Machine Translation: Evolution, Multilingual Models, Evaluation, and Challenges

Machine Translation (MT) is one of the most influential research areas in Natural Language Processing (NLP). The automated process of translating text from one language to another, has witnessed significant advancements in recent years. Over roughly seven decades it has moved from rule-based systems (RBMT) to statistical (SMT), neural (NMT) and Transformer-based models, and more recently to massively multilingual systems and Large Language Models (LLMs). This survey is written as an entry point for researchers new to the field. It reviews the main paradigms and their architectural milestones, massively multilingual systems, LLM-based translation, speech and image-grounded translation, and a dedicated review of machine translation for Indian languages covering early systems, parallel corpora, transliteration resources and Indic-specific evaluation. It also surveys automatic and human evaluation methods, and discusses open challenges including low-resource languages, hallucination, bias, document-level translation, evaluation validity and privacy. The narrative review includes selected literature available up to November 2026 and closes with future research directions.