Перегляд за Автор "Troitskyi, Kyrylo"

Зараз показуємо 1 - 1 з 1
  • Ескіз
    Документ
    Digital technologies for grammatical error correction: deep learning methods & syntactic n-grams
    (Одеський національний університет імені І. І. Мечникова, 2021) Pozharytska, Olena O.; Troitskyi, Kyrylo; Пожарицька, Олена Олександрівна; Троїцький, Кирило Володимирович
    The object of this article is automated grammatical error detection as a field of linguistics. The subject of the article is the variety of methods and techniques used in grammatical error detection along with their applications and evaluation. The article considers the most productive methods used in the field of grammatical error detection and correction in computational linguistics. The purpose of the article is to review major rule-based and deep learning methods used in the area, evaluate and compare them. The methods of research used in this article are data analysis, description of abstract computational models and observation of their performance. The article offers and defines a model based on syntactic n-grams, describes the ways of its implementation and the necessary pre-processing steps for the model to work. The particular error types that the model is capable of detecting are noun-verb agreement errors, preposition errors, noun number errors and some article error types. Also, the article analyses a recent model based on the transformer architecture — GECToR (Grammatical Error Correction: Tag, Not Rewrite). This deep learning model is aimed at detecting and correcting much more complicated errors, including those that rely on extralinguistic realia. Additionally, it is very useful because in contrast to other models that just replace incorrect tokens without explanations, GECToR assigns labels that can be further interpreted for educational purposes. Also, conclusions were made about the advantages and disadvantages of the described models that were discovered after their practical implementation.