Recurrent neural networks are a subtype of neural networks that are recurring, in the sense that the next input of the neural network is based (partly) on the previous one's output. This lets it have a memory of sorts, where the neural network outputs a partial result, as well as its current "thoughts", which then the next step takes and processes again, resulting in a slightly more progressed result and slightly more in-depth "thoughts". This repeats many times over some length of time, until a final result is reached.