Fine-tuning and multilingual pre-training for abstractive summarization task for the Arabic language

The main task of our research is to train various abstractive summarization models for the Arabic language. work text has hardly begun so far due unavailability datasets needed that. In previous research, we created first monolingual corpus in language summarization. Based on this corpus, fine-tuned transformer models. We tested PreSumm and multilingual BART achieved a “state art” result area with method. present study continues same series research. extended “AraSum” managed reach up 50 thousand items, each consisting an article its corresponding lead. addition, pretrained own trilingual them addition mT5 model language, using AraSum corpus. While there room improvement resources infrastructure possess, results clearly demonstrate that most surpassed XL-Sum which considered be state art far. Our will released facilitate future