Greedy inference methods perform well for tokenizers, with SaGe outperforming others in morphological alignment.
