A Maximum Entropy / Minimum Divergence Translation

  • ModelGeorge FosterRALI
  • Published 2000
I present empirical comparisons between a linear combination of standard statistical language and translation models and an equivalent Maximum Entropy/Minimum Divergence (MEMD) model, using several diierent methods for automatic feature selection. The MEMD model signiicantly outperforms the standard model in test corpus per-plexity, even though it has far… CONTINUE READING