Spaces:
Edmond98
/
Running on A100

sts / uroman /lib /NLP /Chinese.pm
multimodalart's picture
First commit
7bcf8d7
raw
history blame
8.91 kB
################################################################
# #
# Chinese #
# #
################################################################
package NLP::Chinese;
$utf8 = NLP::UTF8;
%empty_ht = ();
sub read_chinese_tonal_pinyin_files {
local($caller, *ht, @filenames) = @_;
$n_kHanyuPinlu = 0;
$n_kXHC1983 = 0;
$n_kHanyuPinyin = 0;
$n_kMandarin = 0;
$n_cedict = 0;
$n_simple_pinyin = 0;
foreach $filename (@filenames) {
if ($filename =~ /unihan/i) {
my $line_number = 0;
if (open(IN, $filename)) {
while (<IN>) {
$line_number++;
next if /^#/;
s/\s*$//;
if (($u, $type, $value) = split(/\t/, $_)) {
if ($type =~ /^(kHanyuPinlu|kXHC1983|kHanyuPinyin|kMandarin)$/) {
$u = $util->trim($u);
$type = $util->trim($type);
$value = $util->trim($value);
$f = $utf8->unicode_string2string($u);
if ($type eq "kHanyuPinlu") {
$value =~ s/\(.*?\)//g;
$value = $util->trim($value);
$translit = $caller->number_to_accent_tone($value);
$ht{"kHanyuPinlu"}->{$f} = $translit;
$n_kHanyuPinlu++;
} elsif ($type eq "kXHC1983") {
@translits = ($value =~ /:(\S+)/g);
$translit = join(" ", @translits);
$ht{"kXHC1983"}->{$f} = $translit;
$n_kXHC1983++;
} elsif ($type eq "kHanyuPinyin") {
$value =~ s/^.*://;
$value =~ s/,/ /g;
$ht{"kHanyuPinyin"}->{$f} = $value;
$n_kHanyuPinyin++;
} elsif ($type eq "kMandarin") {
$ht{"kMandarin"}->{$f} = $value;
$n_kMandarin++;
}
}
}
}
close(IN);
print "Read in $n_kHanyuPinlu kHanyuPinlu, $n_kXHC1983 n_kXHC1983, $n_kHanyuPinyin n_kHanyuPinyin $n_kMandarin n_kMandarin\n";
} else {
print STDERR "Can't open $filename\n";
}
} elsif ($filename =~ /cedict/i) {
if (open(IN, $filename)) {
my $line_number = 0;
while (<IN>) {
$line_number++;
next if /^#/;
s/\s*$//;
if (($f, $translit) = ($_ =~ /^\S+\s+(\S+)\s+\[([^\[\]]+)\]/)) {
$translit = $utf8->extended_lower_case($translit);
$translit = $caller->number_to_accent_tone($translit);
$translit =~ s/\s//g;
if ($old_translit = $ht{"cedict"}->{$f}) {
# $ht{CONFLICT}->{("DUPLICATE " . $f)} = "CEDICT($f): $old_translit\nCEDICT($f): $translit (duplicate)\n" unless $translit eq $old_translit;
$ht{"cedicts"}->{$f} = join(" ", $ht{"cedicts"}->{$f}, $translit) unless $old_translit eq $translit;
} else {
$ht{"cedict"}->{$f} = $translit;
$ht{"cedicts"}->{$f} = $translit;
}
$n_cedict++;
}
}
close(IN);
# print "Read in $n_cedict n_cedict\n";
} else {
print STDERR "Can't open $filename";
}
} elsif ($filename =~ /chinese_to_pinyin/i) {
if (open(IN, $filename)) {
my $line_number = 0;
while (<IN>) {
$line_number++;
next if /^#/;
if (($f, $translit) = ($_ =~ /^(\S+)\t(\S+)\s*$/)) {
$ht{"simple_pinyin"}->{$f} = $translit;
$n_simple_pinyin++;
}
}
close(IN);
# print "Read in $n_simple_pinyin n_simple_pinyin\n";
} else {
print STDERR "Can't open $filename";
}
} else {
print STDERR "Don't know what to do with file $filename (in read_chinese_tonal_pinyin_files)\n";
}
}
}
sub tonal_pinyin {
local($caller, $s, *ht, $gloss) = @_;
return $result if defined($result = $ht{COMBINED}->{$s});
$cedict_pinyin = $ht{"cedict"}->{$s} || "";
$cedicts_pinyin = $ht{"cedicts"}->{$s} || "";
$unihan_pinyin = "";
@characters = $utf8->split_into_utf8_characters($s, "return only chars", *empty_ht);
foreach $c (@characters) {
if ($pinyin = $ht{"simple_pinyin"}->{$c}) {
$unihan_pinyin .= $pinyin;
} elsif ($pinyin = $ht{"kHanyuPinlu"}->{$c}) {
$pinyin =~ s/^(\S+)\s.*$/$1/;
$unihan_pinyin .= $pinyin;
} elsif ($pinyin = $ht{"kXHC1983"}->{$c}) {
$pinyin =~ s/^(\S+)\s.*$/$1/;
$unihan_pinyin .= $pinyin;
} elsif ($pinyin = $ht{"kHanyuPinyin"}->{$c}) {
$pinyin =~ s/^(\S+)\s.*$/$1/;
$unihan_pinyin .= $pinyin;
} elsif ($pinyin = $ht{"cedicts"}->{$c}) {
$pinyin =~ s/^(\S+)\s.*$/$1/;
$unihan_pinyin .= $pinyin;
# middle dot, katakana middle dot, multiplication sign
} elsif ($c =~ /^(\xC2\xB7|\xE3\x83\xBB|\xC3\x97)$/) {
$unihan_pinyin .= $c;
# ASCII
} elsif ($c =~ /^([\x21-\x7E])$/) {
$unihan_pinyin .= $c;
} else {
$unihan_pinyin .= "?";
$hex = $utf8->utf8_to_hex($c);
$unicode = uc $utf8->utf8_to_4hex_unicode($c);
# print STDERR "Tonal pinyin: Unknown character $c ($hex/U+$unicode) -> ?\n";
}
}
$pinyin_title = "";
if (($#characters >= 1) && $cedicts_pinyin) {
foreach $pinyin (split(/\s+/, $cedicts_pinyin)) {
$pinyin_title .= "$s $pinyin (CEDICT)\n";
}
$pinyin_title .= "\n";
}
foreach $c (@characters) {
my %local_ht = ();
@pinyins = ();
foreach $type (("kHanyuPinlu", "kXHC1983", "kHanyuPinyin", "cedicts")) {
if ($pinyin_s = $ht{$type}->{$c}) {
foreach $pinyin (split(/\s+/, $pinyin_s)) {
push(@pinyins, $pinyin) unless $util->member($pinyin, @pinyins);
$type2 = ($type eq "cedicts") ? "CEDICT" : $type;
$local_ht{$pinyin} = ($local_ht{$pinyin}) ? join(", ", $local_ht{$pinyin}, $type2) : $type2;
}
}
}
foreach $pinyin (@pinyins) {
$type_s = $local_ht{$pinyin};
$pinyin_title .= "$c $pinyin ($type_s)\n";
}
}
$pinyin_title =~ s/\n$//;
$pinyin_title =~ s/\n/&#xA;/g;
$unihan_pinyin = "" if $unihan_pinyin =~ /^\?+$/;
if (($#characters >= 1) && $cedict_pinyin && $unihan_pinyin && ($unihan_pinyin ne $cedict_pinyin)) {
$log = "Gloss($s): $gloss\nCEdict($s): $cedicts_pinyin\nUnihan($s): $unihan_pinyin\n";
foreach $type (("kHanyuPinlu", "kXHC1983", "kHanyuPinyin")) {
$log_line = "$type($s): ";
foreach $c (@characters) {
$pinyin = $ht{$type}->{$c} || "";
if ($pinyin =~ / /) {
$log_line .= "($pinyin)";
} elsif ($pinyin) {
$log_line .= $pinyin;
} else {
$log_line .= "?";
}
}
$log .= "$log_line\n";
}
$ht{CONFLICT}->{$s} = $log;
}
$result = $unihan_pinyin || $cedict_pinyin;
$result = $cedict_pinyin if ($#characters > 0) && $cedict_pinyin;
$ht{COMBINED}->{$s} = $result;
$ht{PINYIN_TITLE}->{$s} = $pinyin_title;
return $result;
}
%number_to_accent_tone_ht = (
"a1", "\xC4\x81", "a2", "\xC3\xA1", "a3", "\xC7\x8E", "a4", "\xC3\xA0",
"e1", "\xC4\x93", "e2", "\xC3\xA9", "e3", "\xC4\x9B", "e4", "\xC3\xA8",
"i1", "\xC4\xAB", "i2", "\xC3\xAD", "i3", "\xC7\x90", "i4", "\xC3\xAC",
"o1", "\xC5\x8D", "o2", "\xC3\xB3", "o3", "\xC7\x92", "o4", "\xC3\xB2",
"u1", "\xC5\xAB", "u2", "\xC3\xBA", "u3", "\xC7\x94", "u4", "\xC3\xB9",
"u:1","\xC7\x96", "u:2","\xC7\x98", "u:3","\xC7\x9A", "u:4","\xC7\x9C",
"\xC3\xBC1","\xC7\x96","\xC3\xBC2","\xC7\x98","\xC3\xBC3","\xC7\x9A","\xC3\xBC4","\xC7\x9C"
);
sub number_to_accent_tone {
local($caller, $s) = @_;
my $result = "";
while (($pre,$alpha,$tone_number,$rest) = ($s =~ /^(.*?)((?:[a-z]|u:|\xC3\xBC)+)([1-5])(.*)$/i)) {
if ($tone_number eq "5") {
$result .= "$pre$alpha";
} elsif ((($pre_acc,$acc_letter,$post_acc) = ($alpha =~ /^(.*)([ae])(.*)$/))
|| (($pre_acc,$acc_letter,$post_acc) = ($alpha =~ /^(.*)(o)(u.*)$/))
|| (($pre_acc,$acc_letter,$post_acc) = ($alpha =~ /^(.*)(u:|[iou]|\xC3\xBC)([^aeiou]*)$/))) {
$result .= "$pre$pre_acc" . ($number_to_accent_tone_ht{($acc_letter . $tone_number)} || ($acc_letter . $tone_number)) . $post_acc;
} else {
$result .= "$pre$alpha$tone_number";
}
$s = $rest;
}
$result .= $s;
$result =~ s/u:/\xC3\xBC/g;
return $result;
}
sub string_contains_utf8_cjk_unified_ideograph_p {
local($caller, $s) = @_;
return ($s =~ /([\xE4-\xE9]|\xE3[\x90-\xBF]|\xF0[\xA0-\xAC])/);
}
1;