Also very neat is the Unigram algorithm. IIRC that paper also introduces subword regularization which I think is really cool.

https://arxiv.org/abs/1804.10959