Zero-shot Cross-lingual Transfer for Low Resource NLP

The rapid advancement of Deep Learning, particularly through Large Language Models (LLMs) such as GPT and BERT, has significantly transformed Natural Language Processing (NLP), enabling remarkable progress in tasks like machine translation, sentiment analysis, and text generation. However, these advancements are primarily driven by the availability of vast amounts of high-quality training data, a resource that is scarce for low-resource languages. This disparity has led to a digital divide where NLP technologies perform well for high-resource languages but poorly for low-resource ones, limiting the accessibility and effectiveness of these models for many linguistic communities. To address this challenge, our research explores the potential of zero-shot cross-lingual transfer learning, where models trained on high-resource languages are adapted to low-resource languages without requiring extensive datasets. The research focuses on adapting anchor-based methods to transform the latent embedding spaces of high- and low-resource languages into a common space, enhancing cross-lingual transfer. Additionally, the study examines architectural modifications and cross-lingual approaches to improve the performance of these models across various NLP tasks. Ultimately, the goal is to develop scalable and versatile methodologies that can be generalized to multiple low-resource languages, thereby bridging the performance gap and promoting more inclusive NLP development.