive task, which involves collecting text samples, transcribing, annotating/classifying errors, and providing corrections for all errors that the learners made, as well as collecting meta- data. Nevertheless, this massive effort is justified, because learner corpora are useful for a wide range of critical tasks. These tasks include discovering and studying: • common errors and patterns of errors that learners (or certain groups of learners) make, • common learning paths—how and in what order learn- ers acquire linguistic skills, 1Second Language Learning, Foreign Language Teaching 2https://uclouvain.be/en/research-institutes/ilc/cecl/learner- corpora-around-the-world.html 3http://www.web-corpora.net/RLC/, http://rus-ltc.org/search • how learner …

Content truncated to 500 words