Permalink
Switch branches/tags
Find file Copy path
8aa8988 Jul 29, 2017
1 contributor

Users who have contributed to this file

executable file 1180 lines (1068 sloc) 40.8 KB
#!/usr/bin/env perl
use warnings;
use strict;
use utf8;
use Encode;
use XML::Twig;
use Sort::Naturally;
binmode STDOUT, ":utf8";
binmode STDERR, ":utf8";
#################################
# History:
#
# version 14
# (2016-03-29 lukas.diduch@nist.gov)
# * Fixed warning message in case seg-id is a string, by sorting in correct order using Sort::Naturally.
#
# version 13b
# * Fixed die 'bug' in case seg->id = 0
#
# version 13a
# * modified the scoring functions to prevent division-by-zero errors when a system segment is empty
# * affected methods: 'bleu_score' and 'bleu_score_smoothing'
#
# version 13
# * Uses a XML parser to read data (only when extension is .xml)
# * Smoothing of the segment-level BLEU scores, done by default
# * smoothing method similar to that of bleu-1.04.pl (IBM)
# * see comments above the 'bleu_score' method for more details on how the smoothing is computed
# * added a '--no-smoothing' option to simulate old scripts behavior
# * Introduction of the 'brevity-penalty' option, taking one of two values:
# * 'closest' (default) : act as IBM BLEU (taking the closest reference translation length)
# * in case two reference translations are at the same distance, will take the shortest one
# * for more details regarding how the BP is computed, see comments of the 'brevity_penalty_closest' function
# * 'shortest' : act as previous versions of the script (taking shortest reference translation length)
# * Introduction of the 'international-tokenization' option, boolean, disabled by default
# by default (when the option is not provided), uses 11b's tokenization function
# when option specified, uses v12's tokenization function
# * Introduction of a 'Metrics MATR output' flag (option '--metricsMATR')
# when used, creates three files for both BLEU score and NIST score:
# * BLEU-seg.scr and NIST-seg.scr: contain segment-level scores
# * BLEU-doc.scr and NIST-doc.scr: contain document-level scores
# * BLEU-sys.scr and NIST-sys.scr: contain system-level scores
# * SGML parsing
# * script will halt if source, reference and test files don't share the same setid attribute value (used for metricsMATR output)
# * correct segment IDs extracted from the files (was previously using an array, and using the index as a segID for output)
# * detailed output flag (-d) can now be used when running both BLEU and NIST
#
# version 12
# * Text normalization changes:
# * convert entity references (only the entities declared in the DTD)
# * now uses unicode categories
# * tokenize punctuation unless followed AND preceded by digits
# * tokenize symbols
# * UTF-8 handling:
# * files are now read using utf8 mode
# * Added the '-e' command-line option to enclose non-ASCII characters between spaces
#
# version 11b -- text normalization modified:
# * take out the join digit line because it joins digits
# when it shouldn't have
# $norm_text =~ s/(\d)\s+(?=\d)/$1/g; #join digits
#
# version 11a -- corrected output of individual n-gram precision values
#
# version 11 -- bug fixes:
# * make filehandle operate in binary mode to prevent Perl from operating
# (by default in Red Hat 9) in UTF-8
# * fix failure on joining digits
# version 10 -- updated output to include more details of n-gram scoring.
# Defaults to generate both NIST and BLEU scores. Use -b for BLEU
# only, use -n for NIST only
#
# version 09d -- bug fix (for BLEU scoring, ngrams were fixed at 4
# being the max, regardless what was entered on the command line.)
#
# version 09c -- bug fix (During the calculation of ngram information,
# each ngram was being counted only once for each segment. This has
# been fixed so that each ngram is counted correctly in each segment.)
#
# version 09b -- text normalization modified:
# * option flag added to preserve upper case
# * non-ASCII characters left in place.
#
# version 09a -- text normalization modified:
# * " and & converted to "" and &, respectively
# * non-ASCII characters kept together (bug fix)
#
# version 09 -- modified to accommodate sgml tag and attribute
# names revised to conform to default SGML conventions.
#
# version 08 -- modifies the NIST metric in accordance with the
# findings on the 2001 Chinese-English dry run corpus. Also
# incorporates the BLEU metric as an option and supports the
# output of ngram detail.
#
# version 07 -- in response to the MT meeting on 28 Jan 2002 at ISI
# Keep strings of non-ASCII characters together as one word
# (rather than splitting them into one-character words).
# Change length penalty so that translations that are longer than
# the average reference translation are not penalized.
#
# version 06
# Prevent divide-by-zero when a segment has no evaluation N-grams.
# Correct segment index for level 3 debug output.
#
# version 05
# improve diagnostic error messages
#
# version 04
# tag segments
#
# version 03
# add detailed output option (intermediate document and segment scores)
#
# version 02
# accommodation of modified sgml tags and attributes
#
# version 01
# same as bleu version 15, but modified to provide formal score output.
#
# original IBM version
# Author: Kishore Papineni
# Date: 06/10/2001
#################################
######
# Intro
my ($date, $time) = date_time_stamp();
print "MT evaluation scorer began on $date at $time\n";
print "\ncommand line: ", $0, " ", join(" ", @ARGV), "\n";
my $usage = "\n\nUsage: $0 -r <ref_file> -s <src_file> -t <tst_file>\n\n".
"Description: This Perl script evaluates MT system performance.\n".
"\n".
"Required arguments:\n".
" -r <ref_file> is a file containing the reference translations for\n".
" the documents to be evaluated.\n".
" -s <src_file> is a file containing the source documents for which\n".
" translations are to be evaluated\n".
" -t <tst_file> is a file containing the translations to be evaluated\n".
"\n".
"Optional arguments:\n".
" -h prints this help message to STDOUT\n".
" -c preserves upper-case alphabetic characters\n".
" -b generate BLEU scores only\n".
" -n generate NIST scores only\n".
" -d detailed output flag:\n".
" 0 (default) for system-level score only\n".
" 1 to include document-level scores\n".
" 2 to include segment-level scores\n".
" 3 to include ngram-level scores\n".
" -e enclose non-ASCII characters between spaces\n".
" --brevity-penalty ( closest | shortest )\n" .
" closest (default) : acts as IBM BLEU (takes the closest reference translation length)\n" .
" shortest : acts as previous versions of the script (takes the shortest reference translation length)\n" .
" --international-tokenization\n" .
" when specified, uses Unicode-based (only) tokenization rules\n" .
" when not specified (default), uses default tokenization (some language-dependant rules)\n" .
" --metricsMATR : create three files for both BLEU scores and NIST scores:\n" .
" BLEU-seg.scr and NIST-seg.scr : segment-level scores\n" .
" BLEU-doc.scr and NIST-doc.scr : document-level scores\n" .
" BLEU-sys.scr and NIST-sys.scr : system-level scores\n" .
" --no-smoothing : disable smoothing on BLEU scores\n" .
"\n";
use vars qw ($opt_r $opt_s $opt_t $opt_d $opt_h $opt_b $opt_n $opt_c $opt_x $opt_e);
use Getopt::Long;
my $ref_file = '';
my $src_file = '';
my $tst_file = '';
my $detail = 0;
my $help = '';
my $preserve_case = '';
my $split_non_ASCII = '';
my $brevity_penalty = 'closest';
my $international_tokenization;
my $metricsMATR_output = '';
my $no_smoothing = '';
our $opt_x = '';
our $opt_b = '';
our $opt_n = '';
GetOptions(
'r=s' => \$ref_file,
's=s' => \$src_file,
't=s' => \$tst_file,
'd:i' => \$detail,
'h|help' => \$help,
'b',
'n',
'c' => \$preserve_case,
'x:s',
'e' => \$split_non_ASCII,
'brevity-penalty:s' => \$brevity_penalty,
'international-tokenization' => \$international_tokenization,
'metricsMATR-output' => \$metricsMATR_output,
'no-smoothing' => \$no_smoothing
);
die $usage if $help;
die "Error in command line: ref_file not defined$usage" unless ( $ref_file );
die "Error in command line: src_file not defined$usage" unless ( $src_file );
die "Error in command line: tst_file not defined$usage" unless ( $tst_file );
my $BLEU_BP;
if ( !( $brevity_penalty cmp 'closest' ) )
{
$BLEU_BP = \&brevity_penalty_closest;
}
elsif ( !( $brevity_penalty cmp 'shortest' ) )
{
$BLEU_BP = \&brevity_penalty_shortest;
}
else
{
die "Incorrect value supplied for 'brevity_penalty'$usage";
}
my $TOKENIZATION = \&tokenization;
$TOKENIZATION = \&tokenization_international if ( $international_tokenization );
my $BLEU_SCORE = \&bleu_score;
$BLEU_SCORE = \&bleu_score_nosmoothing if ( $no_smoothing );
my $max_Ngram = 9;
my $METHOD = "BOTH";
if ( $opt_b ) { $METHOD = "BLEU"; }
if ( $opt_n ) { $METHOD = "NIST"; }
my $method;
######
# Global variables
my ($src_lang, $tgt_lang, @tst_sys, @ref_sys); # evaluation parameters
my (%tst_data, %ref_data); # the data -- with structure: {system}{document}{segments}
my ($src_id, $ref_id, $tst_id); # unique identifiers for ref and tst translation sets
my %eval_docs; # document information for the evaluation data set
my %ngram_info; # the information obtained from (the last word in) the ngram
######
# Get source document ID's
($src_id) = get_source_info ($src_file);
######
# Get reference translations
($ref_id) = get_MT_data (\%ref_data, "RefSet", $ref_file);
compute_ngram_info ();
######
# Get translations to evaluate
($tst_id) = get_MT_data (\%tst_data, "TstSet", $tst_file);
######
# Check data for completeness and correctness
check_MT_data ();
######
#
my %NISTmt;
my %NISTOverall;
my %BLEUmt;
my %BLEUOverall;
######
# Evaluate
print "\nEvaluation of $src_lang-to-$tgt_lang translation using:\n";
my $cum_seg = 0;
foreach my $doc (sort keys %eval_docs)
{
$cum_seg += scalar( keys( %{$eval_docs{$doc}{SEGS}} ) );
}
print " src set \"$src_id\" (", scalar keys %eval_docs, " docs, $cum_seg segs)\n";
print " ref set \"$ref_id\" (", scalar keys %ref_data, " refs)\n";
print " tst set \"$tst_id\" (", scalar keys %tst_data, " systems)\n\n";
foreach my $sys (sort @tst_sys)
{
for (my $n=1; $n<=$max_Ngram; $n++)
{
$NISTmt{$n}{$sys}{cum} = 0;
$NISTmt{$n}{$sys}{ind} = 0;
$BLEUmt{$n}{$sys}{cum} = 0;
$BLEUmt{$n}{$sys}{ind} = 0;
}
if ( ($METHOD eq "BOTH") || ($METHOD eq "NIST") )
{
$method="NIST";
score_system ($sys, \%NISTmt, \%NISTOverall);
}
if ( ($METHOD eq "BOTH") || ($METHOD eq "BLEU") )
{
$method="BLEU";
score_system ($sys, \%BLEUmt, \%BLEUOverall);
}
}
######
printout_report ();
if ( $metricsMATR_output )
{
outputMetricsMATR( 'NIST', %NISTOverall ) if ( ( $METHOD eq 'BOTH' ) || ( $METHOD eq 'NIST' ) );
outputMetricsMATR( 'BLEU', %BLEUOverall ) if ( ( $METHOD eq 'BOTH' ) || ( $METHOD eq 'BLEU' ) );
}
($date, $time) = date_time_stamp();
print "\nMT evaluation scorer ended on $date at $time\n";
exit 0;
#################################
sub get_source_info
{
my ($file) = @_;
my ($name, $id, $src, $doc, $seg);
my ($data, $tag, $span);
# Extension of the file determines the parser used:
# .xml : XML::Twig
# otherwise : simple SGML parsing functions
if ( $file =~ /\.xml$/i )
{
my $twig = XML::Twig->new();
$twig->parsefile( $file );
my $root = $twig->root;
my $currentSet = $root->first_child( 'srcset' );
die "Source XML file '$file' does not contain the 'srcset' element" if ( not $currentSet );
$id = $currentSet->{ 'att' }->{ 'setid' } or die "No 'setid' attribute value in '$file'";
$src = $currentSet->{ 'att' }->{ 'srclang' } or die "No srcset 'srclang' attribute value in '$file'";
die "Not the same srclang attribute values across sets" unless ( not defined $src_lang or $src eq $src_lang );
$src_lang = $src;
foreach my $currentDoc ( $currentSet->get_xpath( './/doc' ) )
{
my $docID = $currentDoc->{ 'att' }->{ 'docid' } or die "No document 'docid' attribute value in '$file'";
foreach my $currentSeg ( $currentDoc->get_xpath( './/seg' ) )
{
my $segID = $currentSeg->{ 'att' }->{ 'id' };
die "No segment 'id' attribute value in '$file'" if (! defined $segID);
my $segData = $currentSeg->text;
($eval_docs{$docID}{SEGS}{$segID}) = &{ $TOKENIZATION }( $segData );
}
}
}
else
{
#read data from file
open (FILE, $file) or die "\nUnable to open translation data file '$file'", $usage;
binmode FILE, ":utf8";
$data .= $_ while <FILE>;
close (FILE);
#get source set info
die "\n\nFATAL INPUT ERROR: no 'src_set' tag in src_file '$file'\n\n"
unless ($tag, $span, $data) = extract_sgml_tag_and_span ("SrcSet", $data);
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless ($id) = extract_sgml_tag_attribute ($name="SetID", $tag);
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless ($src) = extract_sgml_tag_attribute ($name="SrcLang", $tag);
die "\n\nFATAL INPUT ERROR: $name ('$src') in file '$file' inconsistent\n"
." with $name in previous input data ('$src_lang')\n\n"
unless (not defined $src_lang or $src eq $src_lang);
$src_lang = $src;
#get doc info -- ID and # of segs
$data = $span;
while (($tag, $span, $data) = extract_sgml_tag_and_span ("Doc", $data))
{
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless ($doc) = extract_sgml_tag_attribute ($name="DocID", $tag);
die "\n\nFATAL INPUT ERROR: duplicate '$name' in file '$file'\n\n"
if defined $eval_docs{$doc};
$span =~ s/[\s\n\r]+/ /g; # concatenate records
my $nseg=0, my $seg_data = $span;
while (($tag, $span, $seg_data) = extract_sgml_tag_and_span ("Seg", $seg_data))
{
die "\n\nFATAL INPUT ERROR: no attribute '$name' in file '$file'\n\n"
unless ($seg) = extract_sgml_tag_attribute( $name='id', $tag );
($eval_docs{$doc}{SEGS}{$seg}) = &{ $TOKENIZATION }( $span );
$nseg++;
}
die "\n\nFATAL INPUT ERROR: no segments in document '$doc' in file '$file'\n\n"
if $nseg == 0;
}
die "\n\nFATAL INPUT ERROR: no documents in file '$file'\n\n"
unless keys %eval_docs > 0;
}
return $id;
}
#################################
sub get_MT_data
{
my ($docs, $set_tag, $file) = @_;
my ($name, $id, $src, $tgt, $sys, $doc, $seg);
my ($tag, $span, $data);
# Extension of the file determines the parser used:
# .xml : XML::Twig
# otherwise : simple SGML parsing functions
if ( $file =~ /\.xml$/i )
{
my $twig = XML::Twig->new();
$twig->parsefile( $file );
my $root = $twig->root;
foreach my $currentSet ( $root->get_xpath( 'refset' ), $root->get_xpath( 'tstset' ) )
{
$id = $currentSet->{ 'att' }->{ 'setid' } or die "No 'setid' attribute value in '$file'";
$src = $currentSet->{ 'att' }->{ 'srclang' } or die "No 'srclang' attribute value in '$file'";
$tgt = $currentSet->{ 'att' }->{ 'trglang' } or die "No 'trglang' attribute value in '$file'";
die "Not the same 'srclang' attribute value across sets" unless ( $src eq $src_lang );
die "Not the same 'trglang' attribute value across sets" unless ( ( not defined $tgt_lang ) or ( $tgt = $tgt_lang ) );
$tgt_lang = $tgt;
my $sys;
if ( $currentSet->name eq 'tstset' )
{
$sys = $currentSet->{ 'att' }->{ 'sysid' } or die "No 'sysid' attribute value in '$file'";
}
else
{
$sys = $currentSet->{ 'att' }->{ 'refid' } or die "No 'refid' attribute value in '$file'";
}
foreach my $currentDoc ( $currentSet->get_xpath( './/doc' ) )
{
my $docID = $currentDoc->{ 'att' }->{ 'docid' } or die "No document 'docid' attribute value in '$file'";
$docs->{ $sys }{ $docID }{ FILE } = $file;
foreach my $currentSeg ( $currentDoc->get_xpath( './/seg' ) )
{
my $segID = $currentSeg->{ 'att' }->{ 'id' };
die "No segment 'id' attribute value in '$file'" if (! defined $segID);
my $segData = $currentSeg->text;
($docs->{$sys}{$docID}{SEGS}{$segID}) = &{ $TOKENIZATION }( $segData );
}
}
}
}
else
{
#read data from file
open (FILE, $file) or die "\nUnable to open translation data file '$file'", $usage;
binmode FILE, ":utf8";
$data .= $_ while <FILE>;
close (FILE);
#get tag info
while (($tag, $span, $data) = extract_sgml_tag_and_span ($set_tag, $data))
{
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless ($id) = extract_sgml_tag_attribute ($name="SetID", $tag);
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless ($src) = extract_sgml_tag_attribute ($name="SrcLang", $tag);
die "\n\nFATAL INPUT ERROR: $name ('$src') in file '$file' inconsistent\n"
." with $name of source ('$src_lang')\n\n"
unless $src eq $src_lang;
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless ($tgt) = extract_sgml_tag_attribute ($name="TrgLang", $tag);
die "\n\nFATAL INPUT ERROR: $name ('$tgt') in file '$file' inconsistent\n"
." with $name of the evaluation ('$tgt_lang')\n\n"
unless (not defined $tgt_lang or $tgt eq $tgt_lang);
$tgt_lang = $tgt;
my $mtdata = $span;
while (($tag, $span, $mtdata) = extract_sgml_tag_and_span ("Doc", $mtdata))
{
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless (my $sys) = extract_sgml_tag_attribute ($name="SysID", $tag);
die "\n\nFATAL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless $doc = extract_sgml_tag_attribute ($name="DocID", $tag);
die "\n\nFATAL INPUT ERROR: document '$doc' for system '$sys' in file '$file'\n"
." previously loaded from file '$docs->{$sys}{$doc}{FILE}'\n\n"
unless (not defined $docs->{$sys}{$doc});
$span =~ s/[\s\n\r]+/ /g; # concatenate records
my $nseg=0, my $seg_data = $span;
while (($tag, $span, $seg_data) = extract_sgml_tag_and_span ("Seg", $seg_data))
{
die "\n\nFATAIL INPUT ERROR: no tag attribute '$name' in file '$file'\n\n"
unless $seg = extract_sgml_tag_attribute( $name="id", $tag );
($docs->{$sys}{$doc}{SEGS}{$seg}) = &{ $TOKENIZATION }( $span );
$nseg++;
}
die "\n\nFATAL INPUT ERROR: no segments in document '$doc' in file '$file'\n\n" if $nseg == 0;
$docs->{$sys}{$doc}{FILE} = $file;
}
}
}
return $id;
}
#################################
sub check_MT_data
{
@tst_sys = sort keys %tst_data;
@ref_sys = sort keys %ref_data;
die "Not the same 'setid' attribute values across files" unless ( ( $src_id eq $tst_id ) && ( $src_id eq $ref_id ) );
#every evaluation document must be represented for every system and every reference
foreach my $doc (sort keys %eval_docs)
{
my $nseg_source = scalar( keys( %{$eval_docs{$doc}{SEGS}} ) );
foreach my $sys (@tst_sys)
{
die "\n\nFATAL ERROR: no document '$doc' for system '$sys'\n\n" unless defined $tst_data{$sys}{$doc};
my $nseg = scalar( keys( %{$tst_data{$sys}{$doc}{SEGS}} ) );
die "\n\nFATAL ERROR: translated documents must contain the same # of segments as the source, but\n"
." document '$doc' for system '$sys' contains $nseg segments, while\n"
." the source document contains $nseg_source segments.\n\n"
unless $nseg == $nseg_source;
}
foreach my $sys (@ref_sys)
{
die "\n\nFATAL ERROR: no document '$doc' for reference '$sys'\n\n" unless defined $ref_data{$sys}{$doc};
my $nseg = scalar( keys( %{$ref_data{$sys}{$doc}{SEGS}} ) );
die "\n\nFATAL ERROR: translated documents must contain the same # of segments as the source, but\n"
." document '$doc' for system '$sys' contains $nseg segments, while\n"
." the source document contains $nseg_source segments.\n\n"
unless $nseg == $nseg_source;
}
}
}
#################################
sub compute_ngram_info
{
my ($ref, $doc, $seg);
my (@wrds, $tot_wrds, %ngrams, $ngram, $mgram);
my (%ngram_count, @tot_ngrams);
foreach $ref (keys %ref_data)
{
foreach $doc (keys %{$ref_data{$ref}})
{
foreach $seg ( keys %{$ref_data{$ref}{$doc}{SEGS}})
{
@wrds = split /\s+/, $ref_data{ $ref }{ $doc }{ SEGS }{ $seg };
$tot_wrds += @wrds;
%ngrams = %{Words2Ngrams (@wrds)};
foreach $ngram (keys %ngrams)
{
$ngram_count{$ngram} += $ngrams{$ngram};
}
}
}
}
foreach $ngram (keys %ngram_count)
{
@wrds = split / /, $ngram;
pop @wrds, $mgram = join " ", @wrds;
$ngram_info{$ngram} = - log ($mgram ? $ngram_count{$ngram}/$ngram_count{$mgram} : $ngram_count{$ngram}/$tot_wrds) / log 2;
if (defined $opt_x and $opt_x eq "ngram info")
{
@wrds = split / /, $ngram;
printf "ngram info:%9.4f%6d%6d%8d%3d %s\n", $ngram_info{$ngram}, $ngram_count{$ngram},
$mgram ? $ngram_count{$mgram} : $tot_wrds, $tot_wrds, scalar @wrds, $ngram;
}
}
}
#################################
sub score_system
{
my ($sys, $ref, $doc, $SCOREmt, $overallScore);
($sys, $SCOREmt, $overallScore) = @_;
my ($ref_length, $match_cnt, $tst_cnt, $ref_cnt, $tst_info, $ref_info);
my ($cum_ref_length, @cum_match, @cum_tst_cnt, @cum_ref_cnt, @cum_tst_info, @cum_ref_info);
$cum_ref_length = 0;
for (my $j=1; $j<=$max_Ngram; $j++)
{
$cum_match[$j] = $cum_tst_cnt[$j] = $cum_ref_cnt[$j] = $cum_tst_info[$j] = $cum_ref_info[$j] = 0;
}
foreach $doc (sort keys %eval_docs)
{
($ref_length, $match_cnt, $tst_cnt, $ref_cnt, $tst_info, $ref_info) = score_document ($sys, $doc, $overallScore);
if ( $method eq "NIST" )
{
my %DOCmt = ();
my $docScore = nist_score( scalar( @ref_sys ), $match_cnt, $tst_cnt, $ref_cnt, $tst_info, $ref_info, $sys, \%DOCmt );
$overallScore->{ $sys }{ 'documents' }{ $doc }{ 'score' } = $docScore;
if ( $detail >= 1 )
{
printf "$method score using 5-grams = %.4f for system \"$sys\" on document \"$doc\" (%d segments, %d words)\n",
$docScore, scalar keys %{$tst_data{$sys}{$doc}{SEGS}}, $tst_cnt->[1];
}
}
if ( $method eq "BLEU" )
{
my %DOCmt = ();
my $docScore = &{$BLEU_SCORE}( $ref_length, $match_cnt, $tst_cnt, $sys, \%DOCmt );
$overallScore->{ $sys }{ 'documents' }{ $doc }{ 'score' } = $docScore;
if ( $detail >= 1 )
{
printf "$method score using 4-grams = %.4f for system \"$sys\" on document \"$doc\" (%d segments, %d words)\n",
$docScore, scalar keys %{$tst_data{$sys}{$doc}{SEGS}}, $tst_cnt->[1];
}
}
$cum_ref_length += $ref_length;
for (my $j=1; $j<=$max_Ngram; $j++)
{
$cum_match[$j] += $match_cnt->[$j];
$cum_tst_cnt[$j] += $tst_cnt->[$j];
$cum_ref_cnt[$j] += $ref_cnt->[$j];
$cum_tst_info[$j] += $tst_info->[$j];
$cum_ref_info[$j] += $ref_info->[$j];
printf "document info: $sys $doc %d-gram %d %d %d %9.4f %9.4f\n", $j, $match_cnt->[$j],
$tst_cnt->[$j], $ref_cnt->[$j], $tst_info->[$j], $ref_info->[$j]
if (defined $opt_x and $opt_x eq "document info");
}
}
if ($method eq "BLEU")
{
$overallScore->{ $sys }{ 'score' } = &{$BLEU_SCORE}($cum_ref_length, \@cum_match, \@cum_tst_cnt, $sys, $SCOREmt);
}
if ($method eq "NIST")
{
$overallScore->{ $sys }{ 'score' } = nist_score (scalar @ref_sys, \@cum_match, \@cum_tst_cnt, \@cum_ref_cnt, \@cum_tst_info, \@cum_ref_info, $sys, $SCOREmt);
}
}
#################################
sub score_document
{
my ($sys, $ref, $doc, $overallScore);
($sys, $doc, $overallScore) = @_;
my ($ref_length, $match_cnt, $tst_cnt, $ref_cnt, $tst_info, $ref_info);
my ($cum_ref_length, @cum_match, @cum_tst_cnt, @cum_ref_cnt, @cum_tst_info, @cum_ref_info);
$cum_ref_length = 0;
for (my $j=1; $j<=$max_Ngram; $j++)
{
$cum_match[$j] = $cum_tst_cnt[$j] = $cum_ref_cnt[$j] = $cum_tst_info[$j] = $cum_ref_info[$j] = 0;
}
# score each segment
foreach my $seg ( nsort keys( %{$tst_data{$sys}{$doc}{SEGS}} ) )
{
my @ref_segments = ();
foreach $ref (@ref_sys)
{
push @ref_segments, $ref_data{$ref}{$doc}{SEGS}{$seg};
if ( $detail >= 3 )
{
printf "ref '$ref', seg $seg: %s\n", $ref_data{$ref}{$doc}{SEGS}{$seg}
}
}
printf "sys '$sys', seg $seg: %s\n", $tst_data{$sys}{$doc}{SEGS}{$seg} if ( $detail >= 3 );
($ref_length, $match_cnt, $tst_cnt, $ref_cnt, $tst_info, $ref_info) = score_segment ($tst_data{$sys}{$doc}{SEGS}{$seg}, @ref_segments);
if ( $method eq "BLEU" )
{
my %DOCmt = ();
my $segScore = &{$BLEU_SCORE}($ref_length, $match_cnt, $tst_cnt, $sys, %DOCmt);
$overallScore->{ $sys }{ 'documents' }{ $doc }{ 'segments' }{ $seg }{ 'score' } = $segScore;
if ( $detail >= 2 )
{
printf " $method score using 4-grams = %.4f for system \"$sys\" on segment $seg of document \"$doc\" (%d words)\n", $segScore, $tst_cnt->[1]
}
}
if ( $method eq "NIST" )
{
my %DOCmt = ();
my $segScore = nist_score (scalar @ref_sys, $match_cnt, $tst_cnt, $ref_cnt, $tst_info, $ref_info, $sys, %DOCmt);
$overallScore->{ $sys }{ 'documents' }{ $doc }{ 'segments' }{ $seg }{ 'score' } = $segScore;
if ( $detail >= 2 )
{
printf " $method score using 5-grams = %.4f for system \"$sys\" on segment $seg of document \"$doc\" (%d words)\n", $segScore, $tst_cnt->[1];
}
}
$cum_ref_length += $ref_length;
for (my $j=1; $j<=$max_Ngram; $j++)
{
$cum_match[$j] += $match_cnt->[$j];
$cum_tst_cnt[$j] += $tst_cnt->[$j];
$cum_ref_cnt[$j] += $ref_cnt->[$j];
$cum_tst_info[$j] += $tst_info->[$j];
$cum_ref_info[$j] += $ref_info->[$j];
}
}
return ($cum_ref_length, [@cum_match], [@cum_tst_cnt], [@cum_ref_cnt], [@cum_tst_info], [@cum_ref_info]);
}
###############################################################################################################################
# function returning the shortest reference length
# takes as input:
# - currentLength : the current (shortest) reference length
# - referenceSentenceLength : the current reference sentence length
# - candidateSentenceLength : the current candidate sentence length (unused)
###############################################################################################################################
sub brevity_penalty_shortest
{
my ( $currentLength, $referenceSentenceLength, $candidateSentenceLength ) = @_;
return ( $referenceSentenceLength < $currentLength ? $referenceSentenceLength : $currentLength );
}
###############################################################################################################################
# function returning the closest reference length (to the candidate sentence length)
# takes as input:
# - currentLength: the current (closest) reference length.
# - candidateSentenceLength : the current reference sentence length
# - candidateSentenceLength : the current candidate sentence length
# when two reference sentences are at the same distance, it will return the shortest reference sentence length
# example of 4 iterations, given:
# - one candidate sentence containing 7 tokens
# - one reference translation containing 11 tokens
# - one reference translation containing 8 tokens
# - one reference translation containing 6 tokens
# - one reference translation containing 7 tokens
# the multiple invokations will return:
# - currentLength is set to 11 (outside of this function)
# - brevity_penalty_closest( 11, 8, 7 ) returns 8, since abs( 8 - 7 ) < abs( 11 - 7 )
# - brevity_penalty_closest( 8, 6, 7 ) returns 6, since abs( 8 - 7 ) == abs( 6 - 7 ) AND 6 < 8
# - brevity_penalty_closest( 7, 6, 7 ) returns 7, since abs( 7 - 7 ) < abs( 6 - 7 )
###############################################################################################################################
sub brevity_penalty_closest
{
my ( $currentLength, $referenceSentenceLength, $candidateSentenceLength ) = @_;
my $result = $currentLength;
if ( abs( $candidateSentenceLength - $referenceSentenceLength ) <= abs( $candidateSentenceLength - $currentLength ) )
{
if ( abs( $candidateSentenceLength - $referenceSentenceLength ) == abs( $candidateSentenceLength - $currentLength ) )
{
if ( $currentLength > $referenceSentenceLength )
{
$result = $referenceSentenceLength;
}
}
else
{
$result = $referenceSentenceLength;
}
}
return $result;
}
#################################
sub score_segment
{
my ($tst_seg, @ref_segs) = @_;
my (@tst_wrds, %tst_ngrams, @match_count, @tst_count, @tst_info);
my (@ref_wrds, $ref_seg, %ref_ngrams, %ref_ngrams_max, @ref_count, @ref_info);
my ($ngram);
my (@nwrds_ref);
my $ref_length;
for (my $j=1; $j<= $max_Ngram; $j++)
{
$match_count[$j] = $tst_count[$j] = $ref_count[$j] = $tst_info[$j] = $ref_info[$j] = 0;
}
# get the ngram counts for the test segment
@tst_wrds = split /\s+/, $tst_seg;
%tst_ngrams = %{Words2Ngrams (@tst_wrds)};
for (my $j=1; $j<=$max_Ngram; $j++)
{
# compute ngram counts
$tst_count[$j] = $j<=@tst_wrds ? (@tst_wrds - $j + 1) : 0;
}
# get the ngram counts for the reference segments
foreach $ref_seg (@ref_segs)
{
@ref_wrds = split /\s+/, $ref_seg;
%ref_ngrams = %{Words2Ngrams (@ref_wrds)};
foreach $ngram (keys %ref_ngrams)
{
# find the maximum # of occurrences
my @wrds = split / /, $ngram;
$ref_info[@wrds] += $ngram_info{$ngram};
$ref_ngrams_max{$ngram} = defined $ref_ngrams_max{$ngram} ? max ($ref_ngrams_max{$ngram}, $ref_ngrams{$ngram}) : $ref_ngrams{$ngram};
}
for (my $j=1; $j<=$max_Ngram; $j++)
{
# update ngram counts
$ref_count[$j] += $j<=@ref_wrds ? (@ref_wrds - $j + 1) : 0;
}
if ( not defined( $ref_length ) )
{
$ref_length = scalar( @ref_wrds );
}
else
{
$ref_length = &{$BLEU_BP}( $ref_length, scalar( @ref_wrds ), scalar( @tst_wrds ) );
}
}
# accumulate scoring stats for tst_seg ngrams that match ref_seg ngrams
foreach $ngram (keys %tst_ngrams)
{
next unless defined $ref_ngrams_max{$ngram};
my @wrds = split / /, $ngram;
$tst_info[@wrds] += $ngram_info{$ngram} * min($tst_ngrams{$ngram},$ref_ngrams_max{$ngram});
$match_count[@wrds] += my $count = min($tst_ngrams{$ngram},$ref_ngrams_max{$ngram});
printf "%.2f info for each of $count %d-grams = '%s'\n", $ngram_info{$ngram}, scalar @wrds, $ngram
if $detail >= 3;
}
return ($ref_length, [@match_count], [@tst_count], [@ref_count], [@tst_info], [@ref_info]);
}
#################################
sub bleu_score_nosmoothing
{
my ($ref_length, $matching_ngrams, $tst_ngrams, $sys, $SCOREmt) = @_;
my $score = 0;
my $iscore = 0;
for ( my $j = 1; $j <= $max_Ngram; ++$j )
{
if ($matching_ngrams->[ $j ] == 0)
{
$SCOREmt->{ $j }{ $sys }{ cum }=0;
}
else
{
my $len_score = min (0, 1-$ref_length/$tst_ngrams->[1]);
# Cumulative N-Gram score
$score += log( $matching_ngrams->[ $j ] / $tst_ngrams->[ $j ] );
$SCOREmt->{ $j }{ $sys }{ cum } = exp( $score / $j + $len_score );
# Individual N-Gram score
$iscore = log( $matching_ngrams->[ $j ] / $tst_ngrams->[ $j ] );
$SCOREmt->{ $j }{ $sys }{ ind } = exp( $iscore );
}
}
return $SCOREmt->{ 4 }{ $sys }{ cum };
}
###############################################################################################################################
# Default method used to compute the BLEU score, using smoothing.
# Note that the method used can be overridden using the '--no-smoothing' command-line argument
# The smoothing is computed by taking 1 / ( 2^k ), instead of 0, for each precision score whose matching n-gram count is null
# k is 1 for the first 'n' value for which the n-gram match count is null
# For example, if the text contains:
# - one 2-gram match
# - and (consequently) two 1-gram matches
# the n-gram count for each individual precision score would be:
# - n=1 => prec_count = 2 (two unigrams)
# - n=2 => prec_count = 1 (one bigram)
# - n=3 => prec_count = 1/2 (no trigram, taking 'smoothed' value of 1 / ( 2^k ), with k=1)
# - n=4 => prec_count = 1/4 (no fourgram, taking 'smoothed' value of 1 / ( 2^k ), with k=2)
###############################################################################################################################
sub bleu_score
{
my ($ref_length, $matching_ngrams, $tst_ngrams, $sys, $SCOREmt) = @_;
my $score = 0;
my $iscore = 0;
my $exp_len_score = 0;
$exp_len_score = exp( min (0, 1 - $ref_length / $tst_ngrams->[ 1 ] ) ) if ( $tst_ngrams->[ 1 ] > 0 );
my $smooth = 1;
for ( my $j = 1; $j <= $max_Ngram; ++$j )
{
if ( $tst_ngrams->[ $j ] == 0 )
{
$iscore = 0;
}
elsif ( $matching_ngrams->[ $j ] == 0 )
{
$smooth *= 2;
$iscore = log( 1 / ( $smooth * $tst_ngrams->[ $j ] ) );
}
else
{
$iscore = log( $matching_ngrams->[ $j ] / $tst_ngrams->[ $j ] );
}
$SCOREmt->{ $j }{ $sys }{ ind } = exp( $iscore );
$score += $iscore;
$SCOREmt->{ $j }{ $sys }{ cum } = exp( $score / $j ) * $exp_len_score;
}
return $SCOREmt->{ 4 }{ $sys }{ cum };
}
#################################
sub nist_score
{
my ($nsys, $matching_ngrams, $tst_ngrams, $ref_ngrams, $tst_info, $ref_info, $sys, $SCOREmt) = @_;
my $score = 0;
my $iscore = 0;
for (my $n=1; $n<=$max_Ngram; $n++)
{
$score += $tst_info->[$n]/max($tst_ngrams->[$n],1);
$SCOREmt->{$n}{$sys}{cum} = $score * nist_length_penalty($tst_ngrams->[1]/($ref_ngrams->[1]/$nsys));
$iscore = $tst_info->[$n]/max($tst_ngrams->[$n],1);
$SCOREmt->{$n}{$sys}{ind} = $iscore * nist_length_penalty($tst_ngrams->[1]/($ref_ngrams->[1]/$nsys));
}
return $SCOREmt->{5}{$sys}{cum};
}
#################################
sub Words2Ngrams
{
#convert a string of words to an Ngram count hash
my %count = ();
for (; @_; shift)
{
my ($j, $ngram, $word);
for ($j=0; $j<$max_Ngram and defined($word=$_[$j]); $j++)
{
$ngram .= defined $ngram ? " $word" : $word;
$count{$ngram}++;
}
}
return {%count};
}
#################################
sub tokenization
{
my ($norm_text) = @_;
# language-independent part:
$norm_text =~ s/<skipped>//g; # strip "skipped" tags
$norm_text =~ s/-\n//g; # strip end-of-line hyphenation and join lines
$norm_text =~ s/\n/ /g; # join lines
$norm_text =~ s/&quot;/"/g; # convert SGML tag for quote to "
$norm_text =~ s/&amp;/&/g; # convert SGML tag for ampersand to &
$norm_text =~ s/&lt;/</g; # convert SGML tag for less-than to >
$norm_text =~ s/&gt;/>/g; # convert SGML tag for greater-than to <
# language-dependent part (assuming Western languages):
$norm_text = " $norm_text ";
$norm_text =~ tr/[A-Z]/[a-z]/ unless $preserve_case;
$norm_text =~ s/([\{-\~\[-\` -\&\(-\+\:-\@\/])/ $1 /g; # tokenize punctuation
$norm_text =~ s/([^0-9])([\.,])/$1 $2 /g; # tokenize period and comma unless preceded by a digit
$norm_text =~ s/([\.,])([^0-9])/ $1 $2/g; # tokenize period and comma unless followed by a digit
$norm_text =~ s/([0-9])(-)/$1 $2 /g; # tokenize dash when preceded by a digit
$norm_text =~ s/\s+/ /g; # one space only between words
$norm_text =~ s/^\s+//; # no leading space
$norm_text =~ s/\s+$//; # no trailing space
return $norm_text;
}
sub tokenization_international
{
my ($norm_text) = @_;
$norm_text =~ s/<skipped>//g; # strip "skipped" tags
#$norm_text =~ s/\p{Hyphen}\p{Zl}//g; # strip end-of-line hyphenation and join lines
$norm_text =~ s/\p{Zl}/ /g; # join lines
# replace entities
$norm_text =~ s/&quot;/\"/g; # quote to "
$norm_text =~ s/&amp;/&/g; # ampersand to &
$norm_text =~ s/&lt;/</g; # less-than to <
$norm_text =~ s/&gt;/>/g; # greater-than to >
$norm_text =~ s/&apos;/\'/g; # apostrophe to '
$norm_text = lc( $norm_text ) unless $preserve_case; # lowercasing if needed
$norm_text =~ s/([^[:ascii:]])/ $1 /g if ( $split_non_ASCII );
# punctuation: tokenize any punctuation unless followed AND preceded by a digit
$norm_text =~ s/(\P{N})(\p{P})/$1 $2 /g;
$norm_text =~ s/(\p{P})(\P{N})/ $1 $2/g;
$norm_text =~ s/(\p{S})/ $1 /g; # tokenize symbols
$norm_text =~ s/\p{Z}+/ /g; # one space only between words
$norm_text =~ s/^\p{Z}+//; # no leading space
$norm_text =~ s/\p{Z}+$//; # no trailing space
return $norm_text;
}
#################################
sub nist_length_penalty
{
my ($ratio) = @_;
return 1 if $ratio >= 1;
return 0 if $ratio <= 0;
my $ratio_x = 1.5;
my $score_x = 0.5;
my $beta = -log($score_x)/log($ratio_x)/log($ratio_x);
return exp (-$beta*log($ratio)*log($ratio));
}
#################################
sub date_time_stamp
{
my ($sec, $min, $hour, $mday, $mon, $year, $wday, $yday, $isdst) = localtime();
my @months = qw(Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec);
my ($date, $time);
$time = sprintf "%2.2d:%2.2d:%2.2d", $hour, $min, $sec;
$date = sprintf "%4.4s %3.3s %s", 1900+$year, $months[$mon], $mday;
return ($date, $time);
}
#################################
sub extract_sgml_tag_and_span
{
my ($name, $data) = @_;
($data =~ m|<$name\s*([^>]*)>(.*?)</$name\s*>(.*)|si) ? ($1, $2, $3) : ();
}
#################################
sub extract_sgml_tag_attribute
{
my ($name, $data) = @_;
($data =~ m|$name\s*=\s*\"([^\"]*)\"|si) ? ($1) : ();
}
#################################
sub max
{
my ($max, $next);
return unless defined ($max=pop);
while (defined ($next=pop))
{
$max = $next if $next > $max;
}
return $max;
}
#################################
sub min
{
my ($min, $next);
return unless defined ($min=pop);
while (defined ($next=pop))
{
$min = $next if $next < $min;
}
return $min;
}
#################################
sub printout_report
{
if ( $METHOD eq "BOTH" )
{
foreach my $sys (sort @tst_sys)
{
printf "NIST score = %2.4f BLEU score = %.4f for system \"$sys\"\n",$NISTmt{5}{$sys}{cum},$BLEUmt{4}{$sys}{cum};
}
}
elsif ($METHOD eq "NIST" )
{
foreach my $sys (sort @tst_sys)
{
printf "NIST score = %2.4f for system \"$sys\"\n",$NISTmt{5}{$sys}{cum};
}
}
elsif ($METHOD eq "BLEU" )
{
foreach my $sys (sort @tst_sys)
{
printf "\nBLEU score = %.4f for system \"$sys\"\n",$BLEUmt{4}{$sys}{cum};
}
}
printf "\n# ------------------------------------------------------------------------\n\n";
printf "Individual N-gram scoring\n";
printf " 1-gram 2-gram 3-gram 4-gram 5-gram 6-gram 7-gram 8-gram 9-gram\n";
printf " ------ ------ ------ ------ ------ ------ ------ ------ ------\n";
if ( ( $METHOD eq "BOTH" ) || ($METHOD eq "NIST") )
{
foreach my $sys (sort @tst_sys)
{
printf " NIST:";
for (my $i=1; $i<=$max_Ngram; $i++)
{
printf " %2.4f ",$NISTmt{$i}{$sys}{ind}
}
printf " \"$sys\"\n";
}
printf "\n";
}
if ( ( $METHOD eq "BOTH" ) || ($METHOD eq "BLEU") )
{
foreach my $sys (sort @tst_sys)
{
printf " BLEU:";
for (my $i=1; $i<=$max_Ngram; $i++)
{
printf " %2.4f ",$BLEUmt{$i}{$sys}{ind}
}
printf " \"$sys\"\n";
}
}
printf "\n# ------------------------------------------------------------------------\n";
printf "\nCumulative N-gram scoring\n";
printf " 1-gram 2-gram 3-gram 4-gram 5-gram 6-gram 7-gram 8-gram 9-gram\n";
printf " ------ ------ ------ ------ ------ ------ ------ ------ ------\n";
if (( $METHOD eq "BOTH" ) || ($METHOD eq "NIST"))
{
foreach my $sys (sort @tst_sys)
{
printf " NIST:";
for (my $i=1; $i<=$max_Ngram; $i++)
{
printf " %2.4f ",$NISTmt{$i}{$sys}{cum}
}
printf " \"$sys\"\n";
}
}
printf "\n";
if ( ( $METHOD eq "BOTH" ) || ($METHOD eq "BLEU") )
{
foreach my $sys (sort @tst_sys)
{
printf " BLEU:";
for (my $i=1; $i<=$max_Ngram; $i++)
{
printf " %2.4f ",$BLEUmt{$i}{$sys}{cum}
}
printf " \"$sys\"\n";
}
}
}
###############################################################################################################################
# Create three files, by using:
# - $prefix : the prefix used for the output file names
# - %overall : a hash containing seg/doc/sys-level scores:
# - $overall{ $SYSTEM_ID }{ 'score' } => system-level score
# - $overall{ $SYSTEM_ID }{ 'documents' }{ $DOCUMENT_ID }{ 'score' } => document-level score
# - $overall{ $SYSTEM_ID }{ 'documents' }{ $DOCUMENT_ID }{ 'segments' }{ $SEGMENT_ID } => segment-level score
###############################################################################################################################
sub outputMetricsMATR
{
my ( $prefix, %overall ) = @_;
my $fileNameSys = $prefix . '-sys.scr';
my $fileNameDoc = $prefix . '-doc.scr';
my $fileNameSeg = $prefix . '-seg.scr';
open FILEOUT_SYS, '>', $fileNameSys or die "Could not open file: ${fileNameSys}";
open FILEOUT_DOC, '>', $fileNameDoc or die "Could not open file: ${fileNameDoc}";
open FILEOUT_SEG, '>', $fileNameSeg or die "Could not open file: ${fileNameSeg}";
foreach my $sys ( sort( keys( %overall ) ) )
{
my $scoreSys = $overall{ $sys }{ 'score' };
print FILEOUT_SYS "${tst_id}\t${sys}\t${scoreSys}\n";
foreach my $doc ( sort( keys( %{$overall{ $sys }{ 'documents' }} ) ) )
{
my $scoreDoc = $overall{ $sys }{ 'documents' }{ $doc }{ 'score' };
print FILEOUT_DOC "${tst_id}\t${sys}\t${doc}\t${scoreDoc}\n";
foreach my $seg ( nsort keys( %{$overall{ $sys }{ 'documents' }{ $doc }{ 'segments' }} ) )
{
my $scoreSeg = $overall{ $sys }{ 'documents' }{ $doc }{ 'segments' }{ $seg }{ 'score' };
print FILEOUT_SEG "${tst_id}\t${sys}\t${doc}\t${seg}\t${scoreSeg}\n";
}
}
}
close FILEOUT_SEG;
close FILEOUT_DOC;
close FILEOUT_SYS;
}