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
|
#!/usr/bin/perl -w
use strict;
my $SCRIPT_DIR; BEGIN { use Cwd qw/ abs_path getcwd /; use File::Basename; $SCRIPT_DIR = dirname(abs_path($0)); push @INC, $SCRIPT_DIR; }
use Getopt::Long;
my $training_dir = "$SCRIPT_DIR/../training";
die "Can't find training dir: $training_dir" unless -d $training_dir;
my $mkcls = '/Users/cdyer/software/giza-pp/mkcls-v2/mkcls';
my $num_classes = 50;
my $nodes = 40;
my $TRAINING_ITERATIONS = 2000;
my $pmem = "2500mb";
my $DECODER = "cdec";
GetOptions("cdec=s" => \$DECODER,
"jobs=i" => \$nodes,
"pmem=s" => \$pmem,
"mkcls=s" => \$mkcls,
) or usage();
usage() unless (scalar @ARGV == 1);
die "Cannot find mkcls (specify with --mkcls=/path/to/mkcls) at $mkcls\n" unless -f $mkcls;
die "Cannot execute mkcls at $mkcls\n" unless -x $mkcls;
my $in_file = shift @ARGV;
die "Expected format corpus.l1-l2 where l1 & l2 are two-letter abbreviations\nfor the source and target language respectively\n" unless ($in_file =~ /^.+\.([a-z][a-z])-([a-z][a-z])$/);
my $f_lang = $1;
my $e_lang = $2;
print STDERR " Using mkcls in: $mkcls\n\n";
print STDERR "Source language: $f_lang\n";
print STDERR "Target language: $e_lang\n";
die "Don't have an stemmer for $f_lang: please create $SCRIPT_DIR/stemmers/$f_lang.pl\n" unless -f "$SCRIPT_DIR/stemmers/$f_lang.pl";
die "Don't have an stemmer for $e_lang: please create $SCRIPT_DIR/stemmers/$e_lang.pl\n" unless -f "$SCRIPT_DIR/stemmers/$e_lang.pl";
die "Don't have an orthographic normalizer for $f_lang: please create $SCRIPT_DIR/ortho-norm/$f_lang.pl\n" unless -f "$SCRIPT_DIR/ortho-norm/$f_lang.pl";
die "Don't have an orthographic normalizer for $e_lang: please create $SCRIPT_DIR/ortho-norm/$e_lang.pl\n" unless -f "$SCRIPT_DIR/ortho-norm/$e_lang.pl";
my @directions = qw(f-e);
my $corpus = 'c';
my $cwd = getcwd();
my $align_dir = "$cwd/talign";
mkdir $align_dir;
mkdir "$align_dir/grammars";
open IN, "<$in_file" or die "Can't read $in_file: $!";
open E, ">$align_dir/grammars/corpus.e" or die "Can't write: $!";
open F, ">$align_dir/grammars/corpus.f" or die "Can't write: $!";
while(<IN>) {
chomp;
my ($f, $e) = split / \|\|\| /;
die "Bad format, excepted ||| separated line" unless defined $f && defined $e;
print F "$f\n";
print E "$e\n";
}
close F;
close E;
close IN;
`cp $SCRIPT_DIR/makefiles/makefile.grammars $align_dir/grammars/Makefile`;
die unless $? == 0;
my @targets = qw(grammars);
for my $direction (@directions) {
push @targets, "model-$direction";
make_stage($direction);
}
open TOPLEVEL, ">$align_dir/Makefile" or die "Can't write $align_dir/Makefile: $!";
print TOPLEVEL <<EOT;
E_LANG = $e_lang
F_LANG = $f_lang
SCRIPT_DIR = $SCRIPT_DIR
TRAINING_DIR = $training_dir
MKCLS = $mkcls
NCLASSES = $num_classes
TARGETS = @targets
PTRAIN = \$(TRAINING_DIR)/cluster-ptrain.pl --restart_if_necessary
PTRAIN_PARAMS = --gaussian_prior --sigma_squared 1.0 --max_iteration 15
export
all:
\@failcom='exit 1'; \\
list='\$(TARGETS)'; for subdir in \$\$list; do \\
echo "Making \$\$subdir ..."; \\
(cd \$\$subdir && \$(MAKE)) || eval \$\$failcom; \\
done
clean:
\@failcom='exit 1'; \\
list='\$(TARGETS)'; for subdir in \$\$list; do \\
echo "Making \$\$subdir ..."; \\
(cd \$\$subdir && \$(MAKE) clean) || eval \$\$failcom; \\
done
EOT
close TOPLEVEL;
print STDERR "Created alignment task. chdir to talign/ then type make.\n\n";
exit 0;
sub make_stage {
my ($direction) = @_;
my $stage_dir = "$align_dir/model-$direction";
my $first = $direction;
my $second = $direction;
$first =~ s/^(.+)-.*$/$1/;
$second =~ s/^.+-(.+)$/$1/;
mkdir $stage_dir;
open CDEC, ">$stage_dir/cdec.ini" or die "Can't write $stage_dir/cdec.ini: $!";
print CDEC <<EOT;
formalism=lextrans
lextrans_use_null=true
intersection_strategy=full
grammar=$align_dir/grammars/corpus.$direction.lex-grammar.gz
# grammar=$align_dir/grammars/freq_grammar.$direction.gz
# per_sentence_grammar_file=$align_dir/grammars/psg.$direction
feature_function=WordPairFeatures $align_dir/grammars/wordpairs.$direction.features.gz
feature_function=LexicalPairIndicator
# stem translation
feature_function=LexicalPairIndicator S $align_dir/grammars/corpus.stemmed.$first $align_dir/grammars/${second}stem.map
# POS translation
feature_function=LexicalPairIndicator C $align_dir/grammars/corpus.class.$first $align_dir/grammars/voc2class.$second
feature_function=InputIndicator
feature_function=OutputIndicator
feature_function=RelativeSentencePosition $align_dir/grammars/corpus.class.$first
feature_function=LexNullJump
feature_function=NewJump
feature_function=IdentityCycleDetector
feature_function=NewJump use_binned_log_lengths flen
# jump distance and src and destination class type
feature_function=NewJump use_binned_log_lengths f0 fprev f:$align_dir/grammars/corpus.class.$first
feature_function=SourceBigram
feature_function=SourceBigram SC $align_dir/grammars/corpus.class.$first
EOT
close CDEC;
open AGENDA, ">$stage_dir/agenda.txt" or die "Can't write $stage_dir/agenda.txt: $!";
print AGENDA "cdec.ini $TRAINING_ITERATIONS\n";
close AGENDA;
}
sub usage {
die <<EOT;
Usage: $0 [OPTIONS] training_corpus.fr-en
EOT
}
|
