SaralMarathi: A Regional Language Summarizer Using LLM

The amount of online content in Marathi has gone up quite a bit and therefore it is warranted to suggest some intelligent methods of generating summaries. This paper presents a methodology to efficiently condense the Marathi text using the T5 model which is able to serve multilingual purposes. The T5 which is also called mT5 is the model used to implement the proposed Marathi summarizer because it is built on transformer's framework. This is done by making the model more language specific through training on numerous Marathi language texts to better fit the overviewing style of the ensemble and Marathi language itself. For this purpose, as input text documents, we collected such Marathi texts from various websites, then trained the mT5 model specifically for the summarizing task, and evaluated the uspouted summary with the automated metric ROUGE, as well as qualitatively by checking it and other aspects including style. The findings from the study suggest that the developed system is more effective in selective reading of parts of texts and summarising documents with greater informative content and succinct summaries when compared with other extractive methods. Furthermore, it proved the potential to improve the existing NLP tools aimed at the Low-Resource Languages and Small Language population nations. In order to fulfill these objectives, we found a precise and fast method.