DefinePK

Abstract

This paper presents an extractive text summarization method specially designed for Sindhi, a culturally rich but low-resource Indo-Aryan language spoken widely in Pakistan. The study focuses on selecting the most relevant sentences from Sindhi texts, employing Natural Language Processing (NLP) techniques to generate concise summaries.
The proposed system incorporates essential preprocessing steps, including text cleaning, tokenization, and stemming/lemmatization. For future extraction, it utilizes TF-IDF and sentence embeddings. After scoring the sentences, the most significant ones are selected to form the final summary.
To evaluate the system's performance in five test paragraphs, several metrics are used, including F1 score, precision, recall, cosine similarity, normalization level distance, and accuracy. The system demonstrates reliable and accurate summarization, and consistency achieving high precision (1.0), strong F1 score (0.89-0.92), a low a low normalized error (0.04), and an overall accuracy of 0.86. Graphic analysis further confirms the model's coherence, semantic retention, and low error rates.
By leveraging NLP for information summarization, this study contributes to preserving and promoting the Sindhi language—potential applications including digital accessibility, education, and content curation. Future research aims to enhance contextual understanding by exploring transformer-based models like BERT and extending the approach to abstraction summarization.