https://goodeda.github.io/posts/unigram-word-segmentation/