Sequence Models for Document Structure Identification in an Undeciphered Script