-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathpreprocessing.sh
266 lines (184 loc) · 9.54 KB
/
preprocessing.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
#!/bin/bash
# Data preprocessing
# - desgmise
# - train monolingual data
# - train parallel data
# - reference translations and MT outputs
# removes SGM markup
function desgmise {
INPUT=$1
echo "desgm $INPUT"
cat $INPUT | \
perl -ne 'print $1."\n" if /<seg[^>]+>\s*(.*)\s*<.seg>/i;' \
> "${INPUT%.*}"
}
# convert systems' outputs to plain text
function desgmises_systems {
for i in systems/*.sgm; do
desgmise $i
done
}
# convert reference translations to plain text
function desgmises_references {
for i in references/*.sgm; do
desgmise $i
done
}
# pre-processes monolingual data (will be used to build language models)
# tokenisation and true-casing
# NOTE: the data needs to be downloaded first with data_train_mono/get_data_train_mono.sh
function preprocess_train_mono {
echo "----- preprocess train mono -----"
echo " en"
zcat data_train_mono/news.2015.en.shuffled.gz | python code/tokenizer.py english \
> data_train_mono/news.2015.en.shuffled.tok
perl third/moses/recaser/train-truecaser.perl --model data_train_mono/truecasemodel.en --corpus data_train_mono/news.2015.en.shuffled.tok
cat data_train_mono/news.2015.en.shuffled.tok | perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.en \
| gzip > data_train_mono/news.2015.en.shuffled.tok.true.gz
echo " cs"
zcat data_train_mono/news.2015.cs.shuffled.gz | python code/tokenizer.py czech \
> data_train_mono/news.2015.cs.shuffled.tok
perl third/moses/recaser/train-truecaser.perl --model data_train_mono/truecasemodel.cs --corpus data_train_mono/news.2015.cs.shuffled.tok
cat data_train_mono/news.2015.cs.shuffled.tok | perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.cs \
| gzip > data_train_mono/news.2015.cs.shuffled.tok.true.gz
echo " de"
zcat data_train_mono/news.2015.de.shuffled.gz | python code/tokenizer.py german \
> data_train_mono/news.2015.de.shuffled.tok
perl third/moses/recaser/train-truecaser.perl --model data_train_mono/truecasemodel.de --corpus data_train_mono/news.2015.de.shuffled.tok
cat data_train_mono/news.2015.de.shuffled.tok | perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.de \
| gzip > data_train_mono/news.2015.de.shuffled.tok.true.gz
echo " fi"
zcat data_train_mono/news.2015.fi.shuffled.gz | python code/tokenizer.py finnish \
> data_train_mono/news.2015.fi.shuffled.tok
perl third/moses/recaser/train-truecaser.perl --model data_train_mono/truecasemodel.fi --corpus data_train_mono/news.2015.fi.shuffled.tok
cat data_train_mono/news.2015.fi.shuffled.tok | perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.fi \
| gzip > data_train_mono/news.2015.fi.shuffled.tok.true.gz
echo " ro"
zcat data_train_mono/news.2015.ro.shuffled.gz | perl third/moses/tokenizer/tokenizer.perl -l ro \
> data_train_mono/news.2015.ro.shuffled.tok
perl third/moses/recaser/train-truecaser.perl --model data_train_mono/truecasemodel.ro --corpus data_train_mono/news.2015.ro.shuffled.tok
cat data_train_mono/news.2015.ro.shuffled.tok | perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.ro \
| gzip > data_train_mono/news.2015.ro.shuffled.tok.true.gz
echo " ru"
zcat data_train_mono/news.2015.ru.shuffled.gz | perl third/moses/tokenizer/tokenizer.perl -l ru \
> data_train_mono/news.2015.ru.shuffled.tok
perl third/moses/recaser/train-truecaser.perl --model data_train_mono/truecasemodel.ru --corpus data_train_mono/news.2015.ru.shuffled.tok
cat data_train_mono/news.2015.ru.shuffled.tok | perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.ru \
| gzip > data_train_mono/news.2015.ru.shuffled.tok.true.gz
}
# pre-processes parallel data (will be used to build TODO)
# tokenisation and true-casing
# NOTE: the data needs to be downloaded first with data_train_parallel/get_data_train_parallel.sh
function preprocess_train_parallel {
echo "----- preprocess train parallel -----"
DATADIR=data_train_parallel
# NOTE some input files need to be passed through: fromdos and perl -p -i -e "s/\r//g" news-commentary-v11.de-en.?? and sed -i.old $'s/\xE2\x80\xA8/ /g' inFile
echo " deen, de"
cat $DATADIR/news-commentary-v11.de-en.de | head -n 100000 | python code/tokenizer.py german \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.de \
> $DATADIR/news-commentary-v11.tok.true.de-en.de
echo " deen, en"
cat $DATADIR/news-commentary-v11.de-en.en | head -n 100000 | python code/tokenizer.py english \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.en \
> $DATADIR/news-commentary-v11.tok.true.de-en.en
echo " fien, fi"
cat $DATADIR/europarl-v8.fi-en.fi | head -n 100000 | python code/tokenizer.py finnish \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.fi \
> $DATADIR/europarl-v8.tok.true.fi-en.fi
echo " fien, en"
cat $DATADIR/europarl-v8.fi-en.en | head -n 100000 | python code/tokenizer.py english \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.en \
> $DATADIR/europarl-v8.tok.true.fi-en.en
echo " roen, ro"
cat $DATADIR/europarl-v8.ro-en.ro | head -n 100000 \
| perl third/moses/tokenizer/tokenizer.perl -l ro \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.ro \
> $DATADIR/europarl-v8.tok.true.ro-en.ro
echo " roen, en"
cat $DATADIR/europarl-v8.ro-en.en | head -n 100000 | python code/tokenizer.py english \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.en \
> $DATADIR/europarl-v8.tok.true.ro-en.en
echo " ruen, ru"
perl -p -i -e "s/\r//g" data_train_parallel/news-commentary-v11.ru-en.??
sed -i.old $'s/\xE2\x80\xA8/ /g' data_train_parallel/news-commentary-v11.ru-en.en
cat $DATADIR/news-commentary-v11.ru-en.ru | head -n 100000 \
| perl third/moses/tokenizer/tokenizer.perl -l ru \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.ru \
> $DATADIR/news-commentary-v11.tok.true.ru-en.ru
echo " ruen, en"
cat $DATADIR/news-commentary-v11.ru-en.en | head -n 100000 | python code/tokenizer.py english \
| perl third/moses/recaser/truecase.perl --model data_train_mono/truecasemodel.en \
> $DATADIR/news-commentary-v11.tok.true.ru-en.en
}
# pre-processes reference translatoins and systems (i.e. MT outputs)
# tokenisation, true-casing and stemming (for Hjerson)
# NOTE: systems used... TODO
# All languages use NLTK tokenizer+stemmer, except:
# - Czech: stemmer http://research.variancia.com/czech_stemmer/
# - Romanian: tokenizer Moses v3
# - Russian: tokenizer Moses v3
function preprocess_test {
echo "----- preprocess test -----"
for i in references/*cs systems/encs-?mt?; do
echo "Preprocessing (tokenise + truecase + stem) $i"
cat $i | python code/tokenizer.py czech > $i.tok
cat $i.tok | perl third/moses/recaser/truecase.perl \
--model data_train_mono/truecasemodel.cs \
> $i.tok.true
cat $i.tok.true | python3 third/czech_stemmer_rev0/czech_stemmer.py light > $i.tok.true.base-light 2> $i.tok.true.base-light.log
cat $i.tok.true | python3 third/czech_stemmer_rev0/czech_stemmer.py aggressive > $i.tok.true.base-aggressive 2> $i.tok.true.base-aggressive.log
done
for i in references/*de systems/ende-?mt?; do
echo "Preprocessing (tokenise + truecase + stem) $i"
cat $i | python code/tokenizer.py german > $i.tok
cat $i.tok | python code/stemmer.py german > $i.tok.base
cat $i.tok | perl third/moses/recaser/truecase.perl \
--model data_train_mono/truecasemodel.de \
> $i.tok.true
cat $i.tok.true | python code/stemmer.py german > $i.tok.true.base
done
for i in references/*fi systems/enfi-?mt?; do
echo "Preprocessing (tokenise + truecase + stem) $i"
cat $i | python code/tokenizer.py finnish > $i.tok
cat $i.tok | python code/stemmer.py finnish > $i.tok.base
cat $i.tok | perl third/moses/recaser/truecase.perl \
--model data_train_mono/truecasemodel.fi \
> $i.tok.true
cat $i.tok.true | python code/stemmer.py finnish > $i.tok.true.base
done
# Romanian, Russian not in punkt, so we cannot use NLTK tokenizer!
for i in references/*ro systems/enro-?mt?; do
echo "Preprocessing (tokenise + truecase + stem) $i"
cat $i | perl third/moses/tokenizer/tokenizer.perl -l ro > $i.tok
cat $i.tok | python code/stemmer.py romanian > $i.tok.base
cat $i.tok | perl third/moses/recaser/truecase.perl \
--model data_train_mono/truecasemodel.ro \
> $i.tok.true
cat $i.tok.true | python code/stemmer.py romanian > $i.tok.true.base
done
for i in references/*ru systems/enru-?mt?; do
echo "Preprocessing (tokenise + truecase + stem) $i"
cat $i | perl third/moses/tokenizer/tokenizer.perl -l ru > $i.tok
cat $i.tok | python code/stemmer.py russian > $i.tok.base
cat $i.tok | perl third/moses/recaser/truecase.perl \
--model data_train_mono/truecasemodel.ru \
> $i.tok.true
cat $i.tok.true | python code/stemmer.py russian > $i.tok.true.base
done
for i in references/*en systems/??en-?mt?; do
echo "Preprocessing (tokenise + truecase + stem) $i"
cat $i | python code/tokenizer.py english > $i.tok
cat $i.tok | python code/stemmer.py english > $i.tok.base
cat $i.tok | perl third/moses/recaser/truecase.perl \
--model data_train_mono/truecasemodel.en \
> $i.tok.true
cat $i.tok.true | python code/stemmer.py english > $i.tok.true.base
done
}
# ----------- desgmise
desgmises_systems
desgmises_references
# ----------- data preprocessing
preprocess_train_mono
preprocess_train_parallel # run after preprocess_train_mono as it needs truecase models built there
preprocess_test # run after preprocess_train_mono as it needs truecase models built there