the fully annotated sentences are provided in a file of concatenated
  protocol buffers:

delimitedSentences.proto.bz

This file should be read with the Java function
  `CoreNLPProtos.Sentence.parseDelimitedFrom(<input stream>)`,
  or in other languages taking into consideration that every protocol buffer is
  prepended with the size of the buffer, as a VarInt.
Each proto contains all the annotations for the MIML-RE featurizer, in addition to
  some useful additions (e.g., antecedent for every token).