1 package Text::Tradition::Analysis;
5 use Algorithm::Diff; # for word similarity measure
6 use Encode qw/ decode_utf8 encode_utf8 /;
9 use JSON qw/ to_json decode_json /;
12 use Text::Tradition::Analysis::Result;
13 use Text::Tradition::Stemma;
16 use vars qw/ @EXPORT_OK $VERSION /;
17 @EXPORT_OK = qw/ run_analysis group_variants analyze_variant_location wit_stringify /;
21 my $SOLVER_URL = 'http://byzantini.st/cgi-bin/graphcalc.cgi';
22 my $unsolved_problems = {};
26 Text::Tradition::Analysis - functions for stemma analysis of a tradition
30 Text::Tradition is a library for representation and analysis of collated
31 texts, particularly medieval ones. Where the Collation is the central
32 feature of a Tradition, it may also have one or more stemmata associated
33 with it, and these stemmata may be analyzed. This package provides the
38 =item * L<Text::Tradition::HasStemma> - a role that will be composed into
39 Text::Tradition objects, providing the ability for Text::Tradition::Stemma
40 objects to be associated with them.
42 =item * L<Text::Tradition::Stemma> - an object class that represents stemma
43 hypotheses, both rooted (with a single archetype) and unrooted (e.g.
46 =item * Text::Tradition::Analysis (this package). Provides functions for
47 the analysis of a given stemma against the collation within a given
55 use Text::Tradition::Analysis qw/ run_analysis analyze_variant_location /;
56 my $t = Text::Tradition->new(
57 'name' => 'this is a text',
59 'file' => '/path/to/tei_parallel_seg_file.xml' );
60 $t->add_stemma( 'dotfile' => $stemmafile );
62 my $variant_data = run_analysis( $tradition );
66 =head2 run_analysis( $tradition, %opts )
68 Runs the analysis described in analyze_variant_location on every location in the
69 collation of the given tradition, with the given options. These include:
73 =item * stemma_id - Specify which of the tradition's stemmata to use. Default
74 is 0 (i.e. the first).
76 =item * ranks - Specify a list of location ranks to analyze; exclude the rest.
78 =item * merge_types - Specify a list of relationship types, where related readings
79 should be treated as identical for the purposes of analysis.
81 =item * exclude_type1 - Exclude those ranks whose groupings have only type-1 variants.
88 use Text::Tradition::Analysis qw/ run_analysis analyze_variant_location /;
90 my $datafile = 't/data/florilegium_tei_ps.xml';
91 my $tradition = Text::Tradition->new( 'input' => 'TEI',
93 'file' => $datafile );
94 my $s = $tradition->add_stemma( 'dotfile' => 't/data/florilegium.dot' );
95 is( ref( $s ), 'Text::Tradition::Stemma', "Added stemma to tradition" );
97 my %expected_genealogical = (
128 my $data = run_analysis( $tradition, calcdsn => 'dbi:SQLite:dbname=t/data/analysis.db' );
129 my $c = $tradition->collation;
130 foreach my $row ( @{$data->{'variants'}} ) {
131 # Account for rows that used to be "not useful"
132 unless( exists $expected_genealogical{$row->{'id'}} ) {
133 $expected_genealogical{$row->{'id'}} = 1;
135 my $gen_bool = $row->{'genealogical'} ? 1 : 0;
136 is( $gen_bool, $expected_genealogical{$row->{'id'}},
137 "Got correct genealogical flag for row " . $row->{'id'} );
138 # Check that we have the right row with the right groups
139 my $rank = $row->{'id'};
140 foreach my $rdghash ( @{$row->{'readings'}} ) {
141 # Skip 'readings' that aren't really
142 next unless $c->reading( $rdghash->{'readingid'} );
144 is( $c->reading( $rdghash->{'readingid'} )->rank, $rank,
145 "Got correct reading rank" );
146 # Check the witnesses
147 my @realwits = sort $c->reading_witnesses( $rdghash->{'readingid'} );
148 my @sgrp = sort @{$rdghash->{'group'}};
149 is_deeply( \@sgrp, \@realwits, "Reading analyzed with correct groups" );
152 is( $data->{'variant_count'}, 58, "Got right total variant number" );
153 # TODO Make something meaningful of conflict count, maybe test other bits
160 my( $tradition, %opts ) = @_;
161 my $c = $tradition->collation;
162 my $aclabel = $c->ac_label;
164 my $stemma_id = $opts{'stemma_id'} || 0;
165 my @ranks = ref( $opts{'ranks'} ) eq 'ARRAY' ? @{$opts{'ranks'}} : ();
166 my $collapse = Set::Scalar->new();
167 if( $opts{'merge_types'} && ref( $opts{'merge_types'} ) eq 'ARRAY' ) {
168 $collapse->insert( @{$opts{'merge_types'}} );
169 } elsif( $opts{'merge_types'} ) {
170 $collapse->insert( $opts{'merge_types'} );
173 # Make sure we have a lookup DB for graph calculation results; this will die
174 # if calcdir or calcdsn isn't passed.
176 if( exists $opts{'calcdir'} ) {
177 $dir = delete $opts{'calcdir'}
178 } elsif ( exists $opts{'calcdsn'} ) {
179 eval { require Text::Tradition::Directory };
181 throw( "Could not instantiate a directory for " . $opts{'calcdsn'}
184 $dir = Text::Tradition::Directory->new( dsn => $opts{'calcdsn'} );
188 my $stemma = $tradition->stemma( $stemma_id );
190 # Figure out which witnesses we are working with - that is, the ones that
191 # appear both in the stemma and in the tradition. All others are 'lacunose'
193 my $lacunose = Set::Scalar->new( $stemma->hypotheticals );
194 my $stemma_wits = Set::Scalar->new( $stemma->witnesses );
195 my $tradition_wits = Set::Scalar->new( map { $_->sigil } $tradition->witnesses );
196 $lacunose->insert( $stemma_wits->symmetric_difference( $tradition_wits )->members );
198 # Find and mark 'common' ranks for exclusion, unless they were
199 # explicitly specified.
202 foreach my $rdg ( $c->common_readings ) {
203 $common_rank{$rdg->rank} = 1;
205 @ranks = grep { !$common_rank{$_} } ( 1 .. $c->end->rank-1 );
208 # Group the variants to send to the solver
213 foreach my $rank ( @ranks ) {
214 my $missing = $lacunose->clone();
215 my $rankgroup = group_variants( $tradition, $rank, $missing, $moved, $collapse );
216 # Filter out any empty rankgroups
217 # (e.g. from the later rank for a transposition)
218 next unless keys %$rankgroup;
219 # Get the graph for this rankgroup
220 my $rankgraph = _graph_for_grouping( $stemma, $rankgroup, $missing, $aclabel );
221 if( $opts{'exclude_type1'} ) {
222 # Check to see whether this is a "useful" group.
223 next unless _useful_variant( $rankgroup, $rankgraph, $aclabel );
225 push( @use_ranks, $rank );
226 push( @groups, { grouping => $rankgroup, graph => $rankgraph } );
227 $lacunae{$rank} = $missing;
232 $answer = solve_variants( $dir, @groups );
233 } catch ( Text::Tradition::Error $e ) {
234 if( $e->message =~ /IDP/ ) {
235 # Something is wrong with the solver; make the variants table anyway
236 $answer->{'variants'} = [];
237 map { push( @{$answer->{'variants'}}, _init_unsolved( $_, 'IDP error' ) ) }
240 # Something else is wrong; error out.
245 # Do further analysis on the answer
246 my $conflict_count = 0;
247 my $reversion_count = 0;
248 foreach my $idx ( 0 .. $#use_ranks ) {
249 my $location = $answer->{'variants'}->[$idx];
250 # Add the rank back in
251 my $rank = $use_ranks[$idx];
252 $location->{'id'} = $rank;
253 # Note what our lacunae are
255 map { $lmiss{$_} = 1 } @{$lacunae{$use_ranks[$idx]}};
256 $location->{'missing'} = [ keys %lmiss ];
258 # Run the extra analysis we need.
259 ## TODO We run through all the variants in this call, so
260 ## why not add the reading data there instead of here below?
261 my $graph = $groups[$idx]->{graph};
262 analyze_location( $tradition, $graph, $location, \%lmiss );
265 # Do the final post-analysis tidying up of the data.
266 foreach my $rdghash ( @{$location->{'readings'}} ) {
267 $conflict_count++ if $rdghash->{'is_conflict'};
268 $reversion_count++ if $rdghash->{'is_reverted'};
269 # Add the reading text back in, setting display value as needed
270 my $rdg = $c->reading( $rdghash->{'readingid'} );
272 $rdghash->{'text'} = $rdg->text .
273 ( $rdg->rank == $rank ? '' : ' [' . $rdg->rank . ']' );
274 if( $rdg->does( 'Text::Tradition::Morphology' ) ) {
275 $rdghash->{'is_ungrammatical'} = $rdg->grammar_invalid;
276 $rdghash->{'is_nonsense'} = $rdg->is_nonsense;
279 # Remove lacunose witnesses from this reading's list now that the
282 map { push( @realgroup, $_ ) unless $lmiss{$_} } @{$rdghash->{'group'}};
283 $rdghash->{'group'} = \@realgroup;
284 # Note any layered witnesses that appear in this group
285 foreach( @realgroup ) {
286 if( $_ =~ /^(.*)\Q$aclabel\E$/ ) {
287 push( @layerwits, $1 );
291 $location->{'layerwits'} = \@layerwits if @layerwits;
293 $answer->{'conflict_count'} = $conflict_count;
294 $answer->{'reversion_count'} = $reversion_count;
299 =head2 group_variants( $tradition, $rank, $lacunose, $transposed, $merge_relationship_types )
301 Groups the variants at the given $rank of the collation, treating any
302 relationships in the set $merge_relationship_types as equivalent.
303 $lacunose should be a reference to an array, to which the sigla of lacunose
304 witnesses at this rank will be appended; $transposed should be a reference
305 to a hash, wherein the identities of transposed readings and their
306 relatives will be stored.
308 Returns a hash $group_readings where $rdg is attested by the witnesses listed
309 in $group_readings->{$rdg}.
313 # Return group_readings, groups, lacunose
315 my( $tradition, $rank, $lacunose, $transposed, $collapse ) = @_;
316 my $c = $tradition->collation;
317 my $aclabel = $c->ac_label;
318 my $table = $c->alignment_table;
319 # Get the alignment table readings
320 my %readings_at_rank;
321 my $check_for_gaps = Set::Scalar->new();
323 my $has_transposition;
324 foreach my $tablewit ( @{$table->{'alignment'}} ) {
325 my $rdg = $tablewit->{'tokens'}->[$rank-1];
326 my $wit = $tablewit->{'witness'};
327 # Exclude the witness if it is "lacunose" which if we got here
328 # means "not in the stemma".
329 next if _is_lacunose( $wit, $lacunose, $aclabel );
330 # Note if the witness is actually in a lacuna
331 if( $rdg && $rdg->{'t'}->is_lacuna ) {
332 _add_to_witlist( $wit, $lacunose, $aclabel );
333 # Otherwise the witness either has a positive reading...
335 # If the reading has been counted elsewhere as a transposition, ignore it.
336 if( $transposed->{$rdg->{'t'}->id} ) {
337 # TODO Does this cope with three-way transpositions?
338 map { $moved_wits{$_} = 1 } @{$transposed->{$rdg->{'t'}->id}};
341 # Otherwise, record it...
342 $readings_at_rank{$rdg->{'t'}->id} = $rdg->{'t'};
343 # ...and grab any transpositions, and their relations.
344 my @transp = grep { $_->rank != $rank } $rdg->{'t'}->related_readings();
345 foreach my $trdg ( @transp ) {
346 next if exists $readings_at_rank{$trdg->id};
347 $has_transposition = 1;
348 my @affected_wits = _table_witnesses(
349 $table, $trdg, $lacunose, $aclabel );
350 next unless @affected_wits;
351 map { $moved_wits{$_} = 1 } @affected_wits;
352 $transposed->{$trdg->id} =
353 [ _table_witnesses( $table, $rdg->{'t'}, $lacunose, $aclabel ) ];
354 $readings_at_rank{$trdg->id} = $trdg;
356 # ...or it is empty, ergo a gap.
358 _add_to_witlist( $wit, $check_for_gaps, $aclabel );
361 my $gap_wits = Set::Scalar->new();
362 map { _add_to_witlist( $_, $gap_wits, $aclabel )
363 unless $moved_wits{$_} } $check_for_gaps->members;
365 # Group the readings, collapsing groups by relationship if needed.
366 my $grouped_readings = {};
367 foreach my $rdg ( values %readings_at_rank ) {
368 # Skip readings that have been collapsed into others.
369 next if exists $grouped_readings->{$rdg->id}
370 && $grouped_readings->{$rdg->id} eq 'COLLAPSE';
371 # Get the witness list, including from readings collapsed into this one.
372 my @wits = _table_witnesses( $table, $rdg, $lacunose, $aclabel );
373 if( $collapse && $collapse->size ) {
374 my $filter = sub { $collapse->has( $_[0]->type ) };
375 foreach my $other ( $rdg->related_readings( $filter ) ) {
376 my @otherwits = _table_witnesses( $table, $other, $lacunose, $aclabel );
377 push( @wits, @otherwits );
378 $grouped_readings->{$other->id} = 'COLLAPSE';
381 $grouped_readings->{$rdg->id} = Set::Scalar->new( @wits );
383 if( $gap_wits->members ) {
384 $grouped_readings->{'(omitted)'} = $gap_wits;
387 # Get rid of our collapsed readings
388 map { delete $grouped_readings->{$_} if(
389 $grouped_readings->{$_} eq 'COLLAPSE'
390 || $grouped_readings->{$_}->is_empty ) }
391 keys %$grouped_readings;
393 # If something was transposed, check the groups for doubled-up readings
394 if( $has_transposition ) {
395 # print STDERR "Group for rank $rank:\n";
396 # map { print STDERR "\t$_: " . join( ' ' , @{$grouped_readings->{$_}} ) . "\n" }
397 # keys %$grouped_readings;
398 _check_transposed_consistency( $c, $rank, $transposed, $grouped_readings );
402 return $grouped_readings;
405 # Helper function to query the alignment table for all witnesses (a.c. included)
406 # that have a given reading at its rank.
407 sub _table_witnesses {
408 my( $table, $trdg, $lacunose, $aclabel ) = @_;
409 my $tableidx = $trdg->rank - 1;
410 my $has_reading = Set::Scalar->new();
411 foreach my $row ( @{$table->{'alignment'}} ) {
412 my $wit = $row->{'witness'};
413 next if _is_lacunose( $wit, $lacunose, $aclabel );
414 my $rdg = $row->{'tokens'}->[$tableidx];
415 next unless exists $rdg->{'t'} && defined $rdg->{'t'};
416 _add_to_witlist( $wit, $has_reading, $aclabel )
417 if $rdg->{'t'}->id eq $trdg->id;
419 return $has_reading->members;
422 # Helper function to see if a witness is lacunose even if we are asking about
425 my ( $wit, $lac, $acstr ) = @_;
426 if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
429 return $lac->has( $wit );
432 # Helper function to ensure that X and X a.c. never appear in the same list.
433 sub _add_to_witlist {
434 my( $wit, $list, $acstr ) = @_;
435 if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
436 # Don't add X a.c. if we already have X
437 return if $list->has( $1 );
439 # Delete X a.c. if we are about to add X
440 $list->delete( $wit.$acstr );
442 $list->insert( $wit );
445 sub _check_transposed_consistency {
446 my( $c, $rank, $transposed, $groupings ) = @_;
449 # Note which readings are actually at this rank, and which witnesses
450 # belong to which reading.
451 foreach my $rdg ( keys %$groupings ) {
452 my $rdgobj = $c->reading( $rdg );
453 # Count '(omitted)' as a reading at this rank
454 $thisrank{$rdg} = 1 if !$rdgobj || $rdgobj->rank == $rank;
455 map { push( @{$seen_wits{$_}}, $rdg ) } @{$groupings->{$rdg}};
457 # Our work is done if we have no witness belonging to more than one
459 my @doubled = grep { scalar @{$seen_wits{$_}} > 1 } keys %seen_wits;
460 return unless @doubled;
461 # If we have a symmetric related transposition, drop the non-rank readings.
462 if( @doubled == scalar keys %seen_wits ) {
463 foreach my $rdg ( keys %$groupings ) {
464 if( !$thisrank{$rdg} ) {
465 my $groupstr = wit_stringify( $groupings->{$rdg} );
466 my ( $matched ) = grep { $groupstr eq wit_stringify( $groupings->{$_} ) }
468 delete $groupings->{$rdg};
469 # If we found a group match, assume there is a symmetry happening.
470 # TODO think more about this
471 # print STDERR "*** Deleting symmetric reading $rdg\n";
473 delete $transposed->{$rdg};
474 warn "Found problem in evident symmetry with reading $rdg";
478 # Otherwise 'unhook' the transposed reading(s) that have duplicates.
480 foreach my $dup ( @doubled ) {
481 foreach my $rdg ( @{$seen_wits{$dup}} ) {
482 next if $thisrank{$rdg};
483 next unless exists $groupings->{$rdg};
484 # print STDERR "*** Deleting asymmetric doubled-up reading $rdg\n";
485 delete $groupings->{$rdg};
486 delete $transposed->{$rdg};
489 # and put any now-orphaned readings into an 'omitted' reading.
490 foreach my $wit ( keys %seen_wits ) {
491 unless( grep { exists $groupings->{$_} } @{$seen_wits{$wit}} ) {
492 $groupings->{'(omitted)'} = Set::Scalar->new()
493 unless exists $groupings->{'(omitted)'};
494 _add_to_witlist( $wit, $groupings->{'(omitted)'}, $c->ac_label );
500 # For the given grouping, return its situation graph based on the stemma.
501 sub _graph_for_grouping {
502 my( $stemma, $grouping, $lacunose, $aclabel ) = @_;
505 foreach my $gs ( values %$grouping ) {
507 if( $_ =~ /^(.*)\Q$aclabel\E$/ ) {
508 push( @$acwits, $1 ) unless $lacunose->has( $1 );
510 $extant->{$_} = 1 unless $lacunose->has( $_ );
516 # contig contains all extant wits and all hypothetical wits
517 # needed to make up the groups.
518 $graph = $stemma->situation_graph( $extant, $acwits, $aclabel );
519 } catch ( Text::Tradition::Error $e ) {
520 throw( "Could not extend graph with given extant and a.c. witnesses: "
523 throw( "Could not extend graph with a.c. witnesses @$acwits" );
528 =head2 solve_variants( $calcdir, @groups )
530 Looks up the set of groups in the answers provided by the external graph solver
531 service and returns a cleaned-up answer, adding the rank IDs back where they belong.
533 The answer has the form
534 { "variants" => [ array of variant location structures ],
535 "variant_count" => total,
536 "conflict_count" => number of conflicts detected,
537 "genealogical_count" => number of solutions found }
544 # Are we using a local result directory, or did we pass an empty value
547 unless( ref( $groups[0] ) eq 'HASH' ) {
548 $dir = shift @groups;
551 ## For each graph/group combo, make a Text::Tradition::Analysis::Result
552 ## object so that we can send it off for IDP lookup.
554 my $genealogical = 0; # counter
555 # TODO Optimize for unique graph problems
557 foreach my $graphproblem ( @groups ) {
558 # Construct the calc result key and look up its answer
559 my $problem = Text::Tradition::Analysis::Result->new(
560 graph => $graphproblem->{'graph'},
561 setlist => [ values %{$graphproblem->{'grouping'}} ] );
562 if( exists $problems{$problem->object_key} ) {
563 $problem = $problems{$problem->object_key};
565 $problems{$problem->object_key} = $problem;
567 $graphproblem->{'object'} = $problem;
572 my $scope = $dir->new_scope;
573 map { $results{$_} = $dir->lookup( $_ ) || $problems{$_} } keys %problems;
575 my $json = JSON->new->allow_blessed->convert_blessed->utf8->encode(
576 [ values %problems ] );
577 # Send it off and get the result
578 # print STDERR "Sending request: " . decode_utf8( $json ) . "\n";
579 my $ua = LWP::UserAgent->new();
580 my $resp = $ua->post( $SOLVER_URL, 'Content-Type' => 'application/json',
581 'Content' => $json );
583 if( $resp->is_success ) {
584 $answer = decode_json( $resp->content );
585 throw( "Unexpected answer from IDP: $answer" ) unless ref( $answer ) eq 'ARRAY';
587 throw( "IDP solver returned " . $resp->status_line . " / " . $resp->content
588 . "; cannot run graph analysis" );
590 # One more sanity check
591 throw( "Something went wrong with answer symmetricity" )
592 unless keys( %problems ) == @$answer;
593 # Convert the results
594 foreach my $a ( @$answer ) {
595 my $r = Text::Tradition::Analysis::Result->new( $a );
596 $results{$r->object_key} = $r;
600 # We now have a single JSON-encoded Result object per problem sent. Fold its
601 # answers into our variant info structure.
602 foreach my $graphproblem ( @groups ) {
603 my $result = $results{$graphproblem->{'object'}->object_key}
604 || $graphproblem->{'object'};
606 # Initialize the result structure for this graph problem
608 if( $result->status eq 'OK' ) {
609 $vstruct = { readings => [] };
610 push( @$variants, $vstruct );
612 push( @$variants, _init_unsolved( $graphproblem, $result->status ) );
616 # 1. Did the group evaluate as genealogical?
617 $vstruct->{genealogical} = $result->is_genealogical;
618 $genealogical++ if $result->is_genealogical;
620 # 2. What are the calculated minimum groupings for each variant loc?
621 foreach my $rid ( keys %{$graphproblem->{grouping}} ) {
622 my $inputset = $graphproblem->{grouping}->{$rid};
623 my $minset = $result->minimum_grouping_for( $inputset );
624 push( @{$vstruct->{readings}}, { readingid => $rid, group => $minset } );
627 # 3. What are the sources and classes calculated for each witness?
628 $vstruct->{witcopy_types} = { $result->classes };
629 $vstruct->{reading_roots} = {};
630 map { $vstruct->{reading_roots}->{$_} = 1 } $result->sources;
634 return { 'variants' => $variants,
635 'variant_count' => scalar @$variants,
636 'genealogical_count' => $genealogical };
640 my( $graphproblem, $status ) = @_;
641 my $vstruct = { 'readings' => [] };
642 $vstruct->{'unsolved'} = $status;
643 foreach my $rid ( keys %{$graphproblem->{grouping}} ) {
644 push( @{$vstruct->{readings}}, { readingid => $rid,
645 group => [ $graphproblem->{grouping}->{$rid}->members ] } );
650 =head2 analyze_location ( $tradition, $graph, $location_hash )
652 Given the tradition, its stemma graph, and the solution from the graph solver,
653 work out the rest of the information we want. For each reading we need missing,
654 conflict, reading_parents, independent_occurrence, followed, not_followed,
655 and follow_unknown. Alters the location_hash in place.
659 sub analyze_location {
660 my ( $tradition, $graph, $variant_row, $lacunose ) = @_;
661 my $c = $tradition->collation;
663 if( exists $variant_row->{'unsolved'} ) {
666 my $reading_roots = delete $variant_row->{'reading_roots'};
667 my $classinfo = delete $variant_row->{'witcopy_types'};
669 # Make a hash of all known node memberships, and make the subgraphs.
672 my $acstr = $c->ac_label;
675 # Note which witnesses positively belong to which group. This information
676 # comes ultimately from the IDP solver.
677 # Also make a note of the reading's roots.
678 foreach my $rdghash ( @{$variant_row->{'readings'}} ) {
679 my $rid = $rdghash->{'readingid'};
681 foreach my $wit ( @{$rdghash->{'group'}} ) {
682 $contig->{$wit} = $rid;
683 if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
686 if( exists $reading_roots->{$wit} && $reading_roots->{$wit} ) {
687 push( @roots, $wit );
690 $rdghash->{'independent_occurrence'} = \@roots;
693 # Now that we have all the node group memberships, calculate followed/
694 # non-followed/unknown values for each reading. Also figure out the
695 # reading's evident parent(s).
696 foreach my $rdghash ( @{$variant_row->{'readings'}} ) {
697 my $rid = $rdghash->{'readingid'};
698 my $rdg = $c->reading( $rid );
699 my @roots = @{$rdghash->{'independent_occurrence'}};
702 @reversions = grep { $classinfo->{$_} eq 'revert' }
703 $rdghash->{'group'}->members;
704 $rdghash->{'reversions'} = \@reversions;
706 my @group = @{$rdghash->{'group'}};
708 # Start figuring things out.
709 $rdghash->{'followed'} = scalar( @group )
710 - ( scalar( @roots ) + scalar( @reversions ) );
711 # Find the parent readings, if any, of this reading.
712 my $sourceparents = _find_reading_parents( $rid, $graph, $contig, @roots );
713 # Work out relationships between readings and their non-followed parent.
714 _resolve_parent_relationships( $c, $rid, $rdg, $sourceparents );
715 $rdghash->{'source_parents'} = $sourceparents;
718 my $revparents = _find_reading_parents( $rid, $graph, $contig, @reversions );
719 _resolve_parent_relationships( $c, $rid, $rdg, $revparents );
720 $rdghash->{'reversion_parents'} = $revparents;
723 # Find the number of times this reading was altered, and the number of
724 # times we're not sure.
725 my( %nofollow, %unknownfollow );
726 foreach my $wit ( @{$rdghash->{'group'}} ) {
727 foreach my $wchild ( $graph->successors( $wit ) ) {
728 if( $reading_roots->{$wchild} && $contig->{$wchild}
729 && $contig->{$wchild} ne $rid ) {
730 # It definitely changed here.
731 $nofollow{$wchild} = 1;
732 } elsif( !($contig->{$wchild}) ) {
733 # The child is a hypothetical node not definitely in
734 # any group. Answer is unknown.
735 $unknownfollow{$wchild} = 1;
736 } # else it is either in our group, or it is a non-root node in a
737 # known group and therefore is presumed to have its reading from
738 # its group, not this link.
741 $rdghash->{'not_followed'} = keys %nofollow;
742 $rdghash->{'follow_unknown'} = keys %unknownfollow;
744 # Now say whether this reading represents a conflict.
745 unless( $variant_row->{'genealogical'} ) {
746 $rdghash->{'is_conflict'} = @roots != 1;
747 $rdghash->{'is_reverted'} = scalar @reversions;
752 sub _find_reading_parents {
753 my( $rid, $graph, $contig, @list ) = @_;
755 foreach my $wit ( @list ) {
756 # Look in the stemma graph to find this witness's extant or known-reading
757 # immediate ancestor(s), and look up the reading that each ancestor holds.
758 my @check = $graph->predecessors( $wit );
761 foreach my $wparent( @check ) {
762 my $preading = $contig->{$wparent};
763 if( $preading && $preading ne $rid ) {
764 $parenthash->{$preading} = 1;
766 push( @next, $graph->predecessors( $wparent ) );
775 sub _resolve_parent_relationships {
776 my( $c, $rid, $rdg, $rdgparents ) = @_;
777 foreach my $p ( keys %$rdgparents ) {
778 # Resolve the relationship of the parent to the reading, and
779 # save it in our hash.
780 my $pobj = $c->reading( $p );
781 my $prep = $pobj ? $pobj->id . ' (' . $pobj->text . ')' : $p;
782 my $phash = { 'label' => $prep };
784 my $rel = $c->get_relationship( $p, $rid );
786 _add_to_hash( $rel, $phash );
788 # First check for a transposed relationship
789 if( $rdg->rank != $pobj->rank ) {
790 foreach my $ti ( $rdg->related_readings( 'transposition' ) ) {
791 next unless $ti->text eq $rdg->text;
792 $rel = $c->get_relationship( $ti, $pobj );
794 _add_to_hash( $rel, $phash, 1 );
799 foreach my $ti ( $pobj->related_readings( 'transposition' ) ) {
800 next unless $ti->text eq $pobj->text;
801 $rel = $c->get_relationship( $ti, $rdg );
803 _add_to_hash( $rel, $phash, 1 );
810 # and then check for sheer word similarity.
811 my $rtext = $rdg->text;
812 my $ptext = $pobj->text;
813 if( similar( $rtext, $ptext ) ) {
814 # say STDERR "Words $rtext and $ptext judged similar";
815 $phash->{relation} = { type => 'wordsimilar' };
819 $phash->{relation} = { type => 'deletion' };
821 # Get the attributes of the parent object while we are here
822 $phash->{'text'} = $pobj->text if $pobj;
823 if( $pobj && $pobj->does('Text::Tradition::Morphology') ) {
824 $phash->{'is_nonsense'} = $pobj->is_nonsense;
825 $phash->{'is_ungrammatical'} = $pobj->grammar_invalid;
827 } elsif( $p eq '(omitted)' ) {
828 $phash->{relation} = { type => 'addition' };
831 $rdgparents->{$p} = $phash;
836 my( $rel, $phash, $is_transposed ) = @_;
837 $phash->{relation} = { type => $rel->type };
838 $phash->{relation}->{transposed} = 1 if $is_transposed;
839 $phash->{relation}->{annotation} = $rel->annotation
840 if $rel->has_annotation;
843 =head2 similar( $word1, $word2 )
845 Use Algorithm::Diff to get a sense of how close the words are to each other.
846 This will hopefully handle substitutions a bit more nicely than Levenshtein.
853 my( $word1, $word2 ) = sort { length($a) <=> length($b) } @_;
854 my @let1 = split( '', lc( $word1 ) );
855 my @let2 = split( '', lc( $word2 ) );
856 my $diff = Algorithm::Diff->new( \@let1, \@let2 );
858 while( $diff->Next ) {
860 # Take off points for longer strings
861 my $cs = $diff->Range(1) - 2;
864 } elsif( !$diff->Items(1) ) {
865 $mag += $diff->Range(2);
866 } elsif( !$diff->Items(2) ) {
867 $mag += $diff->Range(1);
869 # Split the difference for substitutions
870 my $c1 = $diff->Range(1) || 1;
871 my $c2 = $diff->Range(2) || 1;
872 my $cd = ( $c1 + $c2 ) / 2;
876 return ( $mag <= length( $word1 ) / 2 );
879 sub _useful_variant {
880 my( $rankgroup, $rankgraph, $acstr ) = @_;
882 # Sort by group size and return
884 foreach my $rdg ( keys %$rankgroup ) {
885 my @wits = $rankgroup->{$rdg}->members;
889 $is_useful++ unless( $rankgraph->is_sink_vertex( $wits[0] )
890 || $wits[0] =~ /\Q$acstr\E$/ );
893 return $is_useful > 1;
896 =head2 wit_stringify( $groups )
898 Takes an array of witness groupings and produces a string like
899 ['A','B'] / ['C','D','E'] / ['F']
906 # If we were passed an array of witnesses instead of an array of
907 # groupings, then "group" the witnesses first.
908 unless( ref( $groups->[0] ) ) {
909 my $mkgrp = [ $groups ];
912 foreach my $g ( @$groups ) {
913 push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' );
915 return join( ' / ', @gst );
921 Text::Tradition::Error->throw(
922 'ident' => 'Analysis error',
929 This package is free software and is provided "as is" without express
930 or implied warranty. You can redistribute it and/or modify it under
931 the same terms as Perl itself.
935 Tara L Andrews E<lt>aurum@cpan.orgE<gt>