Navigating Linguistic Diversity: In-Context Learning and Prompt Engineering for Subjectivity Analysis in Low-Resource Languages

This research delves into subjectivity analysis through the lens of in-context learning and prompt engineering in low-resource languages, with a focus on Bengali, Gujarati, Hindi, Kannada, Maithili, Marathi, Tamil, Telugu, and Urdu. Leveraging Large Language Models GPT-4 and BARD, the study unveils critical insights for NLP applications. Language-specific prompts prove effective, enhancing adaptability in various NLP tasks. In-context learning’s influence is profound, emphasizing diverse training datasets’ necessity for nuanced expression understanding. Language-specific accuracy variations underscore tailored approaches’ importance. Additionally, degraded performance in less common languages prompts considerations for linguistic minority contexts.

Addressing challenges of data scarcity, cultural biases, and linguistic nuances, the research advocates for advanced data augmentation, adaptive fine-tuning, culture-aware embeddings, and context-aware prompts. Proposing new evaluation metrics for low-resource languages and ethical considerations stress fairness and inclusivity. The findings contribute to refining subjectivity analysis through Large Language Models, advocating language-specific strategies, and fostering ethical AI deployment in diverse linguistic landscapes.