| use strict; |
| use warnings; |
| use Test::More; |
| use Data::Dumper; |
| use JSON::XS; |
| |
| use Benchmark qw/:hireswallclock/; |
| |
| my $t = Benchmark->new; |
| |
| use utf8; |
| use lib 'lib', '../lib'; |
| |
| use File::Basename 'dirname'; |
| use File::Spec::Functions 'catdir'; |
| |
| use_ok('KorAP::XML::Krill'); |
| |
| # This will check preliminary HNC-Files |
| |
| # HNC/DOC00001/00001 |
| my $path = catdir(dirname(__FILE__), '../corpus/HNC/DOC00001/00001'); |
| |
| ok(my $doc = KorAP::XML::Krill->new( path => $path . '/' ), 'Load Korap::Document'); |
| ok($doc->parse, 'Parse document'); |
| |
| is($doc->text_sigle, 'HNC/DOC00001/00001', 'Correct text sigle'); |
| is($doc->doc_sigle, 'HNC/DOC00001', 'Correct document sigle'); |
| is($doc->corpus_sigle, 'HNC', 'Correct corpus sigle'); |
| |
| my $meta = $doc->meta; |
| is($meta->{title}, 'GNU Free Documentation License', 'Title'); |
| is($meta->{pub_place}, 'H_PUBPLACE', 'PubPlace'); |
| |
| # Defined on document level as |
| # idsHeader > fileDesc > publicationStmt > pubDate == 2005/08/16 |
| # idsHeader > fileDesc > biblFull > publicationStmt > pubDate == 2003/07/08-2014/05/03 |
| # idsHeader > fileDesc > biblFull > publicationStmt > sourceDesc > biblStruct > monogr > imprint > pubDate == 2003/07/08-2014/05/03 |
| # is($meta->{pub_date}, '20030708', 'Publication date'); |
| ok(!$meta->{sub_title}, 'SubTitle'); |
| is($meta->{author}, 'Addbot', 'Author'); |
| |
| is($meta->{publisher}, 'H_PUBLISHER', 'Publisher'); |
| ok(!$meta->{editor}, 'Editor'); |
| ok(!$meta->{translator}, 'Translator'); |
| |
| ok(!$meta->{text_type}, 'Correct Text Type'); |
| ok(!$meta->{text_type_art}, 'Correct Text Type Art'); |
| ok(!$meta->{text_type_ref}, 'Correct Text Type Ref'); |
| ok(!$meta->{text_column}, 'Correct Text Column'); |
| ok(!$meta->{text_domain}, 'Correct Text Domain'); |
| is($meta->{creation_date}, '20130302', 'Creation Date'); |
| |
| ok(!$meta->{pages}, 'Pages'); |
| ok(!$meta->{file_edition_statement}, 'File Ed Statement'); |
| ok(!$meta->{bibl_edition_statement}, 'Bibl Ed Statement'); |
| ok(!$meta->{reference}, 'Reference'); |
| is($meta->{language}, 'hu', 'Language'); |
| |
| ok(!$meta->{corpus_title}, 'Correct Corpus title'); |
| ok(!$meta->{corpus_sub_title}, 'Correct Corpus Sub title'); |
| ok(!$meta->{corpus_author}, 'Correct Corpus author'); |
| ok(!$meta->{corpus_editor}, 'Correct Corpus editor'); |
| |
| is($meta->{doc_title}, 'MNSZ hivatalos korpusz: Wikipédia cikkek', 'Correct Doc title'); |
| ok(!$meta->{doc_sub_title}, 'Correct Doc Sub title'); |
| ok(!$meta->{doc_author}, 'Correct Doc author'); |
| ok(!$meta->{doc_editor}, 'Correct Doc editor'); |
| |
| # Tokenization |
| use_ok('KorAP::XML::Tokenizer'); |
| |
| my ($token_base_foundry, $token_base_layer) = (qw/HNC Morpho/); |
| |
| # Get tokenization |
| my $tokens = KorAP::XML::Tokenizer->new( |
| path => $doc->path, |
| doc => $doc, |
| foundry => $token_base_foundry, |
| layer => $token_base_layer, |
| name => 'tokens' |
| ); |
| ok($tokens, 'Token Object is fine'); |
| ok($tokens->parse, 'Token parsing is fine'); |
| |
| my $output = decode_json( $tokens->to_json ); |
| |
| is($output->{data}->{stream}->[0]->[1], '<>:base/s:t$<b>64<i>0<i>4368<i>577<b>0', 't'); |
| is($output->{data}->{stream}->[0]->[3], 'i:addbot', 't'); |
| |
| |
| ## Base |
| ok($tokens->add('DeReKo', 'Structure', 'base_sentences_paragraphs'), 'DeReKo'); |
| ok($tokens->add('HNC', 'Morpho'), 'Add HNC Morphology'); |
| |
| $output = $tokens->to_data; |
| |
| is($output->{data}->{foundries}, 'dereko dereko/structure dereko/structure/base_sentences_paragraphs hnc hnc/morpho', 'Foundries'); |
| |
| is($output->{data}->{layerInfos}, 'dereko/s=spans hnc/l=tokens hnc/m=tokens hnc/p=tokens', 'layerInfos'); |
| |
| my $token = join('||', @{$output->{data}->{stream}->[7]}); |
| |
| like($token, qr!hnc/l:free!, 'data'); |
| like($token, qr!hnc/m:compound:n!, 'data'); |
| like($token, qr!hnc/m:hyphenated:n!, 'data'); |
| like($token, qr!hnc/m:mboundary:free!, 'data'); |
| like($token, qr!hnc/m:morphemes:ZERO::NOM!, 'data'); |
| like($token, qr!hnc/m:stem:free::FN!, 'data'); |
| like($token, qr!hnc/p:FN\.NOM!, 'data'); |
| like($token, qr!i:free!, 'data'); |
| like($token, qr!s:Free!, 'data'); |
| |
| $token = join('||', @{$output->{data}->{stream}->[30]}); |
| |
| like($token, qr!hnc/l:tervez!, 'data'); |
| like($token, qr!hnc/m:compound:n!, 'data'); |
| like($token, qr!hnc/m:hyphenated:n!, 'data'); |
| like($token, qr!hnc/m:mboundary:tervez\+ett!, 'data'); |
| like($token, qr!hnc/m:morphemes:ett::_MIB ZERO::NOM!, 'data'); |
| like($token, qr!hnc/m:stem:tervez::IGE!, 'data'); |
| like($token, qr!hnc/p:IGE\._MIB\.NOM!, 'data'); |
| like($token, qr!i:tervezett!, 'data'); |
| like($token, qr!s:tervezett!, 'data'); |
| |
| done_testing; |
| __END__ |