Designing the LECOR Learner Corpus for Romanian

Proceedings of Recent Advances in Natural Language Processing, pages 143–152 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_016 143 Designing the LECOR Learner Corpus for Romanian Ana-Maria Barbu1,2, Elena Irimia3, Carmen Mîrzea Vasile1,2, and Vasile P˘ais, 3 1Faculty of Letter, University of Bucharest, 5-7 Edgar Quinet, Bucharest 2"Iorgu Iordan - Al. Rosetti" Institute of Linguistics, 13 Calea 13 Septembrie, Bucharest, Romania 3Romanian Academy Research Institute for Artificial Intelligence, 13 Calea 13 Septembrie, Bucharest, Romania anamaria.barbu@g.unibuc.ro elena@racai.ro carmen_marzea@yahoo.fr vasile@racai.ro Abstract This article presents a work-in-progress project, which aims to build and utilize a corpus of Ro- manian texts written or spoken by non-native students of different nationalities, who learn Romanian as a foreign language in the one- year, intensive academic program organized by the University of Bucharest. This corpus, called LECOR – Learner Corpus for Romanian – is made up of pairs of texts: a version of the student and a corrected one of the teacher. Each version is automatically annotated with lemma and POS-tag, and the two versions are then compared, and the differences are marked as errors at this stage. The corpus also con- tains metadata file sets about students and their samples. In this article, the conceptual frame- work for building and utilization of the corpus is presented, including the acquisition and or- ganization phases of the primary material, the annotation process, and the first attempts to adapt the NoSketch Engine query interface to the project’s objectives. The article concludes by outlining the next steps in the development of the corpus aimed at quantitative accumula- tion and the development of the error correction process and the complex error annotation. 1 Introduction The LECOR corpus is developed through the project "Learner Corpus of Romanian (LECOR). Collection, Annotation and Applications", funded by the Romanian Government as part of the sub- program dedicated to research projects to stimulate independent young teams (TE). The host institution of the project is the University of Bucharest. In this section we present the core features of the LECOR corpus and the different uses for which it has been designed. The main goal of the project is to build and make accessible the first Romanian electronic learner corpus and, at the same time, to profession- alize human resources in this field of research. It will be a corpus in free access, open for search- ing through the provided interface. LECOR will be downloadable only by request and only for re- search purposes, which will be in accordance with the informed consent signed by learners. LECOR is a monolingual general learner corpus, collected over four academic years (2019-2024). The target language is (general, standard) Roma- nian, learned in a native speaking environment and taught by native teachers. The learners are univer- sity students, aged generally between 18 and 25. Their native languages are various (more than 20 mother tongues): Arabic, Chinese, Korean, Alba- nian, Greek, Armenian, Turkmen, Turkish, Persian, Slavic languages (Bulgarian, Serbian, Ukrainian, Belarusian, Russian, etc.). Among the internation- ally spoken languages, Arabic and Mandarin Chi- nese are well …

Content truncated to 500 words