I would like to extend this question and ask whether POS taggers are allowed in a constrained system as well as sentence parser.
I'm further not entirely sure to what extend manual work is permitted. In [1] frequent words have been checked manually and grouped together to form lists of words for different classes which imho does not comply with the rule of using only the provided data.
Best regards and thank you for clarification,
Stefan Falk