Information Technologies for Solving the Problem of Correcting Errors in Ukrainian-language Texts

This article is dedicated to the study and analysis of grammatical error correction (GEC) tasks in Ukrainian language texts, which is a significant issue in the field of natural language processing (NLP). The paper addresses the specific challenges faced by automatic error correction systems due to the peculiarities of the Ukrainian language, such as its morphological complexity and contextuality. Examples of typical errors are provided, and the reasons why existing GEC methods often prove insufficient for Ukrainian are analysed. The literature review covers recent research and publications in the GEC field, particularly those related to other languages, and highlights approaches that can be adapted for Ukrainian. Special attention is given to the analysis of existing Ukrainian text corpora, such as the UA_GEC and others used for training machine learning models. Their volume, text types, specifications, advantages, and disadvantages are described. Tools for natural language processing that support Ukrainian, such as LanguageTool, NLP-uk, Stanza, NLP-Cube, pymorphy2, Tree_stam, are examined. Their functionalities, performance, and accuracy are analysed. Pre-trained machine learning models, including mBART50 and mT5 were adapted for Ukrainian with description of their effectiveness in GEC tasks. The article presents practical aspects of applying these models and corpora for automatic grammatical error correction in Ukrainian texts. The process of adapting models to the specifics of the Ukrainian language is detailed, practical case examples are provided, and results are analysed. A significant part of the paper is devoted to the description of the developed decision support system for correcting errors in Ukrainian language texts. The system’s architecture, its main components, and processes are presented through UML diagrams. The input and output data are described, along with an analysis of the obtained results, demonstrating the effectiveness of the proposed solutions. The results of this study can be useful for NLP system developers, researchers in text processing, and educational institutions focused on improving the quality of written texts in Ukrainian.