sentence-by-sentence.pl.svn-base

解码器是基于短语的统计机器翻译系统的核心模块

#!/usr/bin/perl -w

#sentence-by-sentence: take in a system output, with any number of factors, and a reference translation, also maybe with factors, and show each sentence and its errors
#usage: sentence-by-sentence SYSOUT [REFERENCE]+ > sentences.html

use strict;
use Getopt::Long;

my $sourcefile = undef;
my @truthfiles;
GetOptions(
	"source|s=s" => \$sourcefile,
	"reference|r=s" => \@truthfiles
) or exit(1);

my @sysoutfiles = @ARGV;
if (scalar(@sysoutfiles) == 0 || scalar(@truthfiles) == 0)
{
	print STDERR "usage: $0 system_output(s) > sentence-by-sentence.html
Options:
  --source,-s STRING      foreign input (can be used multiple times)
  --reference,-r STRING   English truth (can be used multiple times)

N-grams are colored by the number of supporting references:
 red for fewest, green for most, mediate shades otherwise.\n";
  exit(1);
}

####################################################################################################################

my @TRUTHS = () x scalar(@truthfiles);
for(my $i = 0; $i < scalar(@truthfiles); $i++)
{
	open($TRUTHS[$i], "<$truthfiles[$i]") or die "couldn't open '$truthfiles[$i]' for read: $!\n";
	binmode($TRUTHS[$i], ":utf8");
}
my @SYSOUTS = () x scalar(@sysoutfiles);
for(my $i = 0; $i < scalar(@sysoutfiles); $i++)
{
	open($SYSOUTS[$i], "<$sysoutfiles[$i]") or die "couldn't open '$sysoutfiles[$i]' for read: $!\n";
	binmode($SYSOUTS[$i], ":utf8");
}
binmode(STDOUT, ":utf8");
if (defined $sourcefile)
{
	open(SOURCE, "<$sourcefile") or die "couldn't open '$sourcefile' for read: $!\n";
	binmode(SOURCE, ":utf8");
}
my @bleuScores;
for(my $i = 0; $i < scalar(@sysoutfiles); $i++) {push @bleuScores, [];}
my @htmlSentences;
my @javascripts;
my @htmlColors = ('#99ff99', '#aaaaff', '#ffff99', '#ff9933', '#ff9999'); #color sentences by rank (split in n tiers)
my $ngramSingleRefColor = '#aaffaa';
my @ngramMultirefColors = ('#ff9999', '#ff9933', '#ffff99', '#a0a0ff', '#99ff99'); #arbitrary-length list; first entry is used for worst n-grams
my $numSentences = 0;
my (@sLines, @eLines);
while(readLines(\@SYSOUTS, \@sLines) && readLines(\@TRUTHS, \@eLines))
{
	#create array of lines of HTML
	my @html = ("<div class=\"sentence_%%%%\" id=\"sentence$numSentences\">"); #%%%% is a flag to be replaced

	my (@sFactors, @eFactors, $sourceFactors);
	#process source
	if (defined $sourcefile)
	{
		my $sourceLine = <SOURCE>;
		escapeMetachars($sourceLine); #remove inconsistencies in encoding
		$sourceFactors = extractFactorArrays($sourceLine);
		push @html, "<tr><td class=\"sent_title\">Source</td><td class=\"source_sentence\" id=\"source$numSentences\">" 
								. getFactoredSentenceHTML($sourceFactors) . "</td></tr>\n";
	}
	#process truth
	for(my $j = 0; $j < scalar(@truthfiles); $j++)
	{
		escapeMetachars($eLines[$j]); #remove inconsistencies in encoding
		push @eFactors, extractFactorArrays($eLines[$j]);
		push @html, "<tr><td class=\"sent_title\">Ref $j</td><td class=\"truth_sentence\" id=\"truth${numSentences}_$j\">" 
								. getFactoredSentenceHTML($eFactors[$j]) . "</td></tr>\n";
	}
	#process sysouts
	my @bleuData;
	for(my $j = 0; $j < scalar(@sysoutfiles); $j++)
	{
		escapeMetachars($sLines[$j]); #remove inconsistencies in encoding
		push @sFactors, extractFactorArrays($sLines[$j]);
		push @bleuData, getBLEUSentenceDetails($sFactors[$j], \@eFactors, 0);
		push @{$bleuScores[$j]}, [$numSentences, $bleuData[$j]->[0], 0]; #the last number will be the rank
		my $pwerData = getPWERSentenceDetails($sFactors[$j], \@eFactors, 0);
		push @html, "<tr><td class=\"sent_title\">Output $j</td><td class=\"sysout_sentence\" id=\"sysout$numSentences\">" 
								. getFactoredSentenceHTML($sFactors[$j], $pwerData) . "</td></tr>\n";
		push @html, "<tr><td class=\"sent_title\">N-grams</td><td class=\"sysout_ngrams\" id=\"ngrams$numSentences\">" 
								. getAllNgramsHTML($sFactors[$j], $bleuData[$j]->[1], scalar(@truthfiles)) . "</td></tr>\n";
	}
	splice(@html, 1, 0, "<div class=\"bleu_report\"><b>Sentence $numSentences)&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;BLEU:</b> " 
							. join("; ", map {sprintf("%.4lg", $_->[0]->[0]) . " (" . join('/', map {sprintf("%.4lg", $_)} @{$_->[0]}[1 .. 4]) . ") "} @bleuData) . "</div><table>\n");
	push @html, "</table></div>\n";
	push @htmlSentences, join('', @html);
	$numSentences++;
	@sLines = (); @eLines = (); #clear writable arrays to be refilled
}
foreach my $sysoutfh (@SYSOUTS) {close($sysoutfh);}
foreach my $truthfh (@TRUTHS) {close($truthfh);}

my $stylesheet = "<style type=\"text/css\">
.legend {background: #fff; border: 1px solid #000; padding: 2px; margin-bottom: 10px; margin-right: 15px}
.legend_title {font-weight: bold; font-size: medium; text-decoration: underline}
div.bleu_report {margin-bottom: 5px}
td.sent_title {font-weight: bold; font-size: medium; margin-bottom: 12px}
.source_sentence {background: #ffcccc; border: 1px solid #bbb}
.truth_sentence {background: #ccffcc; border: 1px solid #bbb}
.sysout_sentence {background: #ccccff; border: 1px solid #bbb}
table.sentence_table {border: none}
.sysout_ngrams {background: #fff; border: 1px solid #bbb}
table.ngram_table {}
td.ngram_cell {padding: 1px}\n";
for(my $i = 0; $i < scalar(@htmlColors); $i++)
{
	$stylesheet .= ".sentence_tier$i {background: $htmlColors[$i]; border: 1px solid #000088; padding: 0px 8px 0px 8px} //entire composition for a given sentence\n";
	$stylesheet .= "div.sentence_tier$i td {margin: 8px 0px 8px 0px}\n";
}
$stylesheet .= "</style>\n";

print "<html><head><meta http-equiv=\"Content-type: text/html; charset=utf-8\">\n";
print "<title>[" . join(', ', @sysoutfiles) . "] vs. [" . join(', ', @truthfiles) . "]: Sentence-by-Sentence Comparison</title>$stylesheet</head><body>\n";

foreach my $systemScores (@bleuScores) {rankSentencesByBLEU($systemScores);}
#javascript to sort by BLEU for any system, by order in corpus ...
print "<script type=\"text/javascript\">
var selectedSysout = 0; //index of system currently being used to rank/sort

function selectSysout(index)
{
	//update the BLEU-range text shown in the legend
	var legend = document.getElementById('legendBLEU');
	var rows = legend.getElementsByTagName('tr');
	for(var i = 0; i < rows.length; i++)
	{
		var cell = rows[i].childNodes[1];
		var spans = cell.getElementsByTagName('span');
		cell.childNodes[0].nodeValue = spans[index].firstChild.nodeValue; //something like '0.1 - 0.3'
	}
	
	//update the background colors of the sentence divs
	var allSentences = document.getElementById('all_sentences');
	var sentences = allSentences.childNodes;
	for(var i = 0; i < sentences.length; i++)
	{
		if(typeof sentences[i].tagName != 'undefined' && sentences[i].tagName.toLowerCase() == 'div') //text nodes have undefined tagName
		{
			var tierSpans = sentences[i].firstChild.childNodes;
			sentences[i].childNodes[2].className = tierSpans[index].firstChild.nodeValue; //something like 'tier3'
		}
	}
	selectedSysout = index; //selectedSysout is a flag to the sort functions
}

function sortByBLEU()
{
	var body = document.getElementById('all_sentences'); var row;
	switch(selectedSysout)
	{\n";
for(my $i = 0; $i < scalar(@sysoutfiles); $i++)
{
	print "case $i:
	{";
	my %rank2index = map {$bleuScores[$i]->[$_]->[2] => $_} (0 .. scalar(@htmlSentences) - 1);
	foreach my $rank (sort {$a <=> $b} keys %rank2index)
	{
		print "\trow = document.getElementById('everything" . $rank2index{$rank} . "');\n";
		print "\tbody.removeChild(row); body.appendChild(row);\n";
	}
	print "break;}\n";
}
print "}
}
function sortByCorpusOrder()
{
	var body = document.getElementById('all_sentences'); var row;\n";
for(my $j = 0; $j < scalar(@htmlSentences); $j++)
{
	print "\trow = document.getElementById('everything$j');\n";
	print "\tbody.removeChild(row); body.appendChild(row);\n";
}
print "}
</script>\n";

#legends for background colors of sentences and n-grams
my (@minBLEU, @maxBLEU);
my @bleuTiers = () x scalar(@htmlSentences); #for each sentence, arrayref of tier indices for each system
for(my $i = 0; $i < scalar(@sysoutfiles); $i++)
{
	my @a = (1e9) x scalar(@htmlColors);
	my @b = (-1e9) x scalar(@htmlColors);
	for(my $k = 0; $k < scalar(@htmlSentences); $k++)
	{
		my $tier = int($bleuScores[$i]->[$k]->[2] / (scalar(@htmlSentences) / scalar(@htmlColors)));
		push @{$bleuTiers[$k]}, $tier;
		if($bleuScores[$i]->[$k]->[1]->[0] < $a[$tier]) {$a[$tier] = $bleuScores[$i]->[$k]->[1]->[0];}
		if($bleuScores[$i]->[$k]->[1]->[0] > $b[$tier]) {$b[$tier] = $bleuScores[$i]->[$k]->[1]->[0];}
	}
	push @minBLEU, \@a;
	push @maxBLEU, \@b;
}
print "<table border=0><tr><td><div id=\"legendBLEU\" class=\"legend\"><span class=\"legend_title\">Sentence Background Colors => BLEU Ranges</span><table border=0>";
for(my $k = 0; $k < scalar(@htmlColors); $k++)
{
	print "<tr><td style=\"width: 15px; height: 15px; background: " . $htmlColors[$k] . "\"></td><td align=left style=\"padding-left: 12px\">" 
							. sprintf("%.4lg", $minBLEU[0]->[$k]) . " - " . sprintf("%.4lg", $maxBLEU[0]->[$k]);
	for(my $j = 0; $j < scalar(@sysoutfiles); $j++)
	{
		print "<span style=\"display: none\">" . sprintf("%.4lg", $minBLEU[$j]->[$k]) . " - " . sprintf("%.4lg", $maxBLEU[$j]->[$k]) . "</span>";
	}
	print "</td></tr>";
}
print "</table></div></td>\n";
print "<td><div class=\"legend\"><span class=\"legend_title\">N-gram Colors => Number of Matching Reference Translations</span><table border=0>";
for(my $k = 1; $k <= scalar(@truthfiles); $k++)
{
	print "<tr><td style=\"width: 15px; height: 15px; background: " . getNgramColorHTML($k, scalar(@truthfiles)) . "\"></td><td align=left style=\"padding-left: 12px\">$k</td></tr>";
}
print "</table></div></td></tr></table><div style=\"font-weight: bold; margin-bottom: 15px\">
PWER errors are marked in red on output sentence displays.</div>
<div style=\"margin-bottom: 8px\">Color by system # " 
						. join(' | ', map {"<a href=\"javascript:selectSysout($_);\">$_</a>" . (($_ == '0') ? " (default)" : "")} (0 .. scalar(@sysoutfiles) - 1)) . "</div>
<div style=\"margin-bottom: 8px\">Sort by <a href=\"javascript:sortByBLEU();\">BLEU score</a> | <a href=\"javascript:sortByCorpusOrder();\">corpus order</a> (default)</div>\n";

#sentence boxes
print "<div id=\"all_sentences\">";
for(my $j = 0; $j < scalar(@htmlSentences); $j++)
{
	print "<div id=\"everything$j\" style=\"margin: 0px; padding: 0px\">";
	print "<div class=\"ranks_container\" style=\"display: none\">" . join('', map {"<span>sentence_tier$_</span>"} @{$bleuTiers[$j]}) . "</div>";
	print "<hr width=98%>";
#	my $bgcolor = getSentenceBGColorHTML($bleuScores[0]->[$j], $i); #i is now # of sentences
	my $tierNum = $bleuTiers[$j]->[0];
	$htmlSentences[$j] =~ s/%%%%/tier$tierNum/;
	print "$htmlSentences[$j]</div>\n";
}
print "</div></body></html>";

##################### utils #####################

#arguments: a, b (scalars)
sub min
{
	my ($a, $b) = @_;
	return ($a < $b) ? $a : $b;
}
#arguments: a, b (scalars)
sub max
{
	my ($a, $b) = @_;
	return ($a > $b) ? $a : $b;
}
#arguments: a list of elements
#return undef for an empty list, the max element otherwise
sub maxN
{
	if(scalar @_ == 0) {return undef;}
	my $val = shift @_;
	foreach my $e (@_) {if($e > $val) {$val = $e;}}
	return $val;
}
#arguments: x
sub my_log
{
  return -9999999999 unless $_[0];
  return log($_[0]);
}
#arguments: x
sub round