summaryrefslogtreecommitdiff
path: root/corpus
diff options
context:
space:
mode:
Diffstat (limited to 'corpus')
-rwxr-xr-xcorpus/support/fix-eos.pl10
-rwxr-xr-xcorpus/support/quote-norm.pl146
-rw-r--r--corpus/support/token_list59
-rwxr-xr-xcorpus/support/tokenizer.pl27
-rwxr-xr-xcorpus/tokenize-anything.sh1
5 files changed, 193 insertions, 50 deletions
diff --git a/corpus/support/fix-eos.pl b/corpus/support/fix-eos.pl
new file mode 100755
index 00000000..584f8b46
--- /dev/null
+++ b/corpus/support/fix-eos.pl
@@ -0,0 +1,10 @@
+#!/usr/bin/perl -w
+use strict;
+use utf8;
+
+binmode(STDIN, ":utf8");
+binmode(STDOUT, ":utf8");
+while(<STDIN>) {
+ s/(\p{Devanagari}{2}[A-Za-z0-9! ,.\@\p{Devanagari}]+?)\s+(\.)(\s*$|\s+\|\|\|)/$1 \x{0964}$3/s;
+ print;
+}
diff --git a/corpus/support/quote-norm.pl b/corpus/support/quote-norm.pl
index 57f4ad77..f677df66 100755
--- a/corpus/support/quote-norm.pl
+++ b/corpus/support/quote-norm.pl
@@ -7,31 +7,98 @@ binmode(STDOUT,"utf8");
while(<STDIN>) {
chomp;
$_ = " $_ ";
- s/&\s*lt\s*;/</gi;
- s/&\s*gt\s*;/>/gi;
- s/&\s*squot\s*;/'/gi;
- s/&\s*quot\s*;/"/gi;
- s/&\s*amp\s*;/&/gi;
- s/&\s*nbsp\s*;/&/gi;
- s/&\s*#\s*160\s*;/ /gi;
+
+ # Regularlize spaces:
+ s/\x{a0}/ /g; # non-breaking space
+ s/\x{2009}/ /g; # thin space
+ s/\x{2028}/ /g; # "line separator"
+ s/\x{2029}/ /g; # "paragraph separator"
+ s/\x{202a}/ /g; # "left-to-right embedding"
+ s/\x{202b}/ /g; # "right-to-left embedding"
+ s/\x{202c}/ /g; # "pop directional formatting"
+ s/\x{202d}/ /g; # "left-to-right override"
+ s/\x{202e}/ /g; # "right-to-left override"
+ s/\x{85}/ /g; # "next line"
+ s/\x{fffd}/ /g; # "replacement character"
+ s/\x{feff}/ /g; # byte-order mark
+ s/\x{fdd3}/ /g; # "unicode non-character"
+
+ # Regularize named HTML/XML escapes:
+ s/&\s*lt\s*;/</gi; # HTML opening angle bracket
+ s/&\s*gt\s*;/>/gi; # HTML closing angle bracket
+ s/&\s*squot\s*;/'/gi; # HTML single quote
+ s/&\s*quot\s*;/"/gi; # HTML double quote
+ s/&\s*nbsp\s*;/ /gi; # HTML non-breaking space
+ s/&apos;/\'/g; # HTML apostrophe
+ s/&\s*amp\s*;/&/gi; # HTML ampersand (last)
+
+ # Regularize known HTML numeric codes:
+ s/&\s*#\s*160\s*;/ /gi; # no-break space
+ s/&\s*#45\s*;\s*&\s*#45\s*;/--/g; # hyphen-minus hyphen-minus
+ s/&\s*#45\s*;/--/g; # hyphen-minus
+
+ # Convert arbitrary hex or decimal HTML entities to actual characters:
+ s/&\#x([0-9A-Fa-f]+);/pack("U", hex($1))/ge;
+ s/&\#([0-9]+);/pack("U", $1)/ge;
+
+ # Convert other Windows 1252 characters to UTF-8
+ s/\x{80}/\x{20ac}/g; # euro sign
+ s/\x{95}/\x{2022}/g; # bullet
+ s/\x{99}/\x{2122}/g; # trademark sign
+
+ # Currency and measure conversions:
s/ (\d\d): (\d\d)/ $1:$2/g;
s/[\x{20a0}]\x{20ac}]/ EUR /g;
s/[\x{00A3}]/ GBP /g;
s/(\W)([A-Z]+\$?)(\d*\.\d+|\d+)/$1$2 $3/g;
s/(\W)(euro?)(\d*\.\d+|\d+)/$1EUR $3/gi;
- s/&\s*#45\s*;\s*&\s*#45\s*;/--/g;
- s/&\s*#45\s*;/--/g;
- s/�c/--/g;
- s/ ,,/ "/g;
- s/„/"/g;
- s/``/"/g;
- s/''/"/g;
- s/[「」]/"/g;
- s/〃/"/g;
- s/¨/"/g;
+
+ # Ridiculous double conversions(?) (news commentary and Giga-FrEn):
+ s/�c/--/g; # long dash
+ s/\x{e2}\x{20ac}oe/\"/g; # opening double quote
+ s/\x{e2}\x{20ac}\x{9c}/\"/g; # opening double quote
+ s/\x{e2}\x{20ac}\x{9d}/\"/g; # closing double quote
+ s/\x{e2}\x{20ac}\x{2122}/\'/g; # apostrophe
+ s/\x{e2}\x{20ac}\x{201c}/ -- /g; # en dash?
+ s/\x{e2}\x{20ac}\x{201d}/ -- /g; # em dash?
+ s/â(\x{80}\x{99}|\x{80}\x{98})/'/g; # single quote?
+ s/â(\x{80}\x{9c}|\x{80}\x{9d})/"/g; # double quote?
+
+ # Regularize quotes:
+ s/ˇ/'/g; # caron
+ s/´/'/g; # acute accent
+ s/`/'/g; # grave accent
+ s/ˉ/'/g; # modified letter macron
+ s/ ,,/ "/g; # ghetto low-99 quote
+ s/``/"/g; # latex-style left quote
+ s/''/"/g; # latex-style right quote
+ s/\x{300c}/"/g; # left corner bracket
+ s/\x{300d}/"/g; # right corner bracket
+ s/\x{3003}/"/g; # ditto mark
+ s/\x{00a8}/"/g; # diaeresis
+ s/\x{92}/\'/g; # curly apostrophe
+ s/\x{2019}/\'/g; # curly apostrophe
+ s/\x{f03d}/\'/g; # curly apostrophe
+ s/\x{b4}/\'/g; # curly apostrophe
+ s/\x{2018}/\'/g; # curly single open quote
+ s/\x{201a}/\'/g; # low-9 quote
+ s/\x{93}/\"/g; # curly left quote
+ s/\x{201c}/\"/g; # curly left quote
+ s/\x{94}/\"/g; # curly right quote
+ s/\x{201d}/\"/g; # curly right quote
+ s/\x{2033}/\"/g; # curly right quote
+ s/\x{201e}/\"/g; # low-99 quote
+ s/\x{84}/\"/g; # low-99 quote (bad enc)
+ s/\x{201f}/\"/g; # high-rev-99 quote
+ s/\x{ab}/\"/g; # opening guillemet
+ s/\x{bb}/\"/g; # closing guillemet
+ s/\x{0301}/'/g; # combining acute accent
+
+ # Space inverted punctuation:
s/¡/ ¡ /g;
s/¿/ ¿ /g;
+ # Russian abbreviations:
s/ п. п. / п.п. /g;
s/ ст. л. / ст.л. /g;
s/ т. е. / т.е. /g;
@@ -45,24 +112,19 @@ while(<STDIN>) {
s/ т. н. / т.н. /g;
s/ т. ч. / т.ч. /g;
s/ н. э. / н.э. /g;
- # â<U+0080><U+0099>
- s/â(\x{80}\x{99}|\x{80}\x{98})/'/g;
- s/â(\x{80}\x{9c}|\x{80}\x{9d})/"/g;
- s/ˇ/'/g;
- s/´/'/g;
- s/`/'/g;
- s/’/'/g;
- s/ ́/'/g;
- s/‘/'/g;
- s/ˉ/'/g;
- s/β/ß/g; # WMT 2010 error
- s/“/"/g;
- s/”/"/g;
- s/«/"/g;
- s/»/"/g;
+
+ # Convert foreign numerals into Arabic numerals
+ tr/०-९/0-9/; # devangari
+ tr/౦-౯/0-9/; # telugu
+ tr/೦-೯/0-9/; # kannada
+ tr/೦-௯/0-9/; # tamil
+ tr/൦-൯/0-9/; # malayalam
+
+ # Random punctuation:
tr/！-～/!-~/;
s/、/,/g;
# s/。/./g;
+ s/\x{85}/.../g;
s/…/.../g;
s/―/--/g;
s/–/--/g;
@@ -77,11 +139,27 @@ while(<STDIN>) {
s/’/'/g;
s/â€"/"/g;
s/؛/;/g;
-
+
+ # Regularize ligatures:
+ s/\x{9c}/oe/g; # "oe" ligature
+ s/\x{0153}/oe/g; # "oe" ligature
+ s/\x{8c}/Oe/g; # "OE" ligature
+ s/\x{0152}/Oe/g; # "OE" ligature
+ s/\x{fb00}/ff/g; # "ff" ligature
+ s/\x{fb01}/fi/g; # "fi" ligature
+ s/\x{fb02}/fl/g; # "fl" ligature
+ s/\x{fb03}/ffi/g; # "ffi" ligature
+ s/\x{fb04}/ffi/g; # "ffl" ligature
+
+ s/β/ß/g; # WMT 2010 error
+
+ # Strip extra spaces:
s/\s+/ /g;
s/^\s+//;
s/\s+$//;
- s/[\x{00}-\x{1f}]//g;
+
+ # Delete control characters:
+ s/[\x{00}-\x{1f}]//g;
print "$_\n";
}
diff --git a/corpus/support/token_list b/corpus/support/token_list
index 43dd80d9..228663f6 100644
--- a/corpus/support/token_list
+++ b/corpus/support/token_list
@@ -1,6 +1,65 @@
##################### hyphenated words added by Fei since 3/7/05
##X-ray
+# hindi abbreviation patterns
+जन.
+फर.
+अग.
+सित.
+अक्टू.
+अक्तू.
+नव.
+दिस.
+डी.एल.
+डी.टी.ओ.
+डी.ए.
+ए.एस.आई.
+डी.टी.ओ.
+एम.एस.आर.टी.सी.
+बी.बी.एम.बी.
+डी.एस.पी.
+सी.आर.पी.
+एस.डी.एम.
+सी.डी.पी.ओ.
+बी.डी.ओ.
+एस.डी.ओ.
+एम.पी.पी.
+पी.एच.ई.
+एस.एच.ओ.
+ए.सी.पी.
+यू.पी.
+पी.एम.
+आर.बी.डी.
+वी.पी.
+सी.ए.डी.पी.
+ए.
+बी.
+सी.
+डी.
+ई.
+एफ.
+जी.
+एच.
+आई.
+जे.
+के.
+एल.
+एम.
+एन.
+ओ.
+पी.
+क़यू.
+आर.
+एस.
+टी.
+यू.
+वी.
+डबल्यू.
+एक्स.
+वाई.
+ज़ेड.
+ज़ी.
+
##################### words made of punct only
:-
:-)
diff --git a/corpus/support/tokenizer.pl b/corpus/support/tokenizer.pl
index e0df16a7..7771201f 100755
--- a/corpus/support/tokenizer.pl
+++ b/corpus/support/tokenizer.pl
@@ -65,7 +65,7 @@ my $Split_AposD = 1; ## 'd
### some patterns
-my $common_right_punc = '\.|\,|\;|:|\!|\?|\"|\)|\]|\}|\>|\-';
+my $common_right_punc = '\x{0964}|\.|\,|\;|\!|:|\?|\"|\)|\]|\}|\>|\-';
#### step 1: read files
@@ -112,7 +112,7 @@ my $new_token_total = 0;
while(<STDIN>){
chomp();
-
+ s/\x{0970}/./g; # dev abbreviation character
if(/^(\[b\s+|\]b|\]f|\[f\s+)/ || (/^\[[bf]$/) || (/^\s*$/) || /^<DOC/ || /^<\/DOC/) {
## markup
print STDOUT "$_\n";
@@ -121,7 +121,7 @@ while(<STDIN>){
my $orig_num = 0;
my $deep_proc_num = 0;
-
+ s/(\x{0964}+)/ $1/g; # Devangari end of sentence
my $new_line = proc_line($_, \$orig_num, \$deep_proc_num);
$orig_token_total += $orig_num;
@@ -148,7 +148,8 @@ while(<STDIN>){
$new_line =~ s/(set|src|tgt|trg)/ $1/g;
}
- print STDOUT " $new_line\n";
+ chomp $new_line;
+ print STDOUT "$new_line\n";
}
########################################################################
@@ -228,6 +229,7 @@ sub proc_token {
## step 1: check the most common case
if($token =~ /^[a-z0-9\p{Cyrillic}\p{Greek}\p{Hebrew}\p{Han}\p{Arabic}\p{Devanagari}]+$/i){
+ #if($token =~ /^[a-z0-9\p{Cyrillic}\p{Greek}\p{Hebrew}\p{Han}\p{Arabic}]+$/i){
### most common cases
return $token;
}
@@ -363,7 +365,7 @@ sub deep_proc_token {
##### step 0: if it mades up of all puncts, remove one punct at a time.
if($line !~ /[\p{Cyrillic}\p{Greek}\p{Hebrew}\p{Han}\p{Arabic}\p{Devanagari}a-zA-Z\d]/){
- if($line =~ /^(\!+|\@+|\++|\=+|\*+|\<+|\>+|\|+|\?+|\.+|\-+|\_+|\&+)$/){
+ if($line =~ /^(\!+|\@+|\++|\=+|\*+|\<+|\>+|\|+|\?+|\x{0964}+|\.+|\-+|\_+|\&+)$/){
## ++ @@@@ !!! ....
return $line;
}
@@ -454,7 +456,7 @@ sub deep_proc_token {
### deal with ': e.g., 's, 't, 'm, 'll, 're, 've, n't
## 'there => ' there '98 => the same
- $suc += ($line =~ s/^(\'+)([a-z]+)/ $1 $2/gi);
+ $suc += ($line =~ s/^(\'+)([a-z\p{Cyrillic}\p{Greek}\p{Hebrew}\p{Han}\p{Arabic}\p{Devanagari}]+)/ $1 $2/gi);
## note that \' and \. could interact: e.g., U.S.'s; 're.
if($Split_NAposT && ($line =~ /^(.*[a-z]+)(n\'t)([\.]*)$/i)){
@@ -664,10 +666,10 @@ sub deep_proc_token {
return $line;
}
- if($line =~ /^(([a-z]\.)+)(\.*)$/i){
+ if ($line =~ /^(([a-z]|ए|बी|सी|डी|ई|एफ|जी|एच|आई|जे|के|एल|एम|एन|ओ|पी|क़यू|आर|एस|टी|यू|वी|डबल्यू|एक्स|वाई|ज़ेड|ज़ी)(\.([a-z]|ए|बी|सी|डी|ई|एफ|जी|एच|आई|जे|के|एल|एम|एन|ओ|पी|क़यू|आर|एस|टी|यू|वी|डबल्यू|एक्स|वाई|ज़ेड|ज़ी))+)(\.?)(\.*)$/i){
## I.B.M.
- my $t1 = $1;
- my $t3 = $3;
+ my $t1 = $1 . $5;
+ my $t3 = $6;
return $t1 . " ". proc_token($t3);
}
@@ -701,10 +703,3 @@ sub deep_proc_token {
return $line;
}
-
-
-
-
-
-
-
diff --git a/corpus/tokenize-anything.sh b/corpus/tokenize-anything.sh
index 5b7933d8..bca954d1 100755
--- a/corpus/tokenize-anything.sh
+++ b/corpus/tokenize-anything.sh
@@ -14,6 +14,7 @@ fi
$SUPPORT/utf8-normalize.sh $NORMARGS |
$SUPPORT/quote-norm.pl |
$SUPPORT/tokenizer.pl |
+ $SUPPORT/fix-eos.pl |
sed $SEDFLAGS -e 's/ al - / al-/g' |
$SUPPORT/fix-contract.pl |
sed $SEDFLAGS -e 's/^ //' | sed $SEDFLAGS -e 's/ $//' |
generated by cgit v1.2.3 (git 2.39.1) at 2024-09-21 23:32:13 +0000