Scholarlink

<p>In the domain of Natural Language Processing (NLP), data scarcity remains a major obstacle for low resource languages such as Bengali. This paper introduces a hybrid embedding generation framework designed to explore and enhance literary style transfer between two iconic Bengali poets Rabindranath Tagore and Kazi Nazrul Islam. The proposed approach conducts a comparative embedding level analysis using five distinct pre-trained language models to investigate how effectively each model captures the stylistic, rhythmic, and semantic differences between the two authors. Experimental results demonstrate that embedding representations significantly influence the generator’s capability in reproducing author specific tone and vocabulary. By integrating embedding-based stylistic understanding with generative modeling, this research bridges a crucial gap in Bengali literary text generation and establishes a scalable foundation for low-resource language modeling. The framework not only preserves semantic coherence but also enhances stylistic fidelity paving the way for future explorations in computational creativity, cultural preservation, and multilingual style transfer.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Literary Style Transfer in Bengali Literature Using Hybrid BERT-GAN Architecture: A Comprehensive Study on Tagore and Nazrul

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Hasanat Nihal,
  • Nakib Aman,
  • Md. Abu Johab,
  • Farjana Yasmin,
  • Md. Monir Hossain

摘要

In the domain of Natural Language Processing (NLP), data scarcity remains a major obstacle for low resource languages such as Bengali. This paper introduces a hybrid embedding generation framework designed to explore and enhance literary style transfer between two iconic Bengali poets Rabindranath Tagore and Kazi Nazrul Islam. The proposed approach conducts a comparative embedding level analysis using five distinct pre-trained language models to investigate how effectively each model captures the stylistic, rhythmic, and semantic differences between the two authors. Experimental results demonstrate that embedding representations significantly influence the generator’s capability in reproducing author specific tone and vocabulary. By integrating embedding-based stylistic understanding with generative modeling, this research bridges a crucial gap in Bengali literary text generation and establishes a scalable foundation for low-resource language modeling. The framework not only preserves semantic coherence but also enhances stylistic fidelity paving the way for future explorations in computational creativity, cultural preservation, and multilingual style transfer.