Bangla Key2Text: Text Generation from Keywords for a Low Resource Language

2026-04-21 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

Bangla Key2Text is a new large-scale dataset comprising 2.6 million Bangla keyword-text pairs, developed to facilitate keyword-driven text generation in Bangla, a low-resource language. The dataset was created by applying a BERT-based keyword extraction pipeline to millions of Bangla news articles, converting raw text into structured keyword-text pairs for supervised learning. Researchers fine-tuned two sequence-to-sequence models, mT5 and BanglaT5, to establish baseline performance on this new benchmark. Evaluations using automatic metrics and human judgments demonstrated that task-specific fine-tuning significantly enhances keyword-conditioned text generation in Bangla, outperforming zero-shot large language models. The dataset, along with the trained models and code, has been made publicly available to support further research in Bangla natural language generation and keyword-to-text tasks.

Key takeaway

For research scientists working on natural language generation in low-resource languages, Bangla Key2Text offers a critical resource. You should consider using this 2.6 million-pair dataset to train and benchmark models for keyword-driven text generation, as task-specific fine-tuning has proven to significantly improve performance over zero-shot approaches. This dataset provides a robust foundation for developing more effective NLG systems in Bangla.

Key insights

A new 2.6M Bangla keyword-text dataset enables supervised keyword-driven text generation for low-resource languages.

Principles

• Task-specific fine-tuning improves generation.
• BERT-based extraction creates structured datasets.

Method

A BERT-based pipeline extracts keywords from raw news articles to form keyword-text pairs, which are then used to fine-tune sequence-to-sequence models like mT5 and BanglaT5 for text generation.

In practice

• Use Bangla Key2Text for Bangla NLG.
• Apply BERT for keyword extraction.

Topics

• Bangla Key2Text
• Keyword-to-Text Generation
• Low-Resource Language NLP
• BERT-based Keyword Extraction
• mT5 and BanglaT5

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Surpassing Scale by Efficiency: A Compact 135M Parameter Foundational LLM Natively Adapted for the Bangla Language — A compact 135M parameter LLM for Bangla achieves 1B-tier performance by efficiently merging tokens, overcoming low-resource script challenges.
• NCTB-QA: A Large-Scale Bangla Educational Question Answering Dataset and Benchmarking Performance — Domain-specific fine-tuning is crucial for robust QA performance in low-resource languages, especially with unanswerable questions.
• Source-Grounded Data Generation for Text-to-JSON Learning — STAGE offers a source-grounded pipeline for scalable, verifiable text-to-JSON training data generation using spreadsheets and LLMs.
• Annotation Challenges in Low-Resource African Languages — High-quality NLP for African languages demands culturally-attuned annotation, robust QA, and fair, community-centric practices.
• Evaluation of Chunking Strategies for Effective Text Embedding in Low-Resource Language on Agricultural Documents — Structural preservation in text chunking significantly outperforms linguistic or semantic methods for effective retrieval in low-resource, technical documents.
• Parsing Nheengatu: Performance Gains for a Brazilian Indigenous Universal Dependencies Treebank — Corpus expansion and standardized annotation improve parsing accuracy and reduce variance for low-resource languages.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.