文献目録

1

Peter F. Brown, Stephen A. Della Pietra, Vincent J. Della Pietra, and Robert L. Mercer. "The machinematics of machine translation: Parameter estimation", Computational Linguistics, 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

Pierre Isabelle, Cyril Goutte, and Michel Simard., ``Domain Adaptation of MT systems through automatic post-editing'', MT Summit XI, 102, 2007.

4

Yushi Xu and Stephanie Seneff, "Two-Stage Translation: A Combined Linguistic and Statistical Machine Translation Framework", Proceedings of the Eighth Conference of the Association for Machine Translation (AMTA) 2008.

5

Jason Katz-Brown and Michael Collins, " Syntactic Reordering in Preprocessing for Japanese English Translation: MIT System Description for NTCIR-7 Patent Translation Task", Proceedings of the 7th NTCIR Workshop Meeting, 2008.

5

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

6

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

7

Moses, moses.2007-05-29.tgz,
http://www.statmt.org/moses/

8

NIST Open Machine Translation,
http://www.nist.gov/speech/tests/mt

9

The METEOR Automatic Machine Translation
Evaluation System,
http://www.cs.cmu.edu/ alavie/METEOR/

10

training-phrase-model.perl
Evaluation System,
http://www.cs.cmu.edu/ alavie/METEOR/

表 8: Results

TASK	BTEC_CE
case+punc	bleu	meteor	f1	prec	recl	wer	per	ter	gtm	nist
primary	0.3151	0.6169	0.6569	0.6465	0.6676	0.5590	0.4760	48.0710	0.6478	6.3834
contrastive1	0.3311	0.6109	0.6610	0.6758	0.6468	0.5377	0.4567	44.8140	0.6423	6.1511
contrastive2	0.1070	0.4697	0.5619	0.5671	0.5567	0.7017	0.6182	60.0070	0.4863	3.9727
TASK	CT_CE
case+punc	bleu	meteor	f1	prec	recl	wer	per	ter	gtm	nist
primary.CRR	0.2797	0.5971	0.6306	0.6092	0.6536	0.6590	0.5099	61.3850	0.6592	5.5309
contrastive1.CRR	0.2706	0.5881	0.6189	0.5945	0.6453	0.6712	0.5113	62.4990	0.6533	5.4633
contrastive2.CRR	0.0642	0.3953	0.4928	0.5051	0.4811	0.8046	0.6823	74.9560	0.4312	3.2979
primary.ASR.1	0.2482	0.5489	0.5910	0.5773	0.6053	0.6943	0.5456	64.8360	0.6136	5.0705
contrastive1.ASR.1	0.2650	0.5610	0.6000	0.5876	0.6128	0.6647	0.5220	62.0140	0.6307	5.2804
contrastive2.ASR.1	0.0602	0.3654	0.4644	0.4822	0.4479	0.8148	0.7018	76.1960	0.4009	2.9995
TASK	CT_EC
case+punc	bleu	meteor	f1	prec	recl	wer	per	ter	gtm	nist
primary.CRR	0.2759	0.5328	0.5500	0.5150	0.5900	0.7421	0.5382	68.6970	0.6914	5.3888
contrastive1.CRR	0.3391	0.5744	0.6204	0.6430	0.5994	0.5942	0.4356	52.3780	0.6930	6.1764
contrastive2.CRR	0.2300	0.5063	0.5596	0.5599	0.5594	0.6993	0.4987	63.2230	0.6304	5.4766
primary.ASR.1	0.2214	0.4417	0.4516	0.4100	0.5025	0.8518	0.6447	80.8210	0.6399	4.5091
contrastive1.ASR.1	0.2853	0.5134	0.5604	0.5784	0.5435	0.6609	0.4986	59.2510	0.6331	5.4212
contrastive2.ASR.1	0.1902	0.4483	0.4986	0.4948	0.5025	0.7627	0.5683	70.5120	0.5689	4.6699

primary: Proposed method contrastive1: Moses contrastive2: Systran

表 9: Appendix: Results with Parameter Tunings

TASK	BTEC_CE
case+punc	bleu	meteor	f1	prec	recl	wer	per	ter	gtm	nist
primary	0.3351	0.6256	0.6522	0.6301	0.6759	0.5704	0.4874	0.5048	0.6613	6.5972
contrastive1	0.3423	0.6135	0.6500	0.6463	0.6538	0.5436	0.4721	0.4674	0.6551	6.5624
contrastive2	0.107	0.4697	0.5619	0.5671	0.5567	0.7017	0.6182	60.007	0.4863	3.9727

primary: Proposed method contrastive1: Moses contrastive2: Systran