Comparison to MT

• Why are MT alignments hard?

  • sometimes/quelquefois look nothing alike
  • Large scale reordering (SVO to SOV)

• Why are summarization alignments hard?

  • The company/Microsoft look nothing alike
  • HUGE scale reordering
  • Enormous search space (128x1500 words as opposed to 30x30)
  • Much less data (in the same domain), much more uncertainty