Abstractive text summarization for Hungarian

In our research we have created a text summarization software tool for Hungarian using multilingual and BERT-based models. Two types of method exist: abstractive extractive. The is more similar to human generated summarization. Target summaries may include phrases that the original does not necessarily contain. This generates summarized by applying keywords were extracted from text. extractive summarizes most important or sentences built both models Hungarian. For models, used BERT model monolingual summarization, in addition also made experiments with ELECTRA We find outperformed all cases. Furthermore, small achieved higher results than some result because much fewer parameters trained on only 1 GPU within couple days. Another consideration are smaller which end users. To best knowledge first systems reported present paper such