Recurrent neural network
A network that reads a sequence one step at a time, carrying a memory forward. The main language approach before transformers.
Also: RNN
A recurrent neural network processes text word by word, updating a hidden state that summarises everything read so far. RNNs powered speech recognition, translation and early neural language models in the 2010s. Their weaknesses were slow training, because each step waits for the last, and difficulty remembering things from far back. The transformer replaced them for most language work after 2017. The RWKV project and state-space models revived the idea of a recurrent state in forms that train in parallel.