Sadhu-Chalit Bengali Sentence Corpus for Text Classification
Description
This dataset includes 190,434 Bengali sentences selected for binary text categorization into Sadhu (classical/formal prose style) and Chalit (current standard/colloquial style) Bengali language forms. The corpus is made up of selected literary works by prominent Bengali authors such as Rabindranath Tagore, Sarat Chandra Chattopadhyay, Kazi Nazrul Islam, Manik Bandyopadhyay, Bibhutibhushan Bandyopadhyay, and other influential writers who represent various periods and literary traditions in Bengali literature. The primary goal of this dataset is to provide a benchmark corpus for Natural Language Processing (NLP) applications such as text classification, writing style identification, linguistic variation analysis, dialectal and stylistic analysis, and broader computational linguistic research in Bengali. The inclusion of literary works from many authors and periods gives a wide range of linguistic patterns and stylistic traits, making the dataset useful for constructing and testing machine learning and deep learning models for Bengali language style categorization and related NLP tasks. Dataset File Structure: Total Instances: 190,434 Bengali sentences. Columns: Column1 (String): Contains the Bengali sentence text. Column2 (Categorical): Contains the corresponding Bengali linguistic-style label, either 'Shadhu' or 'Chalit'. Class Distribution: Chalit: 100,666 instances (52.86%) Shadhu: 89,768 instances (47.14%) The dataset exhibits a relatively balanced class distribution, with a slight predominance of Chalit sentences. This distribution provides a suitable basis for binary classification experiments while minimizing substantial class-imbalance effects.
Files
Steps to reproduce
Source Extraction: Bengali text was collected and extracted from digitized literary works, including novels, short stories, and other selected literary texts written by prominent Bengali authors such as Rabindranath Tagore, Sarat Chandra Chattopadhyay, Kazi Nazrul Islam, Manik Bandyopadhyay, Bibhutibhushan Bandyopadhyay, and other notable Bengali writers. Text Preprocessing: The collected texts were preprocessed through sentence-level tokenization, structural cleaning, and removal of formatting artifacts to obtain standardized and machine-readable sentence units. Labeling: Each sentence was assigned to one of two linguistic style categories based on characteristic grammatical structures, verb inflections, pronoun forms, and other stylistic markers: Sadhu: Classical or formal literary Bengali sentence structure. Chalit: Modern standard or colloquial Bengali sentence structure. Dataset Packaging: The processed and labeled sentences were organized into a structured CSV file, sadhu_chalit_sentence_with_label.csv, containing the sentence text and corresponding linguistic-style label. The final dataset comprises 190,434 Bengali sentences suitable for binary text-classification and Bengali stylistic analysis tasks.
Institutions
- Pabna University of Science and TechnologyRajshahi Division, Pābna