Annotated Bengali Tense Dataset

Description

This dataset contains Bengali sentences annotated with tense labels of about 12,071 instances: Past (3832), Present (4243), and Future (3996). It was manually curated to support research in temporal analysis, tense classification, and low-resource natural language processing. The dataset spans multiple domains, including newspapers, literature, blogs, and conversational text, ensuring linguistic diversity and balanced class distribution. The dataset complements the publication “Tense Classification for Low-Resource Bengali Language with a Curated Dataset and Transformer-Based Modeling” (IEEE QPAIN 2026), where it was used to benchmark classical machine learning models (Naïve Bayes, SVM), recurrent neural networks (BiLSTM), and transformer-based architectures (BanglaBERT). Research-ready format: XLSX with two columns (Sentence, Tense). Potential Uses: Benchmarking tense classification models, temporal analysis in Bengali NLP, cross-lingual transfer learning studies, low-resource language modeling. License: CC BY 4.0 (open use with attribution).

Steps to reproduce

1. Download dataset. 2. Preprocess text: clean, normalize, tokenize. 3. Apply an appropriate data splitting technique. 4. Train the necessary models. 5. Evaluate performance: accuracy, precision, recall, macro F1; inspect confusion matrices. 6. Interpret results.

Institutions

• Rajshahi University of Engineering and Technology

  Rajshahi Division, Rajshahi

Categories

Licence