Cross-Lingual Transfer Learning in NLP: Enhancing English Language Learning for Non-Native Speakers

In the evolving landscape of language education, this paper delves into the intersection of Natural Language Processing (NLP) and education technology to address the unique challenges faced by non-native speakers, particularly children, in acquiring English proficiency. Leveraging the potential of Cross-Lingual Transfer Learning, the proposed methodology, implemented in Python, aims to enhance language learning outcomes through the innovative use of the NNCES (NNCES) corpus. The NNCES corpus, featuring 50 Telugu-speaking children aged 8 to 12 engaged in English language learning, serves as a rich dataset for exploring cross-lingual transfer learning strategies. The paper introduces Multilingual Transfer Learning with Domain Adaptation (MTL-DA) that fine-tunes pre-trained multilingual models on the NNCES corpus. This strategic adaptation enables models to discern linguistic nuances, phonetic variations, and semantic context inherent in NNCES. The methodology involves a comprehensive pipeline, encompassing dataset collection, preprocessing, feature extraction, and model training. Min-Max Normalization is applied to acoustic features, and Mel-frequency cepstral coefficients (MFCCs) and word embeddings from pre-trained models are integrated into a holistic feature vector. The fine-tuned models exhibit superior performance in English language learning tasks, showcasing an accuracy of $99.6 %$. Comparative analysis with existing methods reveals a significant improvement, with the proposed method surpassing others by $3.6 %$.