Since a lot of multimodal data is created and shared online these days, it’s critical to swiftly and effectively extract pertinent information. Condensing vast amounts of textual data into succinct and insightful summaries is a crucial and practical method that improves information accessibility and saves time. There are three types of text summarizing techniques: hybrid, abstractive, and extractive. While an abstractive summary is creating new sentences that genuinely share the same semantic meaning as the original material, extractive summarization involves choosing key sentences or phrases straight from the source text. In order to improve accuracy and coherence, hybrid approaches combine the two methodologies. The creation of an artificial text summarizing system for the regional language Marathi is the main emphasis of this paper. To do this, high quality Marathi summaries are produced using transformation based models. Models like BART, T5, and Pegasus are employed. Our goal is to assess and compare each model’s performance in order to determine the best method for summarizing Marathi texts and further the field of multilingual natural language processing.
Barphe et al. (2026) studied this question.