Improved Models of Distortion Cost for Statistical Machine Translation Spence Green, Michel Galley, and Christopher D. Manning Stanford University June 4, 2010
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivation Why phrase-based MT? ◮ Fast, simple, and scalable ◮ Good performance for many language pairs (Zollmann et al., 2008; Lopez, 2008; etc.) Reordering in (baseline) phrase-based decoders controlled by: ◮ A distortion cost model ◮ A distortion limit Green, Galley, and Manning Improved Models of Distortion Cost for SMT 2 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivation Why phrase-based MT? ◮ Fast, simple, and scalable ◮ Good performance for many language pairs (Zollmann et al., 2008; Lopez, 2008; etc.) Reordering in (baseline) phrase-based decoders controlled by: ◮ A distortion cost model ◮ A distortion limit Cost model is poor, so a low distortion limit is typically used Green, Galley, and Manning Improved Models of Distortion Cost for SMT 2 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivating Example ������ ���������������������������������������������������������������������������������� �� ������ �� ���#�� ���� �� �� ������� ����� �� ������� �������� � ���� �� ��������� !"��������� Green, Galley, and Manning Improved Models of Distortion Cost for SMT 3 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivating Example ������ ����������������������������������������������� ��� �����!�!������������!��������� �� ������ �� ������ ���� �� �� ������� ����� �� ������� �������� � ���� �� �������� ��������� "#$��������� ���������������� Green, Galley, and Manning Improved Models of Distortion Cost for SMT 4 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivating Example ������ ��������������������������������������������������������� � ������������ ��������� �� ������ �� ���$�� �������� ���� �� �� ������� ����� �� ������� �������� � ���� �� ��������� !"#��������� ���������������� ��������� Green, Galley, and Manning Improved Models of Distortion Cost for SMT 5 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivating Example ������ ����������������������������������������� ���������������!�!������������!��������� �� ������ �� ���%�� �������& ���� �� �� ������� ����� �� ������� �������� � ���� �� ��������� ������������ "#$��������� ���������������� ��������� Green, Galley, and Manning Improved Models of Distortion Cost for SMT 6 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivating Example ������ ��������������������������������������������������������� � ������������ ��������� �� ������ �� ���$�� �������� ���� �� �� ������� ����� �� ������� �������� � ���� �� ��������� !"#��������� ���������������� �� � �� Green, Galley, and Manning Improved Models of Distortion Cost for SMT 7 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivating Example ������ ��������������������������������������������������������� � ������������ ��������� �� ������ �� ���$�� �������� ���� �� �� ������� ����� �� ������� �������� � ���� �� ��������� !"#��������� ���������������� �� � �� ���� Green, Galley, and Manning Improved Models of Distortion Cost for SMT 8 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Motivating Example "����� ��������������������������� �������������#�����$���$������������������������������ �� ������ �� ���%�� ������! ���� �� �� ������� ����� �� ������� �������� � ���� �� ��������� ������������ ���������������� ������� ���� ���������� ��� Green, Galley, and Manning Improved Models of Distortion Cost for SMT 9 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Distortion Limit v. Distortion Cost Cost is a soft constraint ◮ Does not prune the search space ◮ Feature in the log-linear decoder framework Limit is a hard constraint ◮ Prunes translations from the search space Green, Galley, and Manning Improved Models of Distortion Cost for SMT 10 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Distortion Limit v. Distortion Cost Cost is a soft constraint ◮ Does not prune the search space ◮ Feature in the log-linear decoder framework Limit is a hard constraint ◮ Prunes translations from the search space For Moses, low(er) distortion limit improves translation quality! Green, Galley, and Manning Improved Models of Distortion Cost for SMT 10 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Translation Quality Decreases at High Distortion Limits Moses BLEU-4 Performance Arabic-English 44 . 0 43 . 0 42 0 2 4 6 8 10 Chinese-English 32 . 0 31 . 0 30 . 0 29 0 2 4 6 8 10 12 14 Distortion Limit Green, Galley, and Manning Improved Models of Distortion Cost for SMT 11 / 37
Motivation A New Cost Model Phrase-based MT Evaluation Limit v. Cost Conclusion Hard Constraints Reduce Reference Reachability Reference Reachability (%) dlimit = 15 35 dlimit = 12 30 dlimit = 9 25 dlimit = 6 20 15 10 20 220 420 620 820 Translation Option Limit (Auli, Lopez, Hoang, and Koehn, 2009) Green, Galley, and Manning Improved Models of Distortion Cost for SMT 12 / 37
Motivation Future Cost Estimation A New Cost Model Discriminative Distortion Cost Evaluation Implementation Conclusion A New Distortion Cost Model Guide search without hard constraints ◮ Maintain baseline performance at high distortion limits ◮ Solution: Improve heuristic search with future cost estimation (Moore and Quirk, 2007) Encourage linguistically-approriate reorderings ◮ Solution: Transition-based discriminative distortion model Worst-case O ( n ) cost computation ◮ Maintain linear running time of decoding! Green, Galley, and Manning Improved Models of Distortion Cost for SMT 13 / 37
Recommend
More recommend