Leveraging Deep Learning Techniques for Objective and Subjective Assessment of English Language Proficiency
Jillellamoodi Naga Madhuri, Divya Nimma, M. Mythili, Punit Pathak, Veera Ankalu Vuyyuru, M. Karthik · 2024
Objective and subjective assessments are two common methods used to evaluate English language proficiency. In order to measure English language proficiency both objectively and subjectively, this research suggests using a hybrid CNN-BERT model. With the use of Bidirectional Encoder Representations from Transformers (BERT) and Convolutional Neural Networks (CNN), the framework seeks to offer a thorough assessment of English language competency that addresses either objective and subjective elements. Unlike conventional approaches, our method offers a comprehensive assessment by integrating subjective components (like spoken language analysis and reading comprehension) with objective assessments (like grammar and vocabulary). This dataset includes a wide range of text examples, covering various levels of proficiency and linguistic nuances. Next, the collected data is preprocessed using tokenization techniques, breaking down the text into smaller, manageable units. This preprocessing step enhances the efficiency and effectiveness of the subsequent model. Following data preprocessing, a hybrid CNN-BERT model is applied to the processed data. The CNN component captures local features, while the BERT model captures contextual information effectively. By combining the strengths of both CNN and BERT, the proposed model ensures a robust assessment of English language proficiency. The CNN-BERT model offers a dual assessment, covering both objective and subjective aspects of English language proficiency. The objective assessment focuses on grammatical accuracy, vocabulary usage, and syntactic structures. Meanwhile, the subjective assessment delves into fluency, coherence, and overall language proficiency. Experimental results demonstrate that the proposed CNN-BERT model outperforms existing approaches in both objective and subjective assessments. The model achieves 96.5% accuracy in evaluating English language proficiency, making it a valuable tool for language educators, researchers, and learners.