Lattice-based Minimum Error Rate Training for Statistical Machine Translation

   Abstract