Scholarlink

<p>Multilingual conversations involving code-switching and code-mixing are increas-ingly prevalent in social media and digital communication. To understand the challenges, this research focuses on the issues of Language identification in multilingual situations, on the English–Kokborok language pair, a low-resource linguistic combination. To address these opportunities and challenges in the Kok-borok low-resource language, we have applied machine learning techniques, using Hidden Markov Models, Support Vector Machines, N-grams, and Long Short-Term Memory networks using 5800 manually annotated code-mixed sentences on social media platforms. Among the three models, the LSTM model has achieved an accuracy of 91.6% because the LSTM can handle intricate linguistic patterns. Support Vector Machine models also demonstrated simpler language structures with an accuracy of 82.1%. Whereas the Hidden Markov Model provides sequen-tial insights valuable for foundational analysis. By bridging gaps in multilingual NLP, this research contributes to building natural language processing tools that support global initiatives in linguistic inclusion and advancing digital tools for underrepresented languages like Kokborok.</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

Code-Switching Scientific Approaches Using Deep Learning for English–Kokborok Language

  • 期刊论文

  • 发表日期：

    2026年4月24日

查看全文
  • Enjula Uchoi,
  • Koj Sambyo

摘要

Multilingual conversations involving code-switching and code-mixing are increas-ingly prevalent in social media and digital communication. To understand the challenges, this research focuses on the issues of Language identification in multilingual situations, on the English–Kokborok language pair, a low-resource linguistic combination. To address these opportunities and challenges in the Kok-borok low-resource language, we have applied machine learning techniques, using Hidden Markov Models, Support Vector Machines, N-grams, and Long Short-Term Memory networks using 5800 manually annotated code-mixed sentences on social media platforms. Among the three models, the LSTM model has achieved an accuracy of 91.6% because the LSTM can handle intricate linguistic patterns. Support Vector Machine models also demonstrated simpler language structures with an accuracy of 82.1%. Whereas the Hidden Markov Model provides sequen-tial insights valuable for foundational analysis. By bridging gaps in multilingual NLP, this research contributes to building natural language processing tools that support global initiatives in linguistic inclusion and advancing digital tools for underrepresented languages like Kokborok.