Machine Translation
Automatic translation of text or speech between languages using AI, from early rule-based systems to modern neural approaches achieving near-human quality.
Evolution
Rule-based (1950s-1990s): Hand-crafted grammar rules. Statistical (2000s): Learning from parallel corpora. Neural (2014+): Sequence-to-sequence models. Transformer-based (2017+): Attention mechanisms revolutionized quality.
Current State
Google Translate, DeepL, and LLMs handle 100+ languages. High-resource language pairs (English-French) approach human quality. Low-resource languages remain challenging. LLMs can translate with cultural context and style preservation.
Challenges
Low-resource languages with little training data. Preserving idioms, humor, and cultural context. Document-level consistency (maintaining terminology). Real-time interpretation. Handling code-switching (mixing languages).