BERT (Bidirectional Encoder Representations from Transformers)

BERT, or Bidirectional Encoder Representations from Transformers, is a transformer architecture introduced by Google for understanding text. Unlike left-to-right generative models, BERT processes context on both sides of a token and is commonly pretrained by predicting masked words. The resulting encoder representations can be fine-tuned for classification, search, question answering, named entity recognition, and other language-understanding tasks. BERT helped establish large-scale pretraining followed by task adaptation as a standard NLP workflow. It is not primarily designed for free-form text generation, and later variants modify its size, training data, efficiency, vocabulary, or objectives to suit multilingual and domain-specific applications.