CL-CNN: A Low-Resource Character-Level CNN for Bangladeshi Ethnic Language Recognition
Students & Supervisors
Student Authors
Supervisors
Abstract
Bangladesh hosts a diverse set of ethnolinguistic communities; however, most NLP research has centered on Bangla, leaving minority languages underrepresented in digital resources. We study the automatic identification of ethnic languages using a lightweight character-level Convolutional Neural Network (CNN) designed for low-resource settings. Using the “Bd Ethnic Language Classification” dataset (Kag- gle), we constructed a character vocabulary and trained a 1D CNN with global max pooling and dense layers. Due to severe class sparsity, the Tripura subset (93 samples) was excluded to avoid extreme imbalance, and our target set comprised Chakma, Marma, Santali, Garo, and Rakhine. The proposed CNN attains 95.23% accuracy, 95.35% precision, 95.24% recall, and 95.23% F1, outperforming classical and neural baselines (SVM, Random Forest, BiLSTM). We an- alyze confusion patterns (e.g., Chakma-Marma) and discuss the implications for inclusive NLP, accessibility, and cultural preservation. Our study established a strong baseline for low-resource ethnic language identification and highlighted practical design choices for robust character-level models in multilingual environments.
Keywords
Publication Details
- Type of Publication:
- Conference Name: 2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking
- Date of Conference: 16/04/2026 - 16/04/2026
- Venue: IT Business Incubator, Chittagong University of Engineering and Technology (CUET), Chattogram, Bangladesh