Machine translation in natural language processing works by employing algorithms and models to convert text from one language to another automatically. Techniques like statistical and neural machine translation analyze vast datasets to learn the relationships between languages. The system breaks down sentences into components, translates them, and reconstructs them coherently in the target language. This process promotes cross-cultural communication, making information more accessible globally.