A neural network architecture with two parts: an encoder that processes input text and a decoder that generates output text, allowing the model to transform one sequence into another.