Hungarian Sentence Analysis Learning Application with Transformer Models

Abstract

The purpose of our research is to present a project in which we started to develop an educational support tool that helps primary and high school students to use the correct techniques of sentence analysis based on the rules of Hungarian grammar taught in school. The aim was to create an application called LMEZZ that would help students of the Hungarian education system to practise tasks related to native language lessons. In this way, we expect them to have a more accurate understanding of the grammar rules. The application allows them to learn in the comfort of their own homes by providing immediate and accurate feedback on the solutions to various tasks. Natural language processing has made spectacular progress with the application of neural network technology, especially the contextual transformer model. In our research, Hungarian transformer-based BERT models were trained for our sentence analyser task. The results showed that the transformer models were much more condescending than the previously trained convolutional neural network based SpaCy models. This allowed us to increase the reliability of our software.