ELERRANT: Automatic Grammatical Error Type Classification for Greek

Proceedings of Recent Advances in Natural Language Processing, pages 708–717 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_081 708 ELERRANT: Automatic Grammatical Error Type Classification for Greek Katerina Korre†, Marita Chatzipanagiotou†, John Pavlopoulos†♣ † Athens University of Economics and Business, Greece ♣ Stockholm University, Sweden katkorre95,marita.xatzh,annis@aueb.gr Abstract In this paper, we introduce the Greek ver- sion of the automatic annotation tool ERRANT (Bryant et al., 2017), which we named ELER- RANT. ERRANT functions as a rule-based er- ror type classifier and was used as the main evaluation tool of the systems participating in the BEA-2019 (Bryant et al., 2019) shared task. Here, we discuss grammatical and mor- phological differences between English and Greek and how these differences affected the development of ELERRANT. We also intro- duce the first Greek Native Corpus (GNC) and the Greek WikiEdits Corpus (GWE), two new evaluation datasets with errors from na- tive Greek learners and Wikipedia Talk Pages edits respectively. These two datasets are used for the evaluation of ELERRANT. This paper is a sole fragment of a bigger picture which illus- trates the attempt to solve the problem of low- resource languages in NLP, in our case Greek. 1 Introduction Grammatical Error Correction (GEC) is the task of automatically correcting language mistakes in written texts. These mistakes can vary from gram- matical mistakes to punctuation, spelling and mor- phology of a word. The development of a GEC system usually involves the transformation of an erroneous sentence into its correct version, while also keeping the initial meaning intact. Develop- ing those systems requires error annotated data, which can be either learner data or artificial. High- resource languages, such as English, present a va- riety of learner data that cover a relatively wide spectrum of language proficiency levels, native language and topics. Some notable examples are the Cambridge English Write & Improve cor- pus (Yannakoudakis et al., 2018), the LOCNESS corpus (Granger, 1998), and the NUCLE corpus (Dahlmeier et al., 2013). Low-resource languages, on the other hand, are characterized by a scarcity of such corpora, as well as of other GEC resources and Natural Language Processing (NLP) tools. That is also the case of Greek. Although, Greek is only spoken by approx. 13.5 million people (native),1 the fact that Greece has a high immigrant population underlines the need for learning Greek as a Second Language (GSL).2 Therefore, and as technology is being integrated in education (Meurers, 2012; Forcier, 2016), the need for more GEC and NLP tools for Greek becomes evident. In this paper, we present ELERRANT, an auto- matic annotation tool, which is based on ERRANT (Bryant et al., 2017). ERRANT produces an anno- tation mainly consisting of the error location, the error type and the correction of the error, by using an original erroneous sentence along with its cor- rection as input. ERRANT is the first toolkit that not only annotates texts but also provides automatic error typing, offering detailed feedback to Second Language (L2) learners and useful information for language analysis (Bryant et al., 2017). Most im- …

Content truncated to 500 words