Gitiles
Code Review Sign In
korap.ids-mannheim.de / KorAP / KorAP-XML-TEI / 044bd92776786b32925662c4dedf7d4ef7dd9253 / . / script / tei2korapxml
blob: 590daad5edf69804b145d03b1acfd4df5aab62c1 [file] [log] [blame]
#!/usr/bin/env perl
use strict;
use warnings;
use Pod::Usage;
use Getopt::Long qw(GetOptions :config no_auto_abbrev);
use File::Basename qw(dirname);
use open qw(:std :utf8); # assume utf-8 encoding
use Encode qw(encode decode);
use XML::CompactTree::XS;
use XML::LibXML::Reader;
use FindBin;
BEGIN {
unshift @INC, "$FindBin::Bin/../lib";
};
use KorAP::XML::TEI qw!remove_xml_comments escape_xml!;
use KorAP::XML::TEI::Tokenizer::External;
use KorAP::XML::TEI::Tokenizer::Conservative;
use KorAP::XML::TEI::Tokenizer::Aggressive;
use KorAP::XML::TEI::Tokenizer::Collector;
use KorAP::XML::TEI::Zipper;
use KorAP::XML::TEI::Header;
our $VERSION = '0.01';
our $VERSION_MSG = "\ntei2korapxml - v$VERSION\n";
# Parse options from the command line
GetOptions(
"root|r=s" => \(my $_root_dir = '.'), # name of root directory inside zip file
"input|i=s" => \(my $input_fname = ''), # input file (yet only TEI I5 Format accepted)
'tokenizer-call|tc=s' => \(my $tokenizer_call), # Temporary argument for testing purposes
'use-intern-tokenization|ti' => \(my $tokenizer_intern), # use intern tokenization (default = no)
'help|h' => sub {
pod2usage(
-verbose => 99,
-sections => 'NAME|DESCRIPTION|SYNOPSIS|ARGUMENTS|OPTIONS',
-msg => $VERSION_MSG,
-output => '-'
)
},
'version|v' => sub {
pod2usage(
-verbose => 0,
-msg => $VERSION_MSG,
-output => '-'
)
}
);
#
# ~~~ parameter (mandatory) ~~~
#
my $_TEXT_BODY = "text"; # tag (without attributes), which contains the primary text
# optional
my $_CORP_HEADER_BEG = "idsHeader type=\"corpus\""; # just keep the correct order of the attributes and evtl. add an '.*' between them
# optional
my $_DOC_HEADER_BEG = "idsHeader type=\"document\""; # analog
# mandatory
my $_TEXT_HEADER_BEG = "idsHeader type=\"text\""; # analog
#
# ~~~ constants ~~~
#
## extern tokenization
my $_GEN_TOK_EXT = $tokenizer_call ? 1 : 0;
# TODO:
# Read tokenizer call from configuration file.
# was 'java -cp '. join(":", ".", glob(&dirname(__FILE__)."/../target/*.jar")). " de.ids_mannheim.korap.tokenizer.KorAPTokenizerImpl";
my $ext_tok;
if ($tokenizer_call) {
$ext_tok = KorAP::XML::TEI::Tokenizer::External->new($tokenizer_call);
};
my $_tok_file_ext = "tokens.xml";
##
## intern tokenization
my $_GEN_TOK_INT = $tokenizer_intern; # simple tokenization (recommended for testing)
my $_tok_file_con = "tokens_conservative.xml";
my $_tok_file_agg = "tokens_aggressive.xml";
my $aggr_tok = KorAP::XML::TEI::Tokenizer::Aggressive->new;
my $cons_tok = KorAP::XML::TEI::Tokenizer::Conservative->new;
##
my $_tok_dir = "base"; # name of directory for storing tokenization files
my $_DEBUG = 0; # set to 1 for minimal more debug output (no need to be parametrized)
my $_XCT_LN = 0; # only for debugging: include line numbers in elements of $tree_data
# (see also manpage of XML::CompactTree::XS)
my $_header_file = "header.xml"; # name of files containing the text, document and corpus header
my $_data_file = "data.xml"; # name of file containing the primary text data (tokens)
my $_structure_dir = "struct"; # name of directory containing the $_structure_file
my $_structure_file = "structure.xml"; # name of file containing all tags (except ${_TOKEN_TAG}'s) related information
# (= their names and byte offsets in $_data)
## TODO: optional (different annotation tools can produce more zip-files for feeding into KorAP-XML-Krill)
my $_TOKENS_PROC = 1; # on/off: processing of ${_TOKEN_TAG}'s (default: 1)
my $_tokens_dir = "tokens"; # name of directory containing the $_tokens_file
my $_tokens_file = "morpho.xml"; # name of file containing all ${_TOKEN_TAG}'s related information (=their byte offsets in $_data)
# - evtl. with additional inline annotations
my $_TOKENS_TAG = "w"; # name of tag containing all information stored in $_tokens_file
## TODO: optional
# handling inline annotations (inside $_TOKENS_TAG)
my $_INLINE_ANNOT = $ENV{KORAPXMLTEI_INLINE}?1:0; # on/off: set to 1 if inline annotations are present and should be processed (default: 0)
# TODO:
# These parameters are now defunct and moved to Token.pm
my $_INLINE_LEM_RD = "lemma"; # from which attribute to read LEMMA information
my $_INLINE_ATT_RD = "ana"; # from which attribute to read POS information (and evtl. additional MSD - Morphosyntactic Descriptions)
# TODO: The format for the POS and MSD information has to suffice the regular expression ([^ ]+)( (.+))?
# - which means, that the POS information can be followed by an optional blank with additional
# MSD information; unlike the MSD part, the POS part may not contain any blanks.
my $_INLINE_POS_WR = "pos"; # name (inside $_tokens_file) referring to POS information
my $_INLINE_MSD_WR = "msd"; # name (inside $_tokens_file) referring to MSD information
my $_INLINE_LEM_WR = "lemma"; # name (inside $_tokens_file) referring to LEMMA information
##
#
# ~~~ variables ~~~
#
# Initialize Token-Collector
my $tokens = KorAP::XML::TEI::Tokenizer::Collector->new;
# Initialize zipper
my $zipper = KorAP::XML::TEI::Zipper->new;
my $input_fh; # input file handle (default: stdin)
my $buf_in; # text body data extracted from input document ($input_fh), further processed by XML::LibXML::Reader
my $data; # contains the primary text (created by func. 'retr_info' from $buf_in), which is written to '$data_file'
my $dir; # text directory (below $_root_dir)
my ( $text_id, $text_id_esc ); # '$text_id_esc' = escaped version of $text_id (see %ent)
my ( $data_fl );
my ( $data_prfx1, $data_prfx2, $data_sfx ); # $data_* are written to $_data_file
my @structures; # list of arrays, where each array represents a TEI I5 tag (except $_TOKENS_TAG) from the input document
# - the input of this array is written in func. 'write_structures' into the file '$_structure_file'
my ( $ref, $idx, $att_idx ); # needed in func. 'write_structures'
my ( $reader, # instance of 'XML::LibXML::Reader->new' (on input '$buf_in')
$tree_data ); # instance of 'XML::CompactTree::XS::readSubtreeToPerl' (on input '$reader')
# these are only used inside recursive function 'retr_info'
my ( $_IDX, # value is set dependent on $_XCT_LN - for extracting array of child elements from element in $tree_data
$e, # element from $tree_data
$n, # tag name of actual processed element $e
$rl, # recursion level
$dl, # actual length of string $data
@oti, # oti='open tags indizes' - a stack of indizes into @structures, where the top index in @oti
# represents the actual processed element from @structures
$inside_tokens_tag, # flag is set, when inside $_TOKENS_TAG
## variables for handling ~ whitespace related issue ~ (it is sometimes necessary, to correct the from-values for some tags)
$add_one, # ...
$fval, $fval2, # ...
%ws); # hash for indices of whitespace-nodes (needed to recorrect from-values)
# idea: when closing element, check if it's from-index minus 1 refers to a whitespace-node
# (means: 'from-index - 1' is a key in %ws).
# if this is _not_ the case, then the from-value is one to high => correct it by substracting 1
my $output; # temporary variable needed in 'write_*'-functions for writing output to zip-stream $zip)
my ( $i, $c ); # index variables used in loops
#
# ~~~ main ~~~
#
# ~ initializations ~
($_XCT_LN)?($_IDX=5):($_IDX=4);
$data_prfx1 = $data_prfx2 = $data_sfx = "";
$inside_tokens_tag = -1;
$fval = $fval2 = 0;
$_root_dir .= '/'; # base dir must always end with a slash
$_root_dir =~ s/^\.?\///; # remove leading / (only relative paths allowed in IO::Compress::Zip) and redundant ./
$_CORP_HEADER_BEG =~ s#^([^\s]+)(.*)$#$1\[\^>\]*$2#;
$_DOC_HEADER_BEG =~ s#^([^\s]+)(.*)$#$1\[\^>\]*$2#;
$_TEXT_HEADER_BEG =~ s#^([^\s]+)(.*)$#$1\[\^>\]*$2#;
$data_prfx1 = "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n";
$data_prfx1 .= "<?xml-model href=\"text.rng\" type=\"application/xml\" schematypens=\"http://relaxng.org/ns/structure/1.0\"?>\n\n";
$data_prfx1 .= "<raw_text docid=\"";
$data_prfx2 .= "\" xmlns=\"http://ids-mannheim.de/ns/KorAP\">\n";
## TODO: can 'metadata.xml' change or is it constant?
$data_prfx2 .= " <metadata file=\"metadata.xml\" />\n";
##
$data_prfx2 .= " <text>";
$data_sfx = "</text>\n</raw_text>";
# ~ read input and write output (text by text) ~
main();
#
# ~~~ subs ~~~
#
sub main {
my ( $pfx, $sfx );
my $tl = 0; # text line (needed for whitespace handling)
$input_fh = *STDIN; # input file handle (default: stdin)
$data_fl = 0;
$buf_in = $data = $dir = "";
if ( $input_fname ne '' ){
open ( $input_fh, "<", "$input_fname") || die "File \'$input_fname\' could not be opened.\n";
}
# prevents segfaulting of 'XML::LibXML::Reader' inside 'main()' - see notes on 'PerlIO layers' in 'man XML::LibXML')
# removing 'use open qw(:std :utf8)' would fix this problem too, but using binmode on input is more granular
# see in perluniintro: You can switch encodings on an already opened stream by using "binmode()
# see in perlfunc: If LAYER is omitted or specified as ":raw" the filehandle is made suitable for passing binary data.
binmode $input_fh;
# ~ loop (reading input document) ~
while ( <$input_fh> ){
# TODO: yet not tested fo big amounts of data
# must-have, otherwise comments in input could be fatal (e.g.: ...<!--\n<idsHeader...\n-->...)
remove_xml_comments( $input_fh, $_ ); # remove HTML comments (<!--...-->)
if ( $data_fl && m#^(.*)</${_TEXT_BODY}>(.*)$# ){
# ~ end of text body ~
# write data.xml, structure.xml and evtl. morpho.xml and/or tokenization files (s.a.: $_tok_file_ext, $_tok_file_con, $_tok_file_agg)
$pfx = $1; $sfx = $2;
die "ERROR ($0): main(): input line number $.: line with closing text-body tag '${_TEXT_BODY}'"
." contains additional information ... => Aborting\n\tline=$_"
if $pfx !~ /^\s*$/ || $sfx !~ /^\s*$/;
if ( $dir ne "" ){
$reader = XML::LibXML::Reader->new( string => "<text>$buf_in</text>", huge => 1 );
# ~ whitespace handling ~
#
# Every whitespace inside the processed text is 'significant' and recognized as a node of type 'XML_READER_TYPE_SIGNIFICANT_WHITESPACE'
# (see function 'retr_info()').
#
# Definition of significant and insignificant whitespace
# (source: https://www.oracle.com/technical-resources/articles/wang-whitespace.html):
#
# Significant whitespace is part of the document content and should be preserved.
# Insignificant whitespace is used when editing XML documents for readability.
# These whitespaces are typically not intended for inclusion in the delivery of the document.
#
if ( $_XCT_LN ){ # _XCT_LINE_NUMBERS is only for debugging
$tree_data = XML::CompactTree::XS::readSubtreeToPerl( $reader, XCT_DOCUMENT_ROOT | XCT_IGNORE_COMMENTS | XCT_ATTRIBUTE_ARRAY | XCT_LINE_NUMBERS );
} else {
$tree_data = XML::CompactTree::XS::readSubtreeToPerl( $reader, XCT_DOCUMENT_ROOT | XCT_IGNORE_COMMENTS | XCT_ATTRIBUTE_ARRAY );
}
@structures = (); @oti = ();
if ( $_TOKENS_PROC ){
$tokens->reset;
}
$dl = $rl = 0;
# ~ whitespace related issue ~
$add_one = 0;
%ws = ();
# ~ recursion ~
retr_info( \$tree_data->[2] ); # parse input data
$rl--;
# ~ write data.xml ~
# TODO: should not be necessary, because whitespace at the end of every input line is removed: see 'whitespace handling' inside text body
# (...elsif ( $data_fl )....)
$data =~ tr/\n\r/ /; # note: 2 blanks - otherwise offset data would become corrupt
#
# ~ tokenization ~
if ( $_GEN_TOK_EXT ){
# Tokenize and output
$ext_tok->tokenize($data)->to_zip(
$zipper->new_stream("$_root_dir$dir/$_tok_dir/$_tok_file_ext"),
$text_id_esc
);
};
if ( $_GEN_TOK_INT ){
# Tokenize and output
$cons_tok->tokenize($data)->to_zip(
$zipper->new_stream("$_root_dir$dir/$_tok_dir/$_tok_file_con"),
$text_id_esc
);
$aggr_tok->tokenize($data)->to_zip(
$zipper->new_stream("$_root_dir$dir/$_tok_dir/$_tok_file_agg"),
$text_id_esc
);
$aggr_tok->reset;
$cons_tok->reset;
};
# Encode and escape data
$data = escape_xml(encode( "UTF-8", $data ));
# note: the index still refers to the 'single character'-versions,
# which are counted as 1 (search for '&amp;' in data.xml and see
# corresponding indices in $_tokens_file)
print STDERR "DEBUG ($0): main(): Writing (utf8-formatted) xml file $_root_dir$dir/$_data_file\n" if $_DEBUG;
$zipper->new_stream("$_root_dir$dir/$_data_file")
->print("$data_prfx1$text_id_esc$data_prfx2$data$data_sfx");
# ~ write structures ~
write_structures() if @structures;
# ~ write tokens ~
if ($_TOKENS_PROC && !$tokens->empty) {
$tokens->to_zip(
$zipper->new_stream("$_root_dir$dir/$_tokens_dir/${_tokens_file}"),
$text_id_esc,
$_INLINE_ANNOT
);
};
#print STDERR "$0: write_tokenization(): DONE\n";
$data_fl = 0; $buf_in = $data = $dir = ""; # reinit.
} else { # $dir eq ""
print STDERR "WARNING ($0): main(): maybe empty textSigle => skipping this text ...\n";
# print STDERR "WARNING ($0): main(): text header=$header_txt\n";
print STDERR "WARNING ($0): main(): data=$data\n";
}
} elsif ( $data_fl ){
# ~ inside text body ~
#print STDERR "inside text body (\$data_fl set)\n";
# ~ whitespace handling ~
# The idea for the below code fragment was to fix (recreate) missing whitespace in a poorly created corpus, in which linebreaks where inserted
# into the text with the addition that maybe (or not) whitespace before those linebreaks was unintenionally stripped.
#
# It soon turned out, that it was best to suggest considering just avoiding linebreaks and putting all primary text tokens into one line (see
# example further down and notes on 'Input restrictions' in the manpage).
#
# Somehow an old first very poor approach remained, which is not stringent, but also doesn't affect one-line text.
#
# TODO: Maybe it's best, to keep the stripping of whitespace and to just remove the if-clause and to insert a blank by default (with possibly
# an option on how newlines in primary text should be handled (stripped or replaced by a whitespace)).
#
# Examples (how primary text with linebreaks would be converted by below code):
#
# '...<w>end</w>\n<w>.</w>...' -> '...<w>end</w> <w>.</w>...'
# '...<w>,</w>\n<w>this</w>\n<w>is</w>\n<w>it</w>\n<w>!</w>...' -> '<w>,<w> <w>this</w> <w>is</w> <w>it</w> <w>!</w>'.
s/^\s+//; s/\s+$//; # remove consecutive whitespace at beginning and end (mostly one newline)
### NOTE: this is only relevant, if a text consists of more than one line
### TODO: find a better solution, or create a warning, if a text has more than one line ($tl > 1)
### do testing with 2 different corpora (one with only one-line texts, the other with several lines per text)
if ( m/<[^>]+>[^<]/ ){ # line contains at least one tag with at least one character contents
# NOTE: not stringent ('...' stands for text):
#
# beg1............................end1 => no blank before 'beg1'
# beg2....<pb/>...................end2 => no blank before 'beg2'
# beg3....<info attr1="val1"/>....end3 => no blank before 'beg3'
# beg4....<test>ok</test>.........end4 => blank before 'beg4'
#
# => beg1....end1beg2...<pb/>...end2beg3....<info attr1="val1"/>....end3 beg4...<test>ok</test>....end4
# ^
# |_blank between 'end3' and 'beg4'
$tl++; # counter for text lines
s/^(.)/ $1/ if $tl > 1; # insert blank before 1st character (for 2nd line and consecutive lines)
}
###
# add line to buffer
$buf_in .= $_;
} elsif ( m#^(.*)<${_TEXT_BODY}(?: [^>]*)?>(.*)$# ){
# ~ start of text body ~
#print STDERR "inside text body\n";
$pfx = $1; $sfx = $2;
$data_fl = 1;
die "ERROR ($0): main(): input line number $.: line with opening text-body tag '${_TEXT_BODY}'"
." contains additional information ... => Aborting\n\tline=$_"
if $pfx !~ /^\s*$/ || $sfx !~ /^\s*$/;
} elsif ( m#^(.*)(<(?:${_TEXT_HEADER_BEG}|${_DOC_HEADER_BEG}|${_CORP_HEADER_BEG}).*)$# ){
# ~ start of header ~
$pfx = $1;
my $content = "$2\n";
die "ERROR ($0): main(): input line number $.: line with opening header tag"
." is not in expected format ... => Aborting\n\tline=$_"
if $pfx !~ /^\s*$/;
# Parse header
my $header = KorAP::XML::TEI::Header->new($content)->parse($input_fh);
# Header was parseable
if ($header) {
# Write header to zip
my $file = $_root_dir . $header->dir . '/' . $_header_file;
print STDERR "DEBUG ($0): Writing file $file\n" if $_DEBUG;
$header->to_zip($zipper->new_stream($file));
# Header is for text level
if ($header->type eq 'text') {
# Remember dir and sigles
$dir = $header->dir;
$text_id = $header->id;
$text_id_esc = $header->id_esc;
# log output for seeing progression
print STDERR "$0: main(): text_id=".decode("UTF-8", $text_id )."\n";
$tl = 0; # reset (needed for ~ whitespace handling ~)
};
}
}
} #end: while
$zipper->close;
$ext_tok->close if $_GEN_TOK_EXT;
} # end: sub main
sub retr_info { # called from main()
# Notes on how 'XML::CompactTree::XS' works
#
# Example: <node a="v"><node1>some <n/> text</node1><node2>more-text</node2></node>
#
# Print out name of 'node2' for the above example:
#
# echo '<node a="v"><node1>some <n/> text</node1><node2>more-text</node2></node>' | perl -e 'use XML::CompactTree::XS; use XML::LibXML::Reader; $reader = XML::LibXML::Reader->new(IO => STDIN); $data = XML::CompactTree::XS::readSubtreeToPerl( $reader, XCT_DOCUMENT_ROOT | XCT_IGNORE_COMMENTS | XCT_LINE_NUMBERS ); print "\x27".$data->[2]->[0]->[5]->[1]->[1]."\x27\n"'
#
# Exploring the structure of $data ( = reference to below array ):
#
# [ 0: XML_READER_TYPE_DOCUMENT,
# 1: ?
# 2: [ 0: [ 0: XML_READER_TYPE_ELEMENT <- start recursion with array '$data->[2]' (see main(): retr_info( \$tree_data->[2] ))
# 1: 'node'
# 2: ?
# 3: HASH (attributes)
# 4: 1 (line number)
# 5: [ 0: [ 0: XML_READER_TYPE_ELEMENT
# 1: 'node1'
# 2: ?
# 3: undefined (no attributes)
# 4: 1 (line number)
# 5: [ 0: [ 0: XML_READER_TYPE_TEXT
# 1: 'some '
# ]
# 1: [ 0: XML_READER_TYPE_ELEMENT
# 1: 'n'
# 2: ?
# 3: undefined (no attributes)
# 4: 1 (line number)
# 5: undefined (no child-nodes)
# ]
# 2: [ 0: XML_READER_TYPE_TEXT
# 1: ' text'
# ]
# ]
# ]
# 1: [ 0: XML_READER_TYPE_ELEMENT
# 1: 'node2'
# 2: ?
# 3: undefined (not attributes)
# 4: 1 (line number)
# 5: [ 0: [ 0: XML_READER_TYPE_TEXT
# 1: 'more-text'
# ]
# ]
# ]
# ]
# ]
# ]
# ]
#
# $data->[0] = 9 (=> type == XML_READER_TYPE_DOCUMENT)
#
# ref($data->[2]) == ARRAY (with 1 element for 'node')
# ref($data->[2]->[0]) == ARRAY (with 6 elements)
#
# $data->[2]->[0]->[0] == 1 (=> type == XML_READER_TYPE_ELEMENT)
# $data->[2]->[0]->[1] == 'node'
# ref($data->[2]->[0]->[3]) == HASH (=> ${$data->[2]->[0]->[3]}{a} == 'v')
# $data->[2]->[0]->[4] == 1 (line number)
# ref($data->[2]->[0]->[5]) == ARRAY (with 2 elements for 'node1' and 'node2')
# # child-nodes of actual node (see $_IDX)
#
# ref($data->[2]->[0]->[5]->[0]) == ARRAY (with 6 elements)
# $data->[2]->[0]->[5]->[0]->[0] == 1 (=> type == XML_READER_TYPE_ELEMENT)
# $data->[2]->[0]->[5]->[0]->[1] == 'node1'
# $data->[2]->[0]->[5]->[0]->[3] == undefined (=> no attribute)
# $data->[2]->[0]->[5]->[0]->[4] == 1 (line number)
# ref($data->[2]->[0]->[5]->[0]->[5]) == ARRAY (with 3 elements for 'some ', '<n/>' and ' text')
#
# ref($data->[2]->[0]->[5]->[0]->[5]->[0]) == ARRAY (with 2 elements)
# $data->[2]->[0]->[5]->[0]->[5]->[0]->[0] == 3 (=> type == XML_READER_TYPE_TEXT)
# $data->[2]->[0]->[5]->[0]->[5]->[0]->[1] == 'some '
#
# ref($data->[2]->[0]->[5]->[0]->[5]->[1]) == ARRAY (with 5 elements)
# $data->[2]->[0]->[5]->[0]->[5]->[1]->[0] == 1 (=> type == XML_READER_TYPE_ELEMENT)
# $data->[2]->[0]->[5]->[0]->[5]->[1]->[1] == 'n'
# $data->[2]->[0]->[5]->[0]->[5]->[1]->[3] == undefined (=> no attribute)
# $data->[2]->[0]->[5]->[0]->[5]->[1]->[4] == 1 (line number)
# $data->[2]->[0]->[5]->[0]->[5]->[1]->[5] == undefined (=> no child-nodes)
#
# ref($data->[2]->[0]->[5]->[0]->[5]->[2]) == ARRAY (with 2 elements)
# $data->[2]->[0]->[5]->[0]->[5]->[2]->[0] == 3 (=> type == XML_READER_TYPE_TEXT)
# $data->[2]->[0]->[5]->[0]->[5]->[2]->[1] == ' text'
#
#
# retr_info() starts with the array reference ${$_[0]} (= \$tree_data->[2]), which corresponds to ${\$data->[2]} in the above example.
# Hence, the expression @{${$_[0]}} corresponds to @{${\$data->[2]}}, $e to ${${\$data->[2]}}[0] (= $data->[2]->[0]) and $e->[0] to
# ${${\$data->[2]}}[0]->[0] (= $data->[2]->[0]->[0]).
$rl++; # recursion level (1 = topmost level inside retr_info() = should always be level of tag $_TEXT_BODY)
foreach $e ( @{${$_[0]}} ){ # iteration through all array elements ($_[0] is a reference to an array reference)
if ( $e->[0] == XML_READER_TYPE_ELEMENT ){ # element-node (see 'NODE TYPES' in manpage of XML::LibXML::Reader)
#~~~~
# from here: tag-node (opening)
#~~~~
# insert new array (for new tag) into @structures with tag-name and tag-attributes (if present)
# update @oti (open tags indizes) with @structures highest index (= $#structures); e.g.: @a=(1,2,3) => $#a = 2
# ~ tag name ~
$n = $e->[1];
# ~ handle structures ~
my @array;
push @array, $n;
push @structures, \@array;
push @oti, $#structures; # add highest index of @structures to @oti
# ~ handle tokens ~
# Wether to push entry also into tokens
$inside_tokens_tag = $rl if $_TOKENS_PROC && $n eq $_TOKENS_TAG;
my $current_token;
# Add element to token list
if ( $_TOKENS_PROC && $inside_tokens_tag == $rl ){
$current_token = $tokens->add_token($n); # TODO: adding $n is of no use (redundant)
}
# ~ handle attributes ~
if ( defined $e->[3] ){ # only if attributes exist
for ( $c = 0; $c < @{$e->[3]}; $c += 2 ){ # with 'XCT_ATTRIBUTE_ARRAY', $node->[3] is an array reference of the form
# [ name1, value1, name2, value2, ....] of attribute names and corresponding values.
# note: arrays are faster (see: http://makepp.sourceforge.net/2.0/perl_performance.html)
# '$c' references the 'key' and '$c+1' the 'value'
push @{$structures[$#structures]}, ${$e->[3]}[$c], ${$e->[3]}[$c+1];
if ( $_TOKENS_PROC && $inside_tokens_tag == $rl ){
# Add attributes to current token
$current_token->add_attribute(
@{$e->[3]}[$c, $c + 1]
);
}
}
}
# ~ index 'from' ~
# this is, where a normal tag or tokens-tag ($_TOKENS_TAG) starts
push @{$structures[$#structures]}, ( $dl + $add_one ); # see below (text- and whitespace-nodes) for explanation on '$add_one'
if ( $_TOKENS_PROC && $inside_tokens_tag == $rl ){
# Set from value to tokens
$current_token->set_from($dl + $add_one);
};
#~~~~
# until here: tag-node (opening)
#~~~~
# ~~ RECURSION ~~
if ( defined $e->[$_IDX] ){ # do no recursion, if $e->[$_IDX] is not defined (because we have no array of child-nodes, e.g.: <back/>)
retr_info( \$e->[$_IDX] ); # recursion with array of child-nodes
$rl--; # return from recursion
}
#~~~~~
# from here: tag-node (closing)
#~~~~~
# ~ handle structures ~
{
my $ix = pop @oti; # index of just closed tag
my $aix = $#{$structures[$ix]}; # determine highest index from 'array referring to last closed tag' ...
$fval = ${$structures[$ix]}[ $aix ]; # ... and get it's from-value
if ( $fval > 0 && not exists $ws{ $fval - 1 } ){ # ~ whitespace related issue ~
# ~ previous node was a text-node ~
${$structures[$ix]}[ $aix ] = $fval - 1; # recorrect from-value (see below: Notes on ~ whitespace related issue ~)
}
# in case this fails, check input
die "ERROR ($0, retr_info()): text_id='$text_id', processing of \@structures: from-value ($fval) is 2 or more greater"
." than to-value ($dl) => please check. aborting ...\n"
if ( $fval - 1 ) > $dl;
# TODO: find example for which this case applies
# maybe this is not necessary anymore, because the above recorrection of the from-value suffices
# TODO: check, if it's better to remove this line and change above check to 'if ( $fval - 1) >= $dl;
# do testing with bigger corpus excerpt (wikipedia?)
${$structures[$ix]}[ $aix ] = $dl if $fval == $dl + 1; # correct from-value (same as ... if $fval-1 == $dl)
push @{$structures[$ix]}, $dl, $rl; # to-value and recursion-level
# note: use $dl, because the offsets are _between_ the characters (e.g.: word = 'Hello' => from = 0 (before 'H'), to = 5 (after 'o'))
}
# ~ handle tokens ~
if ( $_TOKENS_PROC && $inside_tokens_tag == $rl ){
# Check last added token
my $last_token = $tokens->last_token;
# Get from-value from last added token
my $fval2 = $last_token->from;
if( $fval2 > 0 && not exists $ws{ $fval2 - 1 } ){ # ~ whitespace related issue ~
# ~ previous node was a text-node ~
# recorrect from-value
# (see below: Notes on ~ whitespace related issue ~)
$last_token->set_from($fval2 - 1);
}
# in case this fails, check input
die "ERROR ($0, retr_info()): text_id='$text_id', processing of tokens: from-value ($fval2) is 2 or more greater"
." than to-value ($dl) => please check. aborting ...\n"
if ( $fval2 - 1 ) > $dl;
# TODO:
# find example for which this case applies
# maybe this is not necessary anymore, because the above recorrection of the from-value suffices
#
# TODO:
# check, if it's better to remove this line and change above check to 'if ( $fval2 - 1) >= $dl;
# do testing with bigger corpus excerpt (wikipedia?)
# Correct from-value (same as ... if $fval-1 == $dl)
$last_token->set_from($dl) if $fval2 == $dl + 1;
$last_token->set_to($dl); # Here from == to?
$last_token->set_level($rl);
$inside_tokens_tag = -1; # reset
}
# ~ whitespace related issue ~
# clean up
delete $ws{ $fval - 1 } if $fval > 0 && exists $ws{ $fval - 1 };
delete $ws{ $fval2 - 1 } if $_TOKENS_PROC && $fval2 > 0 && exists $ws{ $fval2 - 1 };
#~~~~
# until here: tag-node (closing)
#~~~~
#~~~~~
# from here: text- and whitespace-nodes
#~~~~~
# The 3rd form of nodes, besides text- (XML_READER_TYPE_TEXT) and tag-nodes (XML_READER_TYPE_ELEMENT) are nodes of the type
# 'XML_READER_TYPE_SIGNIFICANT_WHITESPACE'.
#
# When modifiying the previous example (see: Notes on how 'XML::CompactTree::XS' works) by inserting an additional blank between
# '</node1>' and '<node2>', the output for '$data->[2]->[0]->[5]->[1]->[1]' is a blank (' ') and it's type is '14'
# (XML_READER_TYPE_SIGNIFICANT_WHITESPACE, see 'man XML::LibXML::Reader'):
#
# echo '<node a="v"><node1>some <n/> text</node1> <node2>more-text</node2></node>' | perl -e 'use XML::CompactTree::XS; use XML::LibXML::Reader; $reader = XML::LibXML::Reader->new(IO => STDIN); $data = XML::CompactTree::XS::readSubtreeToPerl( $reader, XCT_DOCUMENT_ROOT | XCT_IGNORE_COMMENTS | XCT_LINE_NUMBERS ); print "node=\x27".$data->[2]->[0]->[5]->[1]->[1]."\x27, type=".$data->[2]->[0]->[5]->[1]->[0]."\n"'
} elsif ( $e->[0] == XML_READER_TYPE_TEXT || $e->[0] == XML_READER_TYPE_SIGNIFICANT_WHITESPACE ){
# Notes on ~ whitespace related issue ~ (referred to the code fragment below)
#
# Example: '... <head type="main"><s>Campagne in Frankreich</s></head><head type="sub"> <s>1792</s> ...'
#
# Two text-nodes should normally be separated by a blank. In the above example, that would be the 2 text-nodes
# 'Campagne in Frankreich' and '1792', which are separated by the whitespace-node ' ' (see [2]).
#
# The text-node 'Campagne in Frankreich' leads to the setting of '$add_one' to 1, so that when opening the 2nd 'head'-tag,
# it's from-index gets set to the correct start-index of '1792' (and not to the start-index of the whitespace-node ' ').
#
# The assumption here is, that in most cases there _is_ a whitespace node between 2 text-nodes. The below code fragment
# enables a way, to check, if this really _was_ the case for the last 2 'non-tag'-nodes, when closing a tag:
#
# When a whitespace-node is read, its from-index is stored as a hash-key (in %ws), to state that it belongs to a ws-node.
# So when closing a tag, it can be checked, if the previous 'non-tag'-node (text or whitespace), which is the one before
# the last read 'non-tag'-node, was a actually _not_ a ws-node, but instead a text-node. In that case, the from-value of
# the last read 'non-tag'-node has to be corrected (see [1]),
#
# For whitespace-nodes $add_one is set to 0, so when opening the next tag (in the above example the 2nd 's'-tag), no
# additional 1 is added (because this was already done by the whitespace-node itself when incrementing the variable $dl).
#
# [1]
# Now, what happens, when 2 text-nodes are _not_ seperated by a whitespace-node (e.g.: <w>Augen<c>,</c></w>)?
# In this case, the falsely increased from-value has to be decreased again by 1 when closing the enclosing tag
# (see above code fragment '... not exists $ws{ $fval - 1 } ...').
#
# [2]
# Comparing the 2 examples '<w>fu</w> <w>bar</w>' and '<w>fu</w><w> </w><w>bar</w>', is ' ' in both cases handled as a
# whitespace-node (XML_READER_TYPE_SIGNIFICANT_WHITESPACE).
#
# The from-index of the 2nd w-tag in the second example refers to 'bar', which may not have been the intention
# (even though '<w> </w>' doesn't make a lot of sense). TODO: could this be a bug?
#
# Empty tags also cling to the next text-token - e.g. in '<w>tok1</w> <w>tok2</w><a><b/></a> <w>tok3</w>' are the from-
# and to-indizes for the tags 'a' and 'b' both 12, which is the start-index of the token 'tok3'.
if( $e->[0] == XML_READER_TYPE_SIGNIFICANT_WHITESPACE ){
# ~ whitespace-node ~
# ~ whitespace related issue ~
$add_one = 0;
$ws{ $dl }++; # state, that this from-index belongs to a whitespace-node
# ('++' doesn't mean a thing here - maybe it could be used for a consistency check)
}else{
# ~ text-node ~
$add_one = 1;
}
# ~ update $data and $dl ~
$data .= $e->[1];
$dl += length( $e->[1] ); # update length of $data
#~~~~~
# until here: text- and whitespace-nodes
#~~~~~
#elsif ( $e->[0] == XML_READER_TYPE_ATTRIBUTE ) # attribute-node
# note: attributes cannot be processed like this ( => use 'XCT_ATTRIBUTE_ARRAY' - see above )
}else{ # not yet handled type
die "ERROR ($0): Not yet handled type (\$e->[0]=".$e->[0].") ... => Aborting\n";
}
} # end: foreach iteration
} # end: sub retr_info
sub write_structures { # called from main()
# ~ write @structures ~
#print STDERR "$0: write_structures(): ...\n";
if ( $dir eq "" ){
print STDERR "WARNING ($0): write_structures(): empty textSigle => nothing to do ...\n";
return;
}
$output = "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<?xml-model href=\"span.rng\" type=\"application/xml\""
." schematypens=\"http://relaxng.org/ns/structure/1.0\"?>\n\n<layer docid=\""
.decode( "UTF-8", $text_id_esc )."\" xmlns=\"http://ids-mannheim.de/ns/KorAP\" version=\"KorAP-0.4\">\n <spanList>\n"; # convert binary string to text string
$c = 0;
foreach $ref ( @structures ){
( @{$ref} == 4 )?( $idx = 1 ):( $idx = @{$ref}-3 ); # if array '@{$ref}' doesn't contain attributes, then the number of elements in this array is 4
# (name, from, to, rec_level), otherwise >4
# correct last from-value ( if the 'second to last' from-value refers to an s-tag, then the last from-value is one to big - see retr_info )
if( $#structures == $c && ${$ref}[ $idx ] == ${$ref}[ $idx+1 ] + 1 ){
${$ref}[$idx] = ${$ref}[ $idx+1 ];
}
# this consistency check is already done in 'retr_info()'
#elsif( ${$ref}[$idx] > ${$ref}[$idx+1] ){ # consistency check: abort, if this doesn't hold
# die "ERROR ($0: write_structures(): \$text_id=$text_id, \$c=$c, tag-name=${$ref}[0]):"
# ." 'from-index=${$ref}[$idx]' > 'to-index=${$ref}[$idx+1]' => please check! aborting ...\n";
# die "ERROR ($0: write_structures(): \$text_id=$text_id, \$c=$c, tag-name=${$ref}[0]):"
# ." 'from-index=${$ref}[$idx]' > 'to-index=${$ref}[$idx+1]' => please check! aborting ...\n\n$output" }
# at least 'POS' should always be there => remove constraint '$_TOKENS_PROC'
#if( $_TOKENS_PROC && ${$ref}[0] ne $_TOKENS_TAG )
if( ${$ref}[0] ne $_TOKENS_TAG ){ # $_TOKENS_TAG is already written in 'write_tokens'
# l (level): insert information about depth of element in XML-tree (top element = level 1)
$output .= " <span id=\"s$c\" from=\"${$ref}[ $idx ]\" to=\"${$ref}[ $idx+1 ]\" l=\"${$ref}[ $idx+2 ]\">\n"
." <fs type=\"struct\" xmlns=\"http://www.tei-c.org/ns/1.0\">\n"
." <f name=\"name\">${$ref}[ 0 ]</f>\n";
if ( $idx > 2 ) # attributes
{
$output .= " <f name=\"attr\">\n <fs type=\"attr\">\n";
for ( $att_idx = 1; $att_idx < $idx; $att_idx += 2 ){
# see explanation in func. 'write_tokens'
${$ref}[ $att_idx+1 ] = escape_xml(${$ref}[ $att_idx+1 ]);
# attribute (at index $att_idx) with value (at index $att_idx+1)
$output .= " <f name=\"${$ref}[ $att_idx ]\">${$ref}[ $att_idx+1 ]</f>\n";
}
$output .= " </fs>\n </f>\n";
}
$output .= " </fs>\n </span>\n";
} # fi: ... ne $_TOKENS_TAG
$c++;
} # end: foreach
$output .= " </spanList>\n</layer>";
$output = encode( "UTF-8", $output ); # convert text string to binary string
$zipper->new_stream("$_root_dir$dir/$_structure_dir/$_structure_file")
->print($output);
#print STDERR "$0: write_structures(): DONE\n";
} # end: sub write_structures
__END__
=pod
=encoding utf8
=head1 NAME
tei2korapxml - Conversion of TEI P5 based formats to KorAP-XML
=head1 SYNOPSIS
cat corpus.i5.xml | tei2korapxml > corpus.korapxml.zip
=head1 DESCRIPTION
C<tei2korapxml> is a script to convert TEI P5 and
L<I5|https://www1.ids-mannheim.de/kl/projekte/korpora/textmodell.html>
based documents to the
L<KorAP-XML format|https://github.com/KorAP/KorAP-XML-Krill#about-korap-xml>.
If no specific input is defined, data is
read from C<STDIN>. If no specific output is defined, data is written
to C<STDOUT>.
This program is usually called from inside another script.
=head1 FORMATS
=head2 Input restrictions
=over 2
=item
utf8 encoded
=item
TEI P5 formatted input with certain restrictions:
=over 4
=item
B<mandatory>: text-header with integrated textsigle, text-body
=item
B<optional>: corp-header with integrated corpsigle,
doc-header with integrated docsigle
=back
=item
all tokens inside the primary text (inside $data) may not be
newline seperated, because newlines are removed
(see code section C<~ inside text body ~>) and a conversion of newlines
into blanks between 2 tokens could lead to additional blanks,
where there should be none (e.g.: punctuation characters like C<,> or
C<.> should not be seperated from their predecessor token).
(see also code section C<~ whitespace handling ~>).
=back
=head2 Notes on the output
=over 2
=item
zip file output (default on C<stdout>) with utf8 encoded entries
(which together form the KorAP-XML format)
=back
=head1 INSTALLATION
C<tei2korapxml> requires L<libxml2-dev> bindings to build. When
these bindings are available, the preferred way to install the script is
to use L<cpanm|App::cpanminus>.
$ cpanm https://github.com/KorAP/KorAP-XML-TEI.git
In case everything went well, the C<tei2korapxml> tool will
be available on your command line immediately.
Minimum requirement for L<KorAP::XML::TEI> is Perl 5.16.
=head1 OPTIONS
=over 2
=item B<--root|-r>
The root directory for output. Defaults to C<.>.
=item B<--help|-h>
Print help information.
=item B<--version|-v>
Print version information.
=item B<--tokenizer-call|-tc>
Call an external tokenizer process, that will tokenize
a single line from STDIN and outputs one token per line.
=item B<--use-intern-tokenization|-ti>
Tokenize the data using two embedded tokenizers,
that will take an I<Aggressive> and a I<conservative>
approach.
=back
=head1 COPYRIGHT AND LICENSE
Copyright (C) 2020, L<IDS Mannheim|https://www.ids-mannheim.de/>
Author: Peter Harders
Contributors: Marc Kupietz, Carsten Schnober, Nils Diewald
L<KorAP::XML::TEI> is developed as part of the L<KorAP|https://korap.ids-mannheim.de/>
Corpus Analysis Platform at the
L<Leibniz Institute for the German Language (IDS)|http://ids-mannheim.de/>,
member of the
L<Leibniz-Gemeinschaft|http://www.leibniz-gemeinschaft.de/>.
This program is free software published under the
L<BSD-2 License|https://raw.githubusercontent.com/KorAP/KorAP-XML-TEI/master/LICENSE>.
=cut