Natural Language Processing in the Era of Large Language Models: A Systematic Review of Methods and Applications

Abstract

Natural Language Processing (NLP) has undergone a substantial transformation with the emergence of large language models (LLMs), which have expanded the capabilities of machines to understand, generate, and interact with human language across increasingly complex tasks. This systematic review examines the evolution of NLP methods in the era of LLMs, with particular emphasis on model architectures, training and adaptation strategies, emerging methodological approaches, and applications across diverse domains. Relevant scholarly literature was systematically reviewed to identify major developments in transformer-based architectures, pre-training and fine-tuning, instruction tuning, reinforcement learning from human feedback, retrieval-augmented generation, prompt engineering, parameter-efficient adaptation, multimodal language processing, and agentic language systems. The findings indicate that LLMs have substantially improved performance in text generation, question answering, summarization, translation, information extraction, dialogue systems, sentiment analysis, and domain-specific language understanding. At the same time, the review identifies persistent challenges involving hallucination, factual reliability, bias, interpretability, computational cost, data privacy, security, evaluation consistency, and the representation of low-resource languages. Applications in healthcare, education, business, scientific research, public services, and software development demonstrate the growing versatility of LLM-based NLP systems, while also highlighting the importance of domain adaptation and human oversight. Overall, the evidence suggests that contemporary NLP is shifting from task-specific language models toward general-purpose, adaptable, and increasingly multimodal language systems. Future research should prioritize reliable and transparent evaluation, efficient model adaptation, multilingual and culturally responsive NLP, responsible deployment, and methods that improve factuality, safety, and human–AI collaboration.