Scholarlink

The Bangla language has been a focal point in the realm of natural language processing, with numerous commendable research endeavors. Despite the wealth of studies, there exists a noticeable gap in significant research dedicated to its intricately enriched dialects, primarily hindered by resource scarcity and the intricate grammatical structures inherent to these dialects. In this study, we introduce a Neural Machine Translation (NMT) system designed specifically for one of the prevalent Bangla dialects, namely the Sylheti dialect, with the objective of facilitating seamless translation into standard Bangla. To achieve this, we explore four different Seq2Seq models: Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), and Bidirectional GRU (BiGRU). To facilitate our study, we gathered a dataset consisting of 600 complete sentence examples and 6500 Sylheti words, each paired with their corresponding translations in standard Bangla. Through our experimentation, the BiLSTM model emerged as the most effective for Sylheti to standard Bangla Neural Machine Translation, outperforming other models. Translation quality is evaluated using the BLEU and ROUGE metrics. On the BiLSTM model, we achieved BLEU scores of 57.4, 45.8, 32.0, and 22.8 for BLEU-1, BLEU-2, BLEU-3, and BLEU-4, respectively which demonstrates the effectiveness of our proposed approach in translating the Sylheti dialect to standard Bangla.

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

Sylheti to Standard Bangla Neural Machine Translation: A Deep Learning-Based Dialect Conversion Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tabia Tanzin Prama,
  • Md Musfique Anwar

摘要

The Bangla language has been a focal point in the realm of natural language processing, with numerous commendable research endeavors. Despite the wealth of studies, there exists a noticeable gap in significant research dedicated to its intricately enriched dialects, primarily hindered by resource scarcity and the intricate grammatical structures inherent to these dialects. In this study, we introduce a Neural Machine Translation (NMT) system designed specifically for one of the prevalent Bangla dialects, namely the Sylheti dialect, with the objective of facilitating seamless translation into standard Bangla. To achieve this, we explore four different Seq2Seq models: Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), and Bidirectional GRU (BiGRU). To facilitate our study, we gathered a dataset consisting of 600 complete sentence examples and 6500 Sylheti words, each paired with their corresponding translations in standard Bangla. Through our experimentation, the BiLSTM model emerged as the most effective for Sylheti to standard Bangla Neural Machine Translation, outperforming other models. Translation quality is evaluated using the BLEU and ROUGE metrics. On the BiLSTM model, we achieved BLEU scores of 57.4, 45.8, 32.0, and 22.8 for BLEU-1, BLEU-2, BLEU-3, and BLEU-4, respectively which demonstrates the effectiveness of our proposed approach in translating the Sylheti dialect to standard Bangla.