62% accuracy, uses transformer models.
   I started reviewing the paper with an eye toward why the accuracy was
   only 62%, but did not find their training information in the time
   provided.