Generative Adversarial Networks for Synthetic Data Augmentation in Low-Resource Language Modeling with Cross-Lingual Knowledge Transfer
Dr.E.M. Roopa DeviAssociate Professor, Department of Information Technology, Kongu Engineering College, Erode, Tamil Nadu, India. roopadevi.it@kongu.edu0000-0002-1127-2701
S. VinothkumarAssistant Professor (SRG), Department of Information Technology, Kongu Engineering College, Erode, Tamil Nadu, India. vinoths.it@kongu.edu0000-0002-1690-6654
Dr.B. VinodhiniAssociate Professor, Department of Computer Science and Engineering, SNS College of Technology, Coimbatore, Tamil Nadu, India. vinodhni.b.cse@snsct.org0009-0005-7762-5716
Keywords: Generative Adversarial Networks (GANs), Synthetic Data Augmentation, Low-Resource Language Modeling, Cross-Lingual Knowledge Transfer, Natural Language Processing (NLP), Machine Translation, Text Generation.
Abstract
Low-resource language modeling is a challenge addressed in this research using a Generative Adversarial Network (GAN) to generate synthetic data and cross-lingual knowledge transfer. Data scarcity is a challenge with low-resource languages that hinders the creation of high-performance natural language processing (NLP) models. The proposed approach is based on using GANs to create synthetic data similar in statistical characteristics to the real-world data, thereby expanding the data set and enhancing the accuracy of the model. Additionally, the model integrates cross-lingual knowledge transfer from HRLs, which further improves the transfer of linguistic features like syntax, grammar and semantics. The effectiveness of the model is showcased through the results and analysis, which show that the proposed GAN with cross-lingual transfer model outperforms the baseline model and other models in various metrics, including Perplexity (34.5), Accuracy (82.9%), F1 Score (79.8%), and BLEU (28.7%). These enhancements are amongst the model's greatest strengths in providing more fluent, semantically coherent, and relevant generated data than the baseline. The results highlight the potential of fusing GANs with cross-lingual knowledge transfer to improve the results in low-resource language tasks such as MT, sentiment analysis, and speech recognition. This will enable the development of more inclusive NLP technologies for traditionally underrepresented languages.