准备语料
/disknew/zw/moses/work/corpus translated.ench 1035224句已分词
数据预处理
过滤掉长句子
/tools/moses-scripts/scripts-20131121-1744/training/clean-corpus-n.perl/disknew/zw/moses/work/corpus/translated en chdisknew/zw/moses/work/corpus/translated.clean 1 100
单词小写处理
tools/scripts/lowercase.perl <disknew/zw/moses/work/corpus/translated.clean.en >disknew/zw/moses/work/corpus/translated.lowercased.en
tools/scripts/lowercase.perl <disknew/zw/moses/work/corpus/translated.clean.ch >disknew/zw/moses/work/corpus/translated.lowercased.ch
创建语言模型
准备训练数据
/tools/scripts/lowercase.perl <disknew/zw/moses/work/corpus/translated.ch >disknew/zw/moses/work/lm/translated.lowercased.ch
创建语言模型
/tools/Srilm/bin/i686-m64/ngram-count-order 5 -interpolate -wbdiscount -unk -textdisknew/zw/moses/work/lm/translated.lowercased.ch -lmdisknew/zw/moses/work/lm/translated.lm
训练短语模型
nohup nice/tools/moses-scripts/scripts-20131121-1744/training/train-model.perl-scripts-root-dir /tools/moses-scripts/scripts-20131121-1744 -root-dir/disknew/zw/moses/work -corpus/disknew/zw/moses/work/corpus/translated.lowercased -f en -e ch -alignmentgrow-diag-final-and -reordering msd-bidirectional-fe -lm 0:5:/disknew/zw/moses/work/lm/translated.lm>& disknew/zw/moses/work/training.out &
数据预处理
开发集/disknew/zw/moses/work/tuning/dev.ch dev.en
最小错误率训练
nohup nice/tools/moses-scripts/scripts-20131121-1744/training/mert-moses.pldisknew/zw/moses/work/tuning/dev.en /disknew/zw/moses/work/tuning/dev.en/tools/moses/moses-cmd/src/moses /disknew/zw/moses/work/model/moses.ini--working-dir /disknew/zw/moses/work/tuning/mert --rootdir/tools/moses-scripts/scripts-20131121-1744 --mertdir /tools/moses/mert--decoder-flags "-v 0" >&/disknew/zw/moses/work/tuning/mert.out &
过滤短语表
/tools/moses-scripts/scripts-20131121-1744/training/filter-model-given-input.pldisknew/zw/moses/work/evaluation/filterdisknew/zw/moses/work/tuning/mert/moses.inidisknew/zw/moses/work/evaluation/test.en
使用调参后的解码器进行解码
nohup nice /tools/moses/moses-cmd/src/moses-config disknew/zw/moses/work/evaluation/filter/moses.ini -input-filedisknew/zw/moses/work/evaluation/test.en 1>disknew/zw/moses/work/evaluation/filter/output 2>disknew/zw/moses/work/evaluation/filter/decode.out &
获取BLEU值
/disknew/sh2013/tools/BLEU./evaluation/test.en ./evaluation/test.cn ./evaluation/filter/output bleu-out