next up previous
Next: About this document ... Up: Statistical Machine Translation without Previous: Acknowledgements

Bibliography

1
Peter F. Brown, Stephen A. Della Pietra, Vincent J. Della Pietra, and Robert L. Mercer. "The machinematics of machine translation: Patamter estimation", Computational Linguestics, 19(2): pp. 263-311. (1993).

2
Philipp Koehn, Franz J. Och, and Daniel Marcu. " Statistical phrase-based translation". In Marti Hearst and Mari Ostendorf, editors, HLT-NAACL 2003: Main Proceedings, pages 127.133, Edmonton, Alberta, Canada, May 27 -June 1. Association for Computational Linguistics. (2003).

3
GIZA++, http://www.fjoch.com/GIZA++.html

4
moses, http://www.statmt.org/moses/

5
training-release-1.3.tgz, http://www.statmt.org/wmt06/shared-task/baseline.html

6
SRILM, The SRI Language Modeling Toolkit,
http://www.speech.sri.com/projects/srilm/


Table 10: Examples of phrase-tables
一 个 日语 导游 |||a Japanese speaking guide |||0.5 0.00339841 0.333333 0.00723042 2.718
一 个 日语 导游 吗 ? |||a Japanese speaking guide ? |||1 0.000771748 0.5 0.00668676 2.718
一 个 时钟 |||a clock |||0.5 0.0113267 1 0.293198 2.718
一 个 时钟 收音机 |||a clock radio |||1 0.00617817 1 0.293198 2.718
一 个 时钟 收音机 谢谢 |||a clock radio , please |||1 0.000602041 1 0.0325873 2.718
一 个 明天 十点 开始 的 |||a tee-off time for ten tomorrow |||1 0.000547434 1 3.02033e-05 2.718
一 个 明治 神殿 的 护身符 可以 预 知 |||A charm from Meiji shrine , a written oracle key holder |||1 3.76995e-05 1 7.77965e-08 2.718


Table 11: Results of experiments
TASK (case+punc) primary ASR.1 bleu nist wer per gtm meteor ter
  9#9 (no-case+ 9#9 contrast 9#9 CRR              
  no-punc)                  
BTEC-CE (case+punc) primary ASR.1 0.2911 6.0333 0.6208 0.5423 0.6013 0.4876 54.1800
      CRR 0.3266 6.4215 0.5873 0.5037 0.6418 0.5189 50.4520
| | |-|-|-|-|-|-|-|-|-|   contrast ASR.1 0.2757 5.8867 0.6360 0.5528 0.5924 0.4809 55.4560
      CRR 0.3185 6.3735 0.6017 0.5141 0.6333 0.5126 52.0960
| |-|-|-|-|-|-|-|-|-|-| (no-case primary ASR.1 0.3126 6.8900 0.6129 0.5164 0.6220 0.5162 53.6510
  +no-punc)   CRR 0.3490 7.4316 0.5703 0.4695 0.6677 0.5527 49.1490
| | |-|-|-|-|-|-|-|-|-|   contrast ASR.1 0.2932 6.7155 0.6338 0.5268 0.6151 0.5077 55.4810
      CRR 0.3366 7.3459 0.5928 0.4846 0.6600 0.5445 51.2980
CT-CE (case+punc) primary ASR.1 0.2331 4.3752 0.6493 0.5767 0.5479 0.4737 56.1700
      CRR 0.2653 4.7015 0.6226 0.5530 0.5750 0.4953 53.6100
| | |-|-|-|-|-|-|-|-|-|   contrast ASR.1 0.2140 4.1510 0.6708 0.5961 0.5278 0.4580 58.0620
      CRR 0.2573 4.5648 0.6385 0.5663 0.5613 0.4828 55.3700
| |-|-|-|-|-|-|-|-|-|-| (no-case primary ASR.1 0.2630 5.2135 0.6340 0.5333 0.5885 0.5219 54.5490
  +no-punc)   CRR 0.3034 5.7842 0.5975 0.5011 0.6272 0.5492 51.7170
| | |-|-|-|-|-|-|-|-|-|   contrast ASR.1 0.2425 5.0223 0.6553 0.5530 0.5738 0.5072 56.4160
      CRR 0.2897 5.6630 0.6105 0.5161 0.6156 0.5379 53.1980
CT-EC (case+punc) primary ASR.1 0.3482 5.6672 0.5441 0.4557 0.8558 0.5501 49.2630
      CRR 0.3975 6.1945 0.4885 0.3957 0.8521 0.5953 43.2030
| | |-|-|-|-|-|-|-|-|-|   contrast ASR.1 0.3553 5.9159 0.5422 0.4481 0.8594 0.5666 48.9160
      CRR 0.4047 6.4814 0.4785 0.3804 0.8612 0.6151 42.2910
| |-|-|-|-|-|-|-|-|-|-| (no-case primary ASR.1 0.3288 5.6212 0.5852 0.4904 0.8368 0.5324 52.4250
  +no-punc)   CRR 0.3827 6.1654 0.5258 0.4233 0.8343 0.5822 45.9450
| | |-|-|-|-|-|-|-|-|-|   contrast ASR.1 0.3324 5.8971 0.5850 0.4827 0.8415 0.5493 52.0320
      CRR 0.3856 6.4656 0.5162 0.4095 0.8448 0.6011 45.1350


Table 12: Delete unknown words
TASK (case+punc) primary ASR.1 bleu nist wer per gtm meteor ter
  9#9 (no-case+ 9#9 contrast 9#9 CRR              
  no-punc)                  
BTEC-CE (case+punc) primary CRR 0.3408 6.0026 0.5761 0.4984 0.6607 0.5350 -
| | |-|-|-|-|-|-|-|-|-|   contrast CRR 0.3329 5.9331 0.5883 0.5065 0.6536 0.5289 -
| |-|-|-|-|-|-|-|-|-|-| (no-case primary CRR 0.3705 7.1515 0.5529 0.4576 0.6904 0.5712 -
| | |-|-|-|-|-|-|-|-|-| +no-punc) contrast CRR 0.3592 7.0826 0.5721 0.4694 0.6844 0.5630 -
CT-EC (case+punc) primary CRR 0.4085 6.1261 0.4829 0.3922 0.8666 0.5973 -
| | |-|-|-|-|-|-|-|-|-|   contrast CRR 0.4132 6.4773 0.4718 0.3765 0.8724 0.6171 -
| |-|-|-|-|-|-|-|-|-|-| (no-case primary CRR 0.3969 6.0371 0.5184 0.4197 0.8505 0.5842 -
| | |-|-|-|-|-|-|-|-|-| +no-punc) contrast CRR 0.3945 6.4120 0.5103 0.4058 0.8573 0.6029 -


next up previous
Next: About this document ... Up: Statistical Machine Translation without Previous: Acknowledgements
Jin'ichi Murakami 2008-10-28