grammar_cop 0.1.0
Sign up to get free protection for your applications and to get access to all the features.
- data/.DS_Store +0 -0
- data/.gitignore +4 -0
- data/Gemfile +4 -0
- data/Rakefile +8 -0
- data/data/.DS_Store +0 -0
- data/data/Makefile +511 -0
- data/data/Makefile.am +4 -0
- data/data/Makefile.in +511 -0
- data/data/de/.DS_Store +0 -0
- data/data/de/4.0.affix +7 -0
- data/data/de/4.0.dict +474 -0
- data/data/de/Makefile +387 -0
- data/data/de/Makefile.am +9 -0
- data/data/de/Makefile.in +387 -0
- data/data/en/.DS_Store +0 -0
- data/data/en/4.0.affix +26 -0
- data/data/en/4.0.batch +1002 -0
- data/data/en/4.0.biolg.batch +411 -0
- data/data/en/4.0.constituent-knowledge +127 -0
- data/data/en/4.0.dict +8759 -0
- data/data/en/4.0.dict.m4 +6928 -0
- data/data/en/4.0.enwiki.batch +14 -0
- data/data/en/4.0.fixes.batch +2776 -0
- data/data/en/4.0.knowledge +306 -0
- data/data/en/4.0.regex +225 -0
- data/data/en/4.0.voa.batch +114 -0
- data/data/en/Makefile +554 -0
- data/data/en/Makefile.am +19 -0
- data/data/en/Makefile.in +554 -0
- data/data/en/README +173 -0
- data/data/en/tiny.dict +157 -0
- data/data/en/words/.DS_Store +0 -0
- data/data/en/words/Makefile +456 -0
- data/data/en/words/Makefile.am +78 -0
- data/data/en/words/Makefile.in +456 -0
- data/data/en/words/currency +205 -0
- data/data/en/words/currency.p +28 -0
- data/data/en/words/entities.given-bisex.sing +39 -0
- data/data/en/words/entities.given-female.sing +4141 -0
- data/data/en/words/entities.given-male.sing +1633 -0
- data/data/en/words/entities.locations.sing +68 -0
- data/data/en/words/entities.national.sing +253 -0
- data/data/en/words/entities.organizations.sing +7 -0
- data/data/en/words/entities.us-states.sing +11 -0
- data/data/en/words/units.1 +45 -0
- data/data/en/words/units.1.dot +4 -0
- data/data/en/words/units.3 +2 -0
- data/data/en/words/units.4 +5 -0
- data/data/en/words/units.4.dot +1 -0
- data/data/en/words/words-medical.adv.1 +1191 -0
- data/data/en/words/words-medical.prep.1 +67 -0
- data/data/en/words/words-medical.v.4.1 +2835 -0
- data/data/en/words/words-medical.v.4.2 +2848 -0
- data/data/en/words/words-medical.v.4.3 +3011 -0
- data/data/en/words/words-medical.v.4.4 +3036 -0
- data/data/en/words/words-medical.v.4.5 +3050 -0
- data/data/en/words/words.adj.1 +6794 -0
- data/data/en/words/words.adj.2 +638 -0
- data/data/en/words/words.adj.3 +667 -0
- data/data/en/words/words.adv.1 +1573 -0
- data/data/en/words/words.adv.2 +67 -0
- data/data/en/words/words.adv.3 +157 -0
- data/data/en/words/words.adv.4 +80 -0
- data/data/en/words/words.n.1 +11464 -0
- data/data/en/words/words.n.1.wiki +264 -0
- data/data/en/words/words.n.2.s +2017 -0
- data/data/en/words/words.n.2.s.biolg +1 -0
- data/data/en/words/words.n.2.s.wiki +298 -0
- data/data/en/words/words.n.2.x +65 -0
- data/data/en/words/words.n.2.x.wiki +10 -0
- data/data/en/words/words.n.3 +5717 -0
- data/data/en/words/words.n.t +23 -0
- data/data/en/words/words.v.1.1 +1038 -0
- data/data/en/words/words.v.1.2 +1043 -0
- data/data/en/words/words.v.1.3 +1052 -0
- data/data/en/words/words.v.1.4 +1023 -0
- data/data/en/words/words.v.1.p +17 -0
- data/data/en/words/words.v.10.1 +14 -0
- data/data/en/words/words.v.10.2 +15 -0
- data/data/en/words/words.v.10.3 +88 -0
- data/data/en/words/words.v.10.4 +17 -0
- data/data/en/words/words.v.2.1 +1253 -0
- data/data/en/words/words.v.2.2 +1304 -0
- data/data/en/words/words.v.2.3 +1280 -0
- data/data/en/words/words.v.2.4 +1285 -0
- data/data/en/words/words.v.2.5 +1287 -0
- data/data/en/words/words.v.4.1 +2472 -0
- data/data/en/words/words.v.4.2 +2487 -0
- data/data/en/words/words.v.4.3 +2441 -0
- data/data/en/words/words.v.4.4 +2478 -0
- data/data/en/words/words.v.4.5 +2483 -0
- data/data/en/words/words.v.5.1 +98 -0
- data/data/en/words/words.v.5.2 +98 -0
- data/data/en/words/words.v.5.3 +103 -0
- data/data/en/words/words.v.5.4 +102 -0
- data/data/en/words/words.v.6.1 +388 -0
- data/data/en/words/words.v.6.2 +401 -0
- data/data/en/words/words.v.6.3 +397 -0
- data/data/en/words/words.v.6.4 +405 -0
- data/data/en/words/words.v.6.5 +401 -0
- data/data/en/words/words.v.8.1 +117 -0
- data/data/en/words/words.v.8.2 +118 -0
- data/data/en/words/words.v.8.3 +118 -0
- data/data/en/words/words.v.8.4 +119 -0
- data/data/en/words/words.v.8.5 +119 -0
- data/data/en/words/words.y +104 -0
- data/data/lt/.DS_Store +0 -0
- data/data/lt/4.0.affix +6 -0
- data/data/lt/4.0.constituent-knowledge +24 -0
- data/data/lt/4.0.dict +135 -0
- data/data/lt/4.0.knowledge +38 -0
- data/data/lt/Makefile +389 -0
- data/data/lt/Makefile.am +11 -0
- data/data/lt/Makefile.in +389 -0
- data/ext/.DS_Store +0 -0
- data/ext/link_grammar/.DS_Store +0 -0
- data/ext/link_grammar/extconf.rb +2 -0
- data/ext/link_grammar/link-grammar/.DS_Store +0 -0
- data/ext/link_grammar/link-grammar/.deps/analyze-linkage.Plo +198 -0
- data/ext/link_grammar/link-grammar/.deps/and.Plo +202 -0
- data/ext/link_grammar/link-grammar/.deps/api.Plo +244 -0
- data/ext/link_grammar/link-grammar/.deps/build-disjuncts.Plo +212 -0
- data/ext/link_grammar/link-grammar/.deps/command-line.Plo +201 -0
- data/ext/link_grammar/link-grammar/.deps/constituents.Plo +201 -0
- data/ext/link_grammar/link-grammar/.deps/count.Plo +202 -0
- data/ext/link_grammar/link-grammar/.deps/disjunct-utils.Plo +126 -0
- data/ext/link_grammar/link-grammar/.deps/disjuncts.Plo +123 -0
- data/ext/link_grammar/link-grammar/.deps/error.Plo +121 -0
- data/ext/link_grammar/link-grammar/.deps/expand.Plo +133 -0
- data/ext/link_grammar/link-grammar/.deps/extract-links.Plo +198 -0
- data/ext/link_grammar/link-grammar/.deps/fast-match.Plo +200 -0
- data/ext/link_grammar/link-grammar/.deps/idiom.Plo +200 -0
- data/ext/link_grammar/link-grammar/.deps/jni-client.Plo +217 -0
- data/ext/link_grammar/link-grammar/.deps/link-parser.Po +1 -0
- data/ext/link_grammar/link-grammar/.deps/massage.Plo +202 -0
- data/ext/link_grammar/link-grammar/.deps/post-process.Plo +202 -0
- data/ext/link_grammar/link-grammar/.deps/pp_knowledge.Plo +202 -0
- data/ext/link_grammar/link-grammar/.deps/pp_lexer.Plo +201 -0
- data/ext/link_grammar/link-grammar/.deps/pp_linkset.Plo +200 -0
- data/ext/link_grammar/link-grammar/.deps/prefix.Plo +102 -0
- data/ext/link_grammar/link-grammar/.deps/preparation.Plo +202 -0
- data/ext/link_grammar/link-grammar/.deps/print-util.Plo +200 -0
- data/ext/link_grammar/link-grammar/.deps/print.Plo +201 -0
- data/ext/link_grammar/link-grammar/.deps/prune.Plo +202 -0
- data/ext/link_grammar/link-grammar/.deps/read-dict.Plo +223 -0
- data/ext/link_grammar/link-grammar/.deps/read-regex.Plo +123 -0
- data/ext/link_grammar/link-grammar/.deps/regex-morph.Plo +131 -0
- data/ext/link_grammar/link-grammar/.deps/resources.Plo +203 -0
- data/ext/link_grammar/link-grammar/.deps/spellcheck-aspell.Plo +1 -0
- data/ext/link_grammar/link-grammar/.deps/spellcheck-hun.Plo +115 -0
- data/ext/link_grammar/link-grammar/.deps/string-set.Plo +198 -0
- data/ext/link_grammar/link-grammar/.deps/tokenize.Plo +160 -0
- data/ext/link_grammar/link-grammar/.deps/utilities.Plo +222 -0
- data/ext/link_grammar/link-grammar/.deps/word-file.Plo +201 -0
- data/ext/link_grammar/link-grammar/.deps/word-utils.Plo +212 -0
- data/ext/link_grammar/link-grammar/.libs/analyze-linkage.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/and.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/api.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/build-disjuncts.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/command-line.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/constituents.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/count.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/disjunct-utils.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/disjuncts.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/error.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/expand.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/extract-links.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/fast-match.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/idiom.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/jni-client.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar-java-symbols.expsym +31 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar-java.4.dylib +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar-java.4.dylib.dSYM/Contents/Info.plist +20 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar-java.4.dylib.dSYM/Contents/Resources/DWARF/liblink-grammar-java.4.dylib +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar-java.a +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar-java.dylib +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar-symbols.expsym +194 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar.4.dylib +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar.4.dylib.dSYM/Contents/Info.plist +20 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar.4.dylib.dSYM/Contents/Resources/DWARF/liblink-grammar.4.dylib +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar.a +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar.dylib +0 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar.la +41 -0
- data/ext/link_grammar/link-grammar/.libs/liblink-grammar.lai +41 -0
- data/ext/link_grammar/link-grammar/.libs/massage.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/post-process.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/pp_knowledge.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/pp_lexer.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/pp_linkset.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/prefix.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/preparation.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/print-util.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/print.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/prune.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/read-dict.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/read-regex.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/regex-morph.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/resources.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/spellcheck-aspell.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/spellcheck-hun.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/string-set.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/tokenize.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/utilities.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/word-file.o +0 -0
- data/ext/link_grammar/link-grammar/.libs/word-utils.o +0 -0
- data/ext/link_grammar/link-grammar/Makefile +900 -0
- data/ext/link_grammar/link-grammar/Makefile.am +202 -0
- data/ext/link_grammar/link-grammar/Makefile.in +900 -0
- data/ext/link_grammar/link-grammar/analyze-linkage.c +1317 -0
- data/ext/link_grammar/link-grammar/analyze-linkage.h +24 -0
- data/ext/link_grammar/link-grammar/and.c +1603 -0
- data/ext/link_grammar/link-grammar/and.h +27 -0
- data/ext/link_grammar/link-grammar/api-structures.h +362 -0
- data/ext/link_grammar/link-grammar/api-types.h +72 -0
- data/ext/link_grammar/link-grammar/api.c +1887 -0
- data/ext/link_grammar/link-grammar/api.h +96 -0
- data/ext/link_grammar/link-grammar/autoit/.DS_Store +0 -0
- data/ext/link_grammar/link-grammar/autoit/README +10 -0
- data/ext/link_grammar/link-grammar/autoit/_LGTest.au3 +22 -0
- data/ext/link_grammar/link-grammar/autoit/_LinkGrammar.au3 +545 -0
- data/ext/link_grammar/link-grammar/build-disjuncts.c +487 -0
- data/ext/link_grammar/link-grammar/build-disjuncts.h +21 -0
- data/ext/link_grammar/link-grammar/command-line.c +458 -0
- data/ext/link_grammar/link-grammar/command-line.h +15 -0
- data/ext/link_grammar/link-grammar/constituents.c +1836 -0
- data/ext/link_grammar/link-grammar/constituents.h +26 -0
- data/ext/link_grammar/link-grammar/corpus/.DS_Store +0 -0
- data/ext/link_grammar/link-grammar/corpus/.deps/cluster.Plo +1 -0
- data/ext/link_grammar/link-grammar/corpus/.deps/corpus.Plo +1 -0
- data/ext/link_grammar/link-grammar/corpus/Makefile +527 -0
- data/ext/link_grammar/link-grammar/corpus/Makefile.am +46 -0
- data/ext/link_grammar/link-grammar/corpus/Makefile.in +527 -0
- data/ext/link_grammar/link-grammar/corpus/README +17 -0
- data/ext/link_grammar/link-grammar/corpus/cluster.c +286 -0
- data/ext/link_grammar/link-grammar/corpus/cluster.h +32 -0
- data/ext/link_grammar/link-grammar/corpus/corpus.c +483 -0
- data/ext/link_grammar/link-grammar/corpus/corpus.h +46 -0
- data/ext/link_grammar/link-grammar/count.c +828 -0
- data/ext/link_grammar/link-grammar/count.h +25 -0
- data/ext/link_grammar/link-grammar/disjunct-utils.c +261 -0
- data/ext/link_grammar/link-grammar/disjunct-utils.h +27 -0
- data/ext/link_grammar/link-grammar/disjuncts.c +138 -0
- data/ext/link_grammar/link-grammar/disjuncts.h +13 -0
- data/ext/link_grammar/link-grammar/error.c +92 -0
- data/ext/link_grammar/link-grammar/error.h +35 -0
- data/ext/link_grammar/link-grammar/expand.c +67 -0
- data/ext/link_grammar/link-grammar/expand.h +13 -0
- data/ext/link_grammar/link-grammar/externs.h +22 -0
- data/ext/link_grammar/link-grammar/extract-links.c +625 -0
- data/ext/link_grammar/link-grammar/extract-links.h +16 -0
- data/ext/link_grammar/link-grammar/fast-match.c +309 -0
- data/ext/link_grammar/link-grammar/fast-match.h +17 -0
- data/ext/link_grammar/link-grammar/idiom.c +373 -0
- data/ext/link_grammar/link-grammar/idiom.h +15 -0
- data/ext/link_grammar/link-grammar/jni-client.c +779 -0
- data/ext/link_grammar/link-grammar/jni-client.h +236 -0
- data/ext/link_grammar/link-grammar/liblink-grammar-java.la +42 -0
- data/ext/link_grammar/link-grammar/liblink-grammar.la +41 -0
- data/ext/link_grammar/link-grammar/link-features.h +37 -0
- data/ext/link_grammar/link-grammar/link-features.h.in +37 -0
- data/ext/link_grammar/link-grammar/link-grammar-java.def +31 -0
- data/ext/link_grammar/link-grammar/link-grammar.def +194 -0
- data/ext/link_grammar/link-grammar/link-includes.h +465 -0
- data/ext/link_grammar/link-grammar/link-parser.c +849 -0
- data/ext/link_grammar/link-grammar/massage.c +329 -0
- data/ext/link_grammar/link-grammar/massage.h +13 -0
- data/ext/link_grammar/link-grammar/post-process.c +1113 -0
- data/ext/link_grammar/link-grammar/post-process.h +45 -0
- data/ext/link_grammar/link-grammar/pp_knowledge.c +376 -0
- data/ext/link_grammar/link-grammar/pp_knowledge.h +14 -0
- data/ext/link_grammar/link-grammar/pp_lexer.c +1920 -0
- data/ext/link_grammar/link-grammar/pp_lexer.h +19 -0
- data/ext/link_grammar/link-grammar/pp_linkset.c +158 -0
- data/ext/link_grammar/link-grammar/pp_linkset.h +20 -0
- data/ext/link_grammar/link-grammar/prefix.c +482 -0
- data/ext/link_grammar/link-grammar/prefix.h +139 -0
- data/ext/link_grammar/link-grammar/preparation.c +412 -0
- data/ext/link_grammar/link-grammar/preparation.h +20 -0
- data/ext/link_grammar/link-grammar/print-util.c +87 -0
- data/ext/link_grammar/link-grammar/print-util.h +32 -0
- data/ext/link_grammar/link-grammar/print.c +1085 -0
- data/ext/link_grammar/link-grammar/print.h +16 -0
- data/ext/link_grammar/link-grammar/prune.c +1864 -0
- data/ext/link_grammar/link-grammar/prune.h +17 -0
- data/ext/link_grammar/link-grammar/read-dict.c +1785 -0
- data/ext/link_grammar/link-grammar/read-dict.h +29 -0
- data/ext/link_grammar/link-grammar/read-regex.c +161 -0
- data/ext/link_grammar/link-grammar/read-regex.h +12 -0
- data/ext/link_grammar/link-grammar/regex-morph.c +126 -0
- data/ext/link_grammar/link-grammar/regex-morph.h +17 -0
- data/ext/link_grammar/link-grammar/resources.c +180 -0
- data/ext/link_grammar/link-grammar/resources.h +23 -0
- data/ext/link_grammar/link-grammar/sat-solver/.DS_Store +0 -0
- data/ext/link_grammar/link-grammar/sat-solver/.deps/fast-sprintf.Plo +1 -0
- data/ext/link_grammar/link-grammar/sat-solver/.deps/sat-encoder.Plo +1 -0
- data/ext/link_grammar/link-grammar/sat-solver/.deps/util.Plo +1 -0
- data/ext/link_grammar/link-grammar/sat-solver/.deps/variables.Plo +1 -0
- data/ext/link_grammar/link-grammar/sat-solver/.deps/word-tag.Plo +1 -0
- data/ext/link_grammar/link-grammar/sat-solver/Makefile +527 -0
- data/ext/link_grammar/link-grammar/sat-solver/Makefile.am +29 -0
- data/ext/link_grammar/link-grammar/sat-solver/Makefile.in +527 -0
- data/ext/link_grammar/link-grammar/sat-solver/clock.hpp +33 -0
- data/ext/link_grammar/link-grammar/sat-solver/fast-sprintf.cpp +26 -0
- data/ext/link_grammar/link-grammar/sat-solver/fast-sprintf.hpp +7 -0
- data/ext/link_grammar/link-grammar/sat-solver/guiding.hpp +244 -0
- data/ext/link_grammar/link-grammar/sat-solver/matrix-ut.hpp +79 -0
- data/ext/link_grammar/link-grammar/sat-solver/sat-encoder.cpp +2811 -0
- data/ext/link_grammar/link-grammar/sat-solver/sat-encoder.h +11 -0
- data/ext/link_grammar/link-grammar/sat-solver/sat-encoder.hpp +381 -0
- data/ext/link_grammar/link-grammar/sat-solver/trie.hpp +118 -0
- data/ext/link_grammar/link-grammar/sat-solver/util.cpp +23 -0
- data/ext/link_grammar/link-grammar/sat-solver/util.hpp +14 -0
- data/ext/link_grammar/link-grammar/sat-solver/variables.cpp +5 -0
- data/ext/link_grammar/link-grammar/sat-solver/variables.hpp +829 -0
- data/ext/link_grammar/link-grammar/sat-solver/word-tag.cpp +159 -0
- data/ext/link_grammar/link-grammar/sat-solver/word-tag.hpp +162 -0
- data/ext/link_grammar/link-grammar/spellcheck-aspell.c +148 -0
- data/ext/link_grammar/link-grammar/spellcheck-hun.c +136 -0
- data/ext/link_grammar/link-grammar/spellcheck.h +34 -0
- data/ext/link_grammar/link-grammar/string-set.c +169 -0
- data/ext/link_grammar/link-grammar/string-set.h +16 -0
- data/ext/link_grammar/link-grammar/structures.h +498 -0
- data/ext/link_grammar/link-grammar/tokenize.c +1049 -0
- data/ext/link_grammar/link-grammar/tokenize.h +15 -0
- data/ext/link_grammar/link-grammar/utilities.c +847 -0
- data/ext/link_grammar/link-grammar/utilities.h +281 -0
- data/ext/link_grammar/link-grammar/word-file.c +124 -0
- data/ext/link_grammar/link-grammar/word-file.h +15 -0
- data/ext/link_grammar/link-grammar/word-utils.c +526 -0
- data/ext/link_grammar/link-grammar/word-utils.h +152 -0
- data/ext/link_grammar/link_grammar.c +202 -0
- data/ext/link_grammar/link_grammar.h +99 -0
- data/grammar_cop.gemspec +24 -0
- data/lib/.DS_Store +0 -0
- data/lib/grammar_cop.rb +9 -0
- data/lib/grammar_cop/.DS_Store +0 -0
- data/lib/grammar_cop/dictionary.rb +19 -0
- data/lib/grammar_cop/linkage.rb +30 -0
- data/lib/grammar_cop/parse_options.rb +32 -0
- data/lib/grammar_cop/sentence.rb +36 -0
- data/lib/grammar_cop/version.rb +3 -0
- data/test/.DS_Store +0 -0
- data/test/grammar_cop_test.rb +27 -0
- metadata +407 -0
@@ -0,0 +1,24 @@
|
|
1
|
+
/*************************************************************************/
|
2
|
+
/* Copyright (c) 2004 */
|
3
|
+
/* Daniel Sleator, David Temperley, and John Lafferty */
|
4
|
+
/* All rights reserved */
|
5
|
+
/* */
|
6
|
+
/* Use of the link grammar parsing system is subject to the terms of the */
|
7
|
+
/* license set forth in the LICENSE file included with this software, */
|
8
|
+
/* and also available at http://www.link.cs.cmu.edu/link/license.html */
|
9
|
+
/* This license allows free redistribution and use in source and binary */
|
10
|
+
/* forms, with or without modification, subject to certain conditions. */
|
11
|
+
/* */
|
12
|
+
/*************************************************************************/
|
13
|
+
|
14
|
+
void init_analyze(Sentence);
|
15
|
+
void free_analyze(Sentence);
|
16
|
+
|
17
|
+
void extract_thin_linkage(Sentence, Parse_Options, Linkage);
|
18
|
+
void extract_fat_linkage (Sentence, Parse_Options, Linkage);
|
19
|
+
Linkage_info analyze_fat_linkage (Sentence, Parse_Options, int pass);
|
20
|
+
Linkage_info analyze_thin_linkage(Sentence, Parse_Options, int pass);
|
21
|
+
void free_DIS_tree(DIS_node *);
|
22
|
+
|
23
|
+
void zero_sublinkage(Sublinkage *s);
|
24
|
+
|
@@ -0,0 +1,1603 @@
|
|
1
|
+
/*************************************************************************/
|
2
|
+
/* Copyright (c) 2004 */
|
3
|
+
/* Daniel Sleator, David Temperley, and John Lafferty */
|
4
|
+
/* All rights reserved */
|
5
|
+
/* */
|
6
|
+
/* Use of the link grammar parsing system is subject to the terms of the */
|
7
|
+
/* license set forth in the LICENSE file included with this software, */
|
8
|
+
/* and also available at http://www.link.cs.cmu.edu/link/license.html */
|
9
|
+
/* This license allows free redistribution and use in source and binary */
|
10
|
+
/* forms, with or without modification, subject to certain conditions. */
|
11
|
+
/* */
|
12
|
+
/*************************************************************************/
|
13
|
+
|
14
|
+
#include "api.h"
|
15
|
+
#include "disjunct-utils.h"
|
16
|
+
|
17
|
+
/*
|
18
|
+
Notes about AND
|
19
|
+
|
20
|
+
A large fraction of the code of this parser seems to deal with handling
|
21
|
+
conjunctions. This comment (combined with reading the paper) should
|
22
|
+
give an idea of how it works.
|
23
|
+
|
24
|
+
First of all, we need a more detailed discussion of strings, what they
|
25
|
+
match, etc. (This entire discussion ignores the labels, which are
|
26
|
+
semantically the same as the leading upper case letters of the
|
27
|
+
connector.)
|
28
|
+
|
29
|
+
We'll deal with infinite strings from an alphabet of three types of
|
30
|
+
characters: "*". "^" and ordinary characters (denoted "a" and "b").
|
31
|
+
(The end of a string should be thought of as an infinite sequence of
|
32
|
+
"*"s).
|
33
|
+
|
34
|
+
Let match(s) be the set of strings that will match the string s. This
|
35
|
+
is defined as follows. A string t is in match(s) if (1) its leading
|
36
|
+
upper case letters exactly match those of s. (2) traversing through
|
37
|
+
both strings, from left to right in step, no missmatch is found
|
38
|
+
between corresponding letters. A missmatch is a pair of differing
|
39
|
+
ordinary characters, or a "^" and any ordinary letter or two "^"s.
|
40
|
+
In other words, a match is exactly a "*" and anything, or two
|
41
|
+
identical ordinary letters.
|
42
|
+
|
43
|
+
Alternative definition of the set match(s):
|
44
|
+
{t | t is obtained from s by replacing each "^" and any other characters
|
45
|
+
by "*"s, and replacing any original "*" in s by any other character
|
46
|
+
(or "^").}
|
47
|
+
|
48
|
+
Theorem: if t in match(s) then s in match(t).
|
49
|
+
|
50
|
+
It is also a theorem that given any two strings s and t, there exists a
|
51
|
+
unique new string u with the property that:
|
52
|
+
|
53
|
+
match(u) = match(s) intersect match(t)
|
54
|
+
|
55
|
+
This string is called the GCD of s and t. Here are some examples.
|
56
|
+
|
57
|
+
GCD(N*a,Nb) = Nba
|
58
|
+
GCD(Na, Nb) = N^
|
59
|
+
GCD(Nab,Nb) = N^b
|
60
|
+
GCD(N^,N*a) = N^a
|
61
|
+
GCD(N^, N) = N^
|
62
|
+
GCD(N^^,N^) = N^^
|
63
|
+
|
64
|
+
We need an algorithm for computing the GCD of two strings. Here is
|
65
|
+
one.
|
66
|
+
|
67
|
+
First get by the upper case letters (which must be equal, otherwise
|
68
|
+
there is no intersection), issuing them. Traverse the rest of the
|
69
|
+
characters of s and t in lockstep until there is nothing left but
|
70
|
+
"*"s. If the two characters are:
|
71
|
+
|
72
|
+
"a" and "a", issue "a"
|
73
|
+
"a" and "b", issue "^"
|
74
|
+
"a" and "*", issue "a"
|
75
|
+
"*" and "*", issue "*"
|
76
|
+
"*" and "^", issue "^"
|
77
|
+
"a" and "^", issue "^"
|
78
|
+
"^" and "^", issue "^"
|
79
|
+
|
80
|
+
A simple case analysis suffices to show that any string that matches
|
81
|
+
the right side, must match both of the left sides, and any string not
|
82
|
+
matching the right side must not match at least one of the left sides.
|
83
|
+
|
84
|
+
This proves that the GCD operator is associative and commutative.
|
85
|
+
(There must be a name for a mathematical structure with these properties.)
|
86
|
+
|
87
|
+
To elaborate further on this theory, define the notion of two strings
|
88
|
+
matching in the dual sense as follows: s and t dual-match if
|
89
|
+
match(s) is contained in match(t) or vice versa---
|
90
|
+
|
91
|
+
Full development of this theory could lead to a more efficient
|
92
|
+
algorithm for this problem. I'll defer this until such time as it
|
93
|
+
appears necessary.
|
94
|
+
|
95
|
+
|
96
|
+
We need a data structure that stores a set of fat links. Each fat
|
97
|
+
link has a number (called its label). The fat link operates in liu of
|
98
|
+
a collection of links. The particular stuff it is a substitute for is
|
99
|
+
defined by a disjunct. This disjunct is stored in the data structure.
|
100
|
+
|
101
|
+
The type of a disjunct is defined by the sequence of connector types
|
102
|
+
(defined by their upper case letters) that comprises it. Each entry
|
103
|
+
of the label_table[] points to a list of disjuncts that have the same
|
104
|
+
type (a hash table is uses so that, given a disjunct, we can efficiently
|
105
|
+
compute the element of the label table in which it belongs).
|
106
|
+
|
107
|
+
We begin by loading up the label table with all of the possible
|
108
|
+
fat links that occur through the words of the sentence. These are
|
109
|
+
obtained by taking every sub-range of the connectors of each disjunct
|
110
|
+
(containing the center). We also compute the closure (under the GCD
|
111
|
+
operator) of these disjuncts and store also store these in the
|
112
|
+
label_table. Each disjunct in this table has a string which represents
|
113
|
+
the subscripts of all of its connectors (and their multi-connector bits).
|
114
|
+
|
115
|
+
It is possible to generate a fat connector for any one of the
|
116
|
+
disjuncts in the label_table. This connector's label field is given
|
117
|
+
the label from the disjunct from which it arose. It's string field
|
118
|
+
is taken from the string of the disjunct (mentioned above). It will be
|
119
|
+
given a priority with a value of UP_priority or DOWN_priority (depending
|
120
|
+
on how it will be used). A connector of UP_priority can match one of
|
121
|
+
DOWN_priority, but neither of these can match any other priority.
|
122
|
+
(Of course, a fat connector can match only another fat connector with
|
123
|
+
the same label.)
|
124
|
+
|
125
|
+
The paper describes in some detail how disjuncts are given to words
|
126
|
+
and to "and" and ",", etc. Each word in the sentence gets many more
|
127
|
+
new disjuncts. For each contiguous set of connectors containing (or
|
128
|
+
adjacent to) the center of the disjunct, we generate a fat link, and
|
129
|
+
replace these connector in the word by a fat link. (Actually we do
|
130
|
+
this twice. Once pointing to the right, once to the left.) These fat
|
131
|
+
links have priority UP_priority.
|
132
|
+
|
133
|
+
What do we generate for ","? For each type of fat link (each label)
|
134
|
+
we make a disjunct that has two down connectors (to the right and left)
|
135
|
+
and one up connector (to the right). There will be a unique way of
|
136
|
+
hooking together a comma-separated and-list.
|
137
|
+
|
138
|
+
The disjuncts on "and" are more complicated. Here we have to do just what
|
139
|
+
we did for comma (but also include the up link to the left), then
|
140
|
+
we also have to allow the process to terminate. So, there is a disjunct
|
141
|
+
with two down fat links, and between them are the original thin links.
|
142
|
+
These are said to "blossom" out. However, this is not all that is
|
143
|
+
necessary. It's possible for an and-list to be part of another and list
|
144
|
+
with a different labeled fat connector. To make this possible, we
|
145
|
+
regroup the just blossomed disjuncts (in all possible ways about the center)
|
146
|
+
and install them as fat links. If this sounds like a lot of disjuncts --
|
147
|
+
it is! The program is currently fairly slow on long sentence with and.
|
148
|
+
|
149
|
+
It is slightly non-obvious that the fat-links in a linkage constructed
|
150
|
+
from disjuncts defined in this way form a binary tree. Naturally,
|
151
|
+
connectors with UP_priority point up the tree, and those with DOWN_priority
|
152
|
+
point down the tree.
|
153
|
+
|
154
|
+
Think of the string x on the connector as representing a set X of strings.
|
155
|
+
X = match(x). So, for example, if x="S^" then match(x) = {"S", "S*a",
|
156
|
+
"S*b", etc}. The matching rules for UP and DOWN priority connectors
|
157
|
+
are such that as you go up (the tree of ands) the X sets get no larger.
|
158
|
+
So, for example, a "Sb" pointing up can match an "S^" pointing down.
|
159
|
+
(Because more stuff can match "Sb" than can match "S^".)
|
160
|
+
This guarantees that whatever connector ultimately gets used after the
|
161
|
+
fat connector blossoms out (see below), it is a powerful enough connector
|
162
|
+
to be able to match to any of the connectors associated with it.
|
163
|
+
|
164
|
+
One problem with the scheme just descibed is that it sometimes generates
|
165
|
+
essentially the same linkage several times. This happens if there is
|
166
|
+
a gap in the connective power, and the mismatch can be moved around in
|
167
|
+
different ways. Here is an example of how this happens.
|
168
|
+
|
169
|
+
(Left is DOWN, right is UP)
|
170
|
+
|
171
|
+
Sa <---> S^ <---> S or Sa <---> Sa <---> S
|
172
|
+
fat thin fat thin
|
173
|
+
|
174
|
+
Here two of the disjunct types are given by "S^" and "Sa". Notice that
|
175
|
+
the criterion of shrinking the matching set is satisfied by the the fat
|
176
|
+
link (traversing from left to right). How do I eliminate one of these?
|
177
|
+
|
178
|
+
I use the technique of canonization. I generate all the linkages. There
|
179
|
+
is then a procedure that can check to see of a linkage is canonical.
|
180
|
+
If it is, it's used, otherwise it's ignored. It's claimed that exactly
|
181
|
+
one canonical one of each equivalence class will be generated.
|
182
|
+
We basically insist that the intermediate fat disjuncts (ones that
|
183
|
+
have a fat link pointing down) are all minimal -- that is, that they
|
184
|
+
cannot be replaced by by another (with a strictly) smaller match set.
|
185
|
+
If one is not minimal, then the linkage is rejected.
|
186
|
+
|
187
|
+
Here's a proof that this is correct. Consider the set of equivalent
|
188
|
+
linkages that are generated. These Pick a disjunct that is the root of
|
189
|
+
its tree. Consider the set of all disjuncts which occur in that positon
|
190
|
+
among the equivalent linkages. The GCD of all of these can fit in that
|
191
|
+
position (it matches down the tree, since its match set has gotten
|
192
|
+
smaller, and it also matches to the THIN links.) Since the GCD is put
|
193
|
+
on "and" this particular one will be generated. Therefore rejecting
|
194
|
+
a linkage in which a root fat disjunct can be replaced by a smaller one
|
195
|
+
is ok (since the smaller one will be generated separately). What about
|
196
|
+
a fat disjunct that is not the root. We consider the set of linkages in
|
197
|
+
which the root is minimal (the ones for which it's not have already been
|
198
|
+
eliminated). Now, consider one of the children of the root in precisely
|
199
|
+
the way we just considered the root. The same argument holds. The only
|
200
|
+
difference is that the root node gives another constraint on how small
|
201
|
+
you can make the disjunct -- so, within these constraints, if we can go
|
202
|
+
smaller, we reject.
|
203
|
+
|
204
|
+
The code to do all of this is fairly ugly, but I think it works.
|
205
|
+
|
206
|
+
|
207
|
+
Problems with this stuff:
|
208
|
+
|
209
|
+
1) There is obviously a combinatorial explosion that takes place.
|
210
|
+
As the number of disjuncts (and the number of their subscripts
|
211
|
+
increase) the number of disjuncts that get put onto "and" will
|
212
|
+
increase tremendously. When we made the transcript for the tech
|
213
|
+
report (Around August 1991) most of the sentence were processed
|
214
|
+
in well under 10 seconds. Now (Jan 1992), some of these sentences
|
215
|
+
take ten times longer. As of this writing I don't really know the
|
216
|
+
reason, other than just the fact that the dictionary entries are
|
217
|
+
more complex than they used to be. The number of linkages has also
|
218
|
+
increased significantly.
|
219
|
+
|
220
|
+
2) Each element of an and list must be attached through only one word.
|
221
|
+
This disallows "there is time enough and space enough for both of us",
|
222
|
+
and many other reasonable sounding things. The combinatorial
|
223
|
+
explosion that would occur if you allowed two different connection
|
224
|
+
points would be tremendous, and the number of solutions would also
|
225
|
+
probably go up by another order of magnitude. Perhaps if there
|
226
|
+
were strong constraints on the type of connectors in which this
|
227
|
+
would be allowed, then this would be a conceivable prospect.
|
228
|
+
|
229
|
+
3) A multi-connector must be either all "outside" or all "inside" the and.
|
230
|
+
For example, "the big black dog and cat ran" has only two ways to
|
231
|
+
linkages (instead of three).
|
232
|
+
|
233
|
+
Possible bug: It seems that the following two linkages should be the
|
234
|
+
same under the canonical linkage test. Could this have to do with the
|
235
|
+
pluralization system?
|
236
|
+
|
237
|
+
> I am big and the bike and the car were broken
|
238
|
+
Accepted (4 linkages, 4 with no P.P. violations) at stage 1
|
239
|
+
Linkage 1, cost vector = (0, 0, 18)
|
240
|
+
|
241
|
+
+------Spx-----+
|
242
|
+
+-----CC-----+------Wd------+-d^^*i^-+ |
|
243
|
+
+-Wd-+Spi+-Pa+ | +--Ds-+d^^*+ +-Ds-+ +--Pv-+
|
244
|
+
| | | | | | | | | | | |
|
245
|
+
///// I.p am big.a and the bike.n and the car.n were broken
|
246
|
+
|
247
|
+
///// RW <---RW----> RW /////
|
248
|
+
///// Wd <---Wd----> Wd I.p
|
249
|
+
I.p CC <---CC----> CC and
|
250
|
+
I.p Sp*i <---Spii--> Spi am
|
251
|
+
am Pa <---Pa----> Pa big.a
|
252
|
+
and Wd <---Wd----> Wd and
|
253
|
+
bike.n d^s** 6<---d^^*i-> d^^*i 6 and
|
254
|
+
the D <---Ds----> Ds bike.n
|
255
|
+
and Sp <---Spx---> Spx were
|
256
|
+
and d^^*i 6<---d^^*i-> d^s** 6 car.n
|
257
|
+
the D <---Ds----> Ds car.n
|
258
|
+
were Pv <---Pv----> Pv broken
|
259
|
+
|
260
|
+
(press return for another)
|
261
|
+
>
|
262
|
+
Linkage 2, cost vector = (0, 0, 18)
|
263
|
+
|
264
|
+
+------Spx-----+
|
265
|
+
+-----CC-----+------Wd------+-d^s**^-+ |
|
266
|
+
+-Wd-+Spi+-Pa+ | +--Ds-+d^s*+ +-Ds-+ +--Pv-+
|
267
|
+
| | | | | | | | | | | |
|
268
|
+
///// I.p am big.a and the bike.n and the car.n were broken
|
269
|
+
|
270
|
+
///// RW <---RW----> RW /////
|
271
|
+
///// Wd <---Wd----> Wd I.p
|
272
|
+
I.p CC <---CC----> CC and
|
273
|
+
I.p Sp*i <---Spii--> Spi am
|
274
|
+
am Pa <---Pa----> Pa big.a
|
275
|
+
and Wd <---Wd----> Wd and
|
276
|
+
bike.n d^s** 6<---d^s**-> d^s** 6 and
|
277
|
+
the D <---Ds----> Ds bike.n
|
278
|
+
and Sp <---Spx---> Spx were
|
279
|
+
and d^s** 6<---d^s**-> d^s** 6 car.n
|
280
|
+
the D <---Ds----> Ds car.n
|
281
|
+
were Pv <---Pv----> Pv broken
|
282
|
+
|
283
|
+
*/
|
284
|
+
|
285
|
+
static void init_LT(Sentence sent)
|
286
|
+
{
|
287
|
+
sent->and_data.LT_bound = 20;
|
288
|
+
sent->and_data.LT_size = 0;
|
289
|
+
sent->and_data.label_table =
|
290
|
+
(Disjunct **) xalloc(sent->and_data.LT_bound * sizeof(Disjunct *));
|
291
|
+
}
|
292
|
+
|
293
|
+
static void grow_LT(Sentence sent)
|
294
|
+
{
|
295
|
+
size_t oldsize = sent->and_data.LT_bound * sizeof(Disjunct *);
|
296
|
+
sent->and_data.LT_bound = (3*sent->and_data.LT_bound)/2;
|
297
|
+
sent->and_data.label_table =
|
298
|
+
(Disjunct **) xrealloc(sent->and_data.label_table,
|
299
|
+
oldsize,
|
300
|
+
sent->and_data.LT_bound * sizeof(Disjunct *));
|
301
|
+
}
|
302
|
+
|
303
|
+
static void init_HT(Sentence sent)
|
304
|
+
{
|
305
|
+
memset(sent->and_data.hash_table, 0, HT_SIZE*sizeof(Label_node *));
|
306
|
+
}
|
307
|
+
|
308
|
+
static void free_HT(Sentence sent)
|
309
|
+
{
|
310
|
+
int i;
|
311
|
+
Label_node * la, * la1;
|
312
|
+
for (i=0; i<HT_SIZE; i++) {
|
313
|
+
for (la=sent->and_data.hash_table[i]; la != NULL; la = la1) {
|
314
|
+
la1 = la->next;
|
315
|
+
xfree((char *)la, sizeof(Label_node));
|
316
|
+
}
|
317
|
+
sent->and_data.hash_table[i] = NULL;
|
318
|
+
}
|
319
|
+
}
|
320
|
+
|
321
|
+
static void free_LT(Sentence sent)
|
322
|
+
{
|
323
|
+
int i;
|
324
|
+
for (i=0; i<sent->and_data.LT_size; i++) {
|
325
|
+
free_disjuncts(sent->and_data.label_table[i]);
|
326
|
+
}
|
327
|
+
xfree((char *) sent->and_data.label_table,
|
328
|
+
sent->and_data.LT_bound * sizeof(Disjunct*));
|
329
|
+
sent->and_data.LT_bound = 0;
|
330
|
+
sent->and_data.LT_size = 0;
|
331
|
+
sent->and_data.label_table = NULL;
|
332
|
+
}
|
333
|
+
|
334
|
+
void free_AND_tables(Sentence sent)
|
335
|
+
{
|
336
|
+
free_LT(sent);
|
337
|
+
free_HT(sent);
|
338
|
+
}
|
339
|
+
|
340
|
+
void initialize_conjunction_tables(Sentence sent)
|
341
|
+
{
|
342
|
+
int i;
|
343
|
+
sent->and_data.LT_bound = 0;
|
344
|
+
sent->and_data.LT_size = 0;
|
345
|
+
sent->and_data.label_table = NULL;
|
346
|
+
for (i=0; i<HT_SIZE; i++) {
|
347
|
+
sent->and_data.hash_table[i] = NULL;
|
348
|
+
}
|
349
|
+
}
|
350
|
+
|
351
|
+
/**
|
352
|
+
* This is a hash function for disjuncts
|
353
|
+
*/
|
354
|
+
static inline int and_hash_disjunct(Disjunct *d)
|
355
|
+
{
|
356
|
+
unsigned int i;
|
357
|
+
Connector *e;
|
358
|
+
i = 0;
|
359
|
+
for (e = d->left ; e != NULL; e = e->next) {
|
360
|
+
i += connector_hash(e);
|
361
|
+
}
|
362
|
+
i += (i<<5);
|
363
|
+
for (e = d->right ; e != NULL; e = e->next) {
|
364
|
+
i += connector_hash(e);
|
365
|
+
}
|
366
|
+
return (i & (HT_SIZE-1));
|
367
|
+
}
|
368
|
+
|
369
|
+
/**
|
370
|
+
* Returns TRUE if the disjunct is appropriate to be made into fat links.
|
371
|
+
* Check here that the connectors are from some small set.
|
372
|
+
* This will disallow, for example "the and their dog ran".
|
373
|
+
*/
|
374
|
+
static int is_appropriate(Sentence sent, Disjunct * d)
|
375
|
+
{
|
376
|
+
Connector * c;
|
377
|
+
|
378
|
+
if (sent->dict->andable_connector_set == NULL) return TRUE;
|
379
|
+
/* if no set, then everything is considered andable */
|
380
|
+
for (c = d->right; c!=NULL; c=c->next) {
|
381
|
+
if (!match_in_connector_set(sent, sent->dict->andable_connector_set, c, '+')) return FALSE;
|
382
|
+
}
|
383
|
+
for (c = d->left; c!=NULL; c=c->next) {
|
384
|
+
if (!match_in_connector_set(sent, sent->dict->andable_connector_set, c, '-')) return FALSE;
|
385
|
+
}
|
386
|
+
return TRUE;
|
387
|
+
}
|
388
|
+
|
389
|
+
/**
|
390
|
+
* Two connectors are said to be of the same type if they have
|
391
|
+
* the same label, and the initial upper case letters of their
|
392
|
+
* strings match.
|
393
|
+
*/
|
394
|
+
static int connector_types_equal(Connector * c1, Connector * c2)
|
395
|
+
{
|
396
|
+
if (c1->label != c2->label) return FALSE;
|
397
|
+
return utf8_upper_match(c1->string, c2->string);
|
398
|
+
}
|
399
|
+
|
400
|
+
/**
|
401
|
+
* Two disjuncts are said to be the same type if they're the same
|
402
|
+
* ignoring the multi fields, the priority fields, and the subscripts
|
403
|
+
* of the connectors (and the string field of the disjunct of course).
|
404
|
+
* Disjuncts of the same type are located in the same label_table list.
|
405
|
+
*
|
406
|
+
* This returns TRUE if they are of the same type.
|
407
|
+
*/
|
408
|
+
static int disjunct_types_equal(Disjunct * d1, Disjunct * d2)
|
409
|
+
{
|
410
|
+
Connector *e1, *e2;
|
411
|
+
|
412
|
+
e1 = d1->left;
|
413
|
+
e2 = d2->left;
|
414
|
+
while((e1!=NULL) && (e2!=NULL)) {
|
415
|
+
if (!connector_types_equal(e1,e2)) break;
|
416
|
+
e1 = e1->next;
|
417
|
+
e2 = e2->next;
|
418
|
+
}
|
419
|
+
if ((e1!=NULL) || (e2!=NULL)) return FALSE;
|
420
|
+
e1 = d1->right;
|
421
|
+
e2 = d2->right;
|
422
|
+
while((e1!=NULL) && (e2!=NULL)) {
|
423
|
+
if (!connector_types_equal(e1,e2)) break;
|
424
|
+
e1 = e1->next;
|
425
|
+
e2 = e2->next;
|
426
|
+
}
|
427
|
+
if ((e1!=NULL) || (e2!=NULL)) return FALSE;
|
428
|
+
return TRUE;
|
429
|
+
}
|
430
|
+
|
431
|
+
/**
|
432
|
+
* This returns a string that is the the GCD of the two given strings.
|
433
|
+
* If the GCD is equal to one of them, a pointer to it is returned.
|
434
|
+
* Otherwise a new string for the GCD is xalloced and put on the
|
435
|
+
* "free later" list.
|
436
|
+
*/
|
437
|
+
const char * intersect_strings(Sentence sent, const char * s, const char * t)
|
438
|
+
{
|
439
|
+
int len, i, j, d;
|
440
|
+
const char *w, *s0;
|
441
|
+
char u0[MAX_TOKEN_LENGTH]; /* Links are *always* less than 10 chars long */
|
442
|
+
char *u;
|
443
|
+
if (strcmp(s,t)==0) return s; /* would work without this */
|
444
|
+
i = strlen(s);
|
445
|
+
j = strlen(t);
|
446
|
+
if (j > i) {
|
447
|
+
w = s; s = t; t = w;
|
448
|
+
len = j;
|
449
|
+
} else {
|
450
|
+
len = i;
|
451
|
+
}
|
452
|
+
/* s is now the longer (at least not the shorter) string */
|
453
|
+
/* and len is its length */
|
454
|
+
u = u0;
|
455
|
+
d = 0;
|
456
|
+
s0 = s;
|
457
|
+
while (*t != '\0') {
|
458
|
+
if ((*s == *t) || (*t == '*')) {
|
459
|
+
*u = *s;
|
460
|
+
} else {
|
461
|
+
d++;
|
462
|
+
if (*s == '*') *u = *t;
|
463
|
+
else *u = '^';
|
464
|
+
}
|
465
|
+
s++; t++; u++;
|
466
|
+
}
|
467
|
+
if (d==0) {
|
468
|
+
return s0;
|
469
|
+
} else {
|
470
|
+
strcpy(u, s); /* get the remainder of s */
|
471
|
+
return string_set_add(u0, sent->string_set);
|
472
|
+
}
|
473
|
+
}
|
474
|
+
|
475
|
+
/**
|
476
|
+
* Two connectors are said to be equal if they are of the same type
|
477
|
+
* (defined above), they have the same multi field, and they have
|
478
|
+
* exactly the same connectors (including lower case chars).
|
479
|
+
* (priorities ignored).
|
480
|
+
*/
|
481
|
+
static int connectors_equal_AND(Connector *c1, Connector *c2)
|
482
|
+
{
|
483
|
+
return (c1->label == c2->label) &&
|
484
|
+
(c1->multi == c2->multi) &&
|
485
|
+
(strcmp(c1->string, c2->string) == 0);
|
486
|
+
}
|
487
|
+
|
488
|
+
/**
|
489
|
+
* Return true if the disjuncts are equal (ignoring priority fields)
|
490
|
+
* and the string of the disjunct.
|
491
|
+
*/
|
492
|
+
static int disjuncts_equal_AND(Sentence sent, Disjunct * d1, Disjunct * d2)
|
493
|
+
{
|
494
|
+
Connector *e1, *e2;
|
495
|
+
sent->and_data.STAT_calls_to_equality_test++;
|
496
|
+
e1 = d1->left;
|
497
|
+
e2 = d2->left;
|
498
|
+
while((e1!=NULL) && (e2!=NULL)) {
|
499
|
+
if (!connectors_equal_AND(e1, e2)) break;
|
500
|
+
e1 = e1->next;
|
501
|
+
e2 = e2->next;
|
502
|
+
}
|
503
|
+
if ((e1!=NULL) || (e2!=NULL)) return FALSE;
|
504
|
+
e1 = d1->right;
|
505
|
+
e2 = d2->right;
|
506
|
+
while((e1!=NULL) && (e2!=NULL)) {
|
507
|
+
if (!connectors_equal_AND(e1, e2)) break;
|
508
|
+
e1 = e1->next;
|
509
|
+
e2 = e2->next;
|
510
|
+
}
|
511
|
+
if ((e1!=NULL) || (e2!=NULL)) return FALSE;
|
512
|
+
return TRUE;
|
513
|
+
}
|
514
|
+
|
515
|
+
/**
|
516
|
+
* Create a new disjunct that is the GCD of d1 and d2.
|
517
|
+
* It assumes that the disjuncts are of the same type, so the
|
518
|
+
* GCD will not be empty.
|
519
|
+
*/
|
520
|
+
static Disjunct * intersect_disjuncts(Sentence sent, Disjunct * d1, Disjunct * d2)
|
521
|
+
{
|
522
|
+
Disjunct * d;
|
523
|
+
Connector *c1, *c2, *c;
|
524
|
+
d = copy_disjunct(d1);
|
525
|
+
c = d->left;
|
526
|
+
c1 = d1->left;
|
527
|
+
c2 = d2->left;
|
528
|
+
while (c1!=NULL) {
|
529
|
+
connector_set_string (c, intersect_strings(sent, c1->string, c2->string));
|
530
|
+
c->multi = (c1->multi) && (c2->multi);
|
531
|
+
c = c->next; c1 = c1->next; c2 = c2->next;
|
532
|
+
}
|
533
|
+
c = d->right;
|
534
|
+
c1 = d1->right;
|
535
|
+
c2 = d2->right;
|
536
|
+
while (c1!=NULL) {
|
537
|
+
connector_set_string (c, intersect_strings(sent, c1->string, c2->string));
|
538
|
+
c->multi = (c1->multi) && (c2->multi);
|
539
|
+
c = c->next; c1 = c1->next; c2 = c2->next;
|
540
|
+
}
|
541
|
+
return d;
|
542
|
+
}
|
543
|
+
|
544
|
+
/**
|
545
|
+
* (1) look for the given disjunct in the table structures
|
546
|
+
* if it's already in the table structures, do nothing
|
547
|
+
* (2) otherwise make a copy of it, and put it into the table structures
|
548
|
+
* (3) also put all of the GCDs of this disjunct with all of the
|
549
|
+
* other matching disjuncts into the table.
|
550
|
+
*
|
551
|
+
* The costs are set to zero.
|
552
|
+
* Note that this has no effect on disjunct d.
|
553
|
+
*/
|
554
|
+
static void put_disjunct_into_table(Sentence sent, Disjunct *d)
|
555
|
+
{
|
556
|
+
Disjunct *d1=NULL, *d2, *di, *d_copy;
|
557
|
+
Label_node * lp;
|
558
|
+
int h, k;
|
559
|
+
|
560
|
+
h = and_hash_disjunct(d);
|
561
|
+
|
562
|
+
for (lp = sent->and_data.hash_table[h]; lp != NULL; lp = lp->next)
|
563
|
+
{
|
564
|
+
d1 = sent->and_data.label_table[lp->label];
|
565
|
+
if (disjunct_types_equal(d,d1)) break;
|
566
|
+
}
|
567
|
+
if (lp != NULL)
|
568
|
+
{
|
569
|
+
/* there is already a label for disjuncts of this type */
|
570
|
+
/* d1 points to the list of disjuncts of this type already there */
|
571
|
+
while(d1 != NULL)
|
572
|
+
{
|
573
|
+
if (disjuncts_equal_AND(sent, d1, d)) return;
|
574
|
+
d1 = d1->next;
|
575
|
+
}
|
576
|
+
/* now we must put the d disjunct in there, and all of the GCDs of
|
577
|
+
it with the ones already there.
|
578
|
+
|
579
|
+
This is done as follows. We scan through the list of disjuncts
|
580
|
+
computing the gcd of the new one with each of the others, putting
|
581
|
+
the resulting disjuncts onto another list rooted at d2.
|
582
|
+
Now insert d into the the list already there. Now for each
|
583
|
+
one on the d2 list, put it in if it isn't already there.
|
584
|
+
|
585
|
+
Here we're making use of the following theorem: Given a
|
586
|
+
collection of sets s1, s2 ... sn closed under intersection,
|
587
|
+
to if we add a new set s to the collection and also add
|
588
|
+
all the intersections between s and s1...sn to the collection,
|
589
|
+
then the collection is still closed under intersection.
|
590
|
+
|
591
|
+
Use a Venn diagram to prove this theorem.
|
592
|
+
|
593
|
+
*/
|
594
|
+
d_copy = copy_disjunct(d);
|
595
|
+
d_copy->cost = 0;
|
596
|
+
k = lp->label;
|
597
|
+
d2 = NULL;
|
598
|
+
for (d1=sent->and_data.label_table[k]; d1!=NULL; d1 = d1->next) {
|
599
|
+
di = intersect_disjuncts(sent, d_copy, d1);
|
600
|
+
di->next = d2;
|
601
|
+
d2 = di;
|
602
|
+
}
|
603
|
+
d_copy->next = sent->and_data.label_table[k];
|
604
|
+
sent->and_data.label_table[k] = d_copy;
|
605
|
+
for (;d2 != NULL; d2 = di) {
|
606
|
+
di = d2->next;
|
607
|
+
for (d1 = sent->and_data.label_table[k]; d1 != NULL; d1 = d1->next) {
|
608
|
+
if (disjuncts_equal_AND(sent, d1, d2)) break;
|
609
|
+
}
|
610
|
+
if (d1 == NULL) {
|
611
|
+
sent->and_data.STAT_N_disjuncts++;
|
612
|
+
d2->next = sent->and_data.label_table[k];
|
613
|
+
sent->and_data.label_table[k] = d2;
|
614
|
+
} else {
|
615
|
+
d2->next = NULL;
|
616
|
+
free_disjuncts(d2);
|
617
|
+
}
|
618
|
+
}
|
619
|
+
} else {
|
620
|
+
/* create a new label for disjuncts of this type */
|
621
|
+
d_copy = copy_disjunct(d);
|
622
|
+
d_copy->cost = 0;
|
623
|
+
d_copy->next = NULL;
|
624
|
+
if (sent->and_data.LT_size == sent->and_data.LT_bound) grow_LT(sent);
|
625
|
+
lp = (Label_node *) xalloc(sizeof(Label_node));
|
626
|
+
lp->next = sent->and_data.hash_table[h];
|
627
|
+
sent->and_data.hash_table[h] = lp;
|
628
|
+
lp->label = sent->and_data.LT_size;
|
629
|
+
sent->and_data.label_table[sent->and_data.LT_size] = d_copy;
|
630
|
+
sent->and_data.LT_size++;
|
631
|
+
sent->and_data.STAT_N_disjuncts++;
|
632
|
+
}
|
633
|
+
}
|
634
|
+
|
635
|
+
/**
|
636
|
+
* A sub disjuct of d is any disjunct obtained by killing the tail
|
637
|
+
* of either connector list at any point.
|
638
|
+
* Here we go through each sub-disjunct of d, and put it into our
|
639
|
+
* table data structure.
|
640
|
+
*
|
641
|
+
* The function has no side effects on d.
|
642
|
+
*/
|
643
|
+
static void extract_all_fat_links(Sentence sent, Disjunct * d)
|
644
|
+
{
|
645
|
+
Connector * cl, * cr, *tl, *tr;
|
646
|
+
tl = d->left;
|
647
|
+
d->left = NULL;
|
648
|
+
for (cr = d->right; cr!=NULL; cr = cr->next) {
|
649
|
+
tr = cr->next;
|
650
|
+
cr->next = NULL;
|
651
|
+
if (is_appropriate(sent, d)) put_disjunct_into_table(sent, d);
|
652
|
+
cr->next = tr;
|
653
|
+
}
|
654
|
+
d->left = tl;
|
655
|
+
|
656
|
+
tr = d->right;
|
657
|
+
d->right = NULL;
|
658
|
+
for (cl = d->left; cl!=NULL; cl = cl->next) {
|
659
|
+
tl = cl->next;
|
660
|
+
cl->next = NULL;
|
661
|
+
if (is_appropriate(sent, d)) put_disjunct_into_table(sent, d);
|
662
|
+
cl->next = tl;
|
663
|
+
}
|
664
|
+
d->right = tr;
|
665
|
+
|
666
|
+
for (cl = d->left; cl!=NULL; cl = cl->next) {
|
667
|
+
for (cr = d->right; cr!=NULL; cr = cr->next) {
|
668
|
+
tl = cl->next;
|
669
|
+
tr = cr->next;
|
670
|
+
cl->next = cr->next = NULL;
|
671
|
+
|
672
|
+
if (is_appropriate(sent, d)) put_disjunct_into_table(sent, d);
|
673
|
+
|
674
|
+
cl->next = tl;
|
675
|
+
cr->next = tr;
|
676
|
+
}
|
677
|
+
}
|
678
|
+
}
|
679
|
+
|
680
|
+
/**
|
681
|
+
* put the next len characters from c->string (skipping upper
|
682
|
+
* case ones) into s. If there are fewer than this, pad with '*'s.
|
683
|
+
* Then put in a character for the multi match bit of c.
|
684
|
+
* Then put in a '\0', and return a pointer to this place.
|
685
|
+
*/
|
686
|
+
static char * stick_in_one_connector(char *s, Connector *c, int len)
|
687
|
+
{
|
688
|
+
const char * t;
|
689
|
+
|
690
|
+
t = skip_utf8_upper(c->string);
|
691
|
+
|
692
|
+
while (*t != '\0') {
|
693
|
+
*s++ = *t++;
|
694
|
+
len--;
|
695
|
+
}
|
696
|
+
while (len > 0) {
|
697
|
+
*s++ = '*';
|
698
|
+
len--;
|
699
|
+
}
|
700
|
+
if (c->multi) *s++ = '*'; else *s++ = '^'; /* check this sometime */
|
701
|
+
*s = '\0';
|
702
|
+
return s;
|
703
|
+
}
|
704
|
+
|
705
|
+
/**
|
706
|
+
* This takes a label k, modifies the list of disjuncts with that
|
707
|
+
* label. For each such disjunct, it computes the string that
|
708
|
+
* will be used in the fat connector that represents it.
|
709
|
+
*
|
710
|
+
* The only hard part is finding the length of each of the strings
|
711
|
+
* so that "*" can be put in. A better explanation will have to wait.
|
712
|
+
*/
|
713
|
+
static void compute_matchers_for_a_label(Sentence sent, int k)
|
714
|
+
{
|
715
|
+
char buff[2*MAX_WORD];
|
716
|
+
int lengths[MAX_LINKS];
|
717
|
+
int N_connectors, i, j;
|
718
|
+
Connector * c;
|
719
|
+
Disjunct * d;
|
720
|
+
const char *cs;
|
721
|
+
char *s;
|
722
|
+
|
723
|
+
d = sent->and_data.label_table[k];
|
724
|
+
|
725
|
+
N_connectors = 0;
|
726
|
+
for (c=d->left; c != NULL; c = c->next) N_connectors ++;
|
727
|
+
for (c=d->right; c != NULL; c = c->next) N_connectors ++;
|
728
|
+
|
729
|
+
for (i=0; i<N_connectors; i++) lengths[i] = 0;
|
730
|
+
while(d != NULL) {
|
731
|
+
i = 0;
|
732
|
+
for (c=d->left; c != NULL; c = c->next) {
|
733
|
+
cs = skip_utf8_upper(c->string);
|
734
|
+
j = strlen(cs);
|
735
|
+
if (j > lengths[i]) lengths[i] = j;
|
736
|
+
i++;
|
737
|
+
}
|
738
|
+
for (c=d->right; c != NULL; c = c->next) {
|
739
|
+
cs = c->string;
|
740
|
+
cs = skip_utf8_upper(cs);
|
741
|
+
j = strlen(cs);
|
742
|
+
if (j > lengths[i]) lengths[i] = j;
|
743
|
+
i++;
|
744
|
+
}
|
745
|
+
d = d->next;
|
746
|
+
}
|
747
|
+
|
748
|
+
for (d = sent->and_data.label_table[k]; d!= NULL; d = d->next)
|
749
|
+
{
|
750
|
+
i=0;
|
751
|
+
s = buff;
|
752
|
+
for (c=d->left; c != NULL; c = c->next) {
|
753
|
+
s = stick_in_one_connector(s, c, lengths[i]);
|
754
|
+
i++;
|
755
|
+
}
|
756
|
+
for (c=d->right; c != NULL; c = c->next) {
|
757
|
+
s = stick_in_one_connector(s, c, lengths[i]);
|
758
|
+
i++;
|
759
|
+
}
|
760
|
+
d->string = string_set_add(buff, sent->string_set);
|
761
|
+
}
|
762
|
+
}
|
763
|
+
|
764
|
+
/**
|
765
|
+
* Goes through the entire sentence and builds the fat link tables
|
766
|
+
* for all the disjuncts of all the words.
|
767
|
+
*/
|
768
|
+
void build_conjunction_tables(Sentence sent)
|
769
|
+
{
|
770
|
+
int w;
|
771
|
+
int k;
|
772
|
+
Disjunct * d;
|
773
|
+
|
774
|
+
init_HT(sent);
|
775
|
+
init_LT(sent);
|
776
|
+
sent->and_data.STAT_N_disjuncts = 0;
|
777
|
+
sent->and_data.STAT_calls_to_equality_test = 0;
|
778
|
+
|
779
|
+
for (w=0; w<sent->length; w++) {
|
780
|
+
for (d=sent->word[w].d; d!=NULL; d=d->next) {
|
781
|
+
extract_all_fat_links(sent, d);
|
782
|
+
}
|
783
|
+
}
|
784
|
+
|
785
|
+
for (k=0; k<sent->and_data.LT_size; k++) {
|
786
|
+
compute_matchers_for_a_label(sent, k);
|
787
|
+
}
|
788
|
+
}
|
789
|
+
|
790
|
+
void print_AND_statistics(Sentence sent)
|
791
|
+
{
|
792
|
+
printf("Number of disjunct types (labels): %d\n", sent->and_data.LT_size);
|
793
|
+
printf("Number of disjuncts in the table: %d\n", sent->and_data.STAT_N_disjuncts);
|
794
|
+
if (sent->and_data.LT_size != 0) {
|
795
|
+
printf("average list length: %f\n",
|
796
|
+
(float)sent->and_data.STAT_N_disjuncts/sent->and_data.LT_size);
|
797
|
+
}
|
798
|
+
printf("Number of equality tests: %d\n", sent->and_data.STAT_calls_to_equality_test);
|
799
|
+
}
|
800
|
+
|
801
|
+
/**
|
802
|
+
* Fill in the fields of c for the disjunct. This must be in
|
803
|
+
* the table data structures. The label field and the string field
|
804
|
+
* are filled in appropriately. Priority is set to UP_priority.
|
805
|
+
*/
|
806
|
+
static void connector_for_disjunct(Sentence sent, Disjunct * d, Connector * c)
|
807
|
+
{
|
808
|
+
int h;
|
809
|
+
Disjunct * d1 = NULL;
|
810
|
+
Label_node * lp;
|
811
|
+
|
812
|
+
h = and_hash_disjunct(d);
|
813
|
+
|
814
|
+
for (lp = sent->and_data.hash_table[h]; lp != NULL; lp = lp->next) {
|
815
|
+
d1 = sent->and_data.label_table[lp->label];
|
816
|
+
if (disjunct_types_equal(d,d1)) break;
|
817
|
+
}
|
818
|
+
assert(lp != NULL, "A disjunct I inserted was not there. (1)");
|
819
|
+
|
820
|
+
while(d1 != NULL) {
|
821
|
+
if (disjuncts_equal_AND(sent, d1, d)) break;
|
822
|
+
d1 = d1->next;
|
823
|
+
}
|
824
|
+
|
825
|
+
assert(d1 != NULL, "A disjunct I inserted was not there. (2)");
|
826
|
+
|
827
|
+
c->label = lp->label;
|
828
|
+
connector_set_string(c, d1->string);
|
829
|
+
c->priority = UP_priority;
|
830
|
+
c->multi = FALSE;
|
831
|
+
}
|
832
|
+
|
833
|
+
|
834
|
+
/**
|
835
|
+
* This function allocates and returns a list of disjuncts.
|
836
|
+
* This is the one obtained by substituting each contiguous
|
837
|
+
* non-empty subrange of d (incident on the center) by an appropriate
|
838
|
+
* fat link, in two possible positions. Does not effect d.
|
839
|
+
* The cost of d is inherited by all of the disjuncts in the result.
|
840
|
+
*/
|
841
|
+
static Disjunct * build_fat_link_substitutions(Sentence sent, Disjunct *d)
|
842
|
+
{
|
843
|
+
Connector * cl, * cr, *tl, *tr, *wc, work_connector;
|
844
|
+
Disjunct *d1, *wd, work_disjunct, *d_list;
|
845
|
+
if (d==NULL) return NULL;
|
846
|
+
wd = &work_disjunct;
|
847
|
+
wc = init_connector(&work_connector);
|
848
|
+
d_list = NULL;
|
849
|
+
*wd = *d;
|
850
|
+
tl = d->left;
|
851
|
+
d->left = NULL;
|
852
|
+
for (cr = d->right; cr!=NULL; cr = cr->next) {
|
853
|
+
tr = cr->next;
|
854
|
+
cr->next = NULL;
|
855
|
+
if (is_appropriate(sent, d)) {
|
856
|
+
connector_for_disjunct(sent, d, wc);
|
857
|
+
wd->left = tl;
|
858
|
+
wd->right = wc;
|
859
|
+
wc->next = tr;
|
860
|
+
d1 = copy_disjunct(wd);
|
861
|
+
d1->next = d_list;
|
862
|
+
d_list = d1;
|
863
|
+
wd->left = wc;
|
864
|
+
wc->next = tl;
|
865
|
+
wd->right = tr;
|
866
|
+
d1 = copy_disjunct(wd);
|
867
|
+
d1->next = d_list;
|
868
|
+
d_list = d1;
|
869
|
+
}
|
870
|
+
cr->next = tr;
|
871
|
+
}
|
872
|
+
d->left = tl;
|
873
|
+
|
874
|
+
tr = d->right;
|
875
|
+
d->right = NULL;
|
876
|
+
for (cl = d->left; cl!=NULL; cl = cl->next) {
|
877
|
+
tl = cl->next;
|
878
|
+
cl->next = NULL;
|
879
|
+
if (is_appropriate(sent, d)) {
|
880
|
+
connector_for_disjunct(sent, d, wc);
|
881
|
+
wd->left = tl;
|
882
|
+
wd->right = wc;
|
883
|
+
wc->next = tr;
|
884
|
+
d1 = copy_disjunct(wd);
|
885
|
+
d1->next = d_list;
|
886
|
+
d_list = d1;
|
887
|
+
wd->left = wc;
|
888
|
+
wc->next = tl;
|
889
|
+
wd->right = tr;
|
890
|
+
d1 = copy_disjunct(wd);
|
891
|
+
d1->next = d_list;
|
892
|
+
d_list = d1;
|
893
|
+
}
|
894
|
+
cl->next = tl;
|
895
|
+
}
|
896
|
+
d->right = tr;
|
897
|
+
|
898
|
+
for (cl = d->left; cl!=NULL; cl = cl->next) {
|
899
|
+
for (cr = d->right; cr!=NULL; cr = cr->next) {
|
900
|
+
tl = cl->next;
|
901
|
+
tr = cr->next;
|
902
|
+
cl->next = cr->next = NULL;
|
903
|
+
if (is_appropriate(sent, d)) {
|
904
|
+
connector_for_disjunct(sent, d, wc);
|
905
|
+
wd->left = tl;
|
906
|
+
wd->right = wc;
|
907
|
+
wc->next = tr;
|
908
|
+
d1 = copy_disjunct(wd);
|
909
|
+
d1->next = d_list;
|
910
|
+
d_list = d1;
|
911
|
+
wd->left = wc;
|
912
|
+
wc->next = tl;
|
913
|
+
wd->right = tr;
|
914
|
+
d1 = copy_disjunct(wd);
|
915
|
+
d1->next = d_list;
|
916
|
+
d_list = d1;
|
917
|
+
}
|
918
|
+
cl->next = tl;
|
919
|
+
cr->next = tr;
|
920
|
+
}
|
921
|
+
}
|
922
|
+
return d_list;
|
923
|
+
}
|
924
|
+
|
925
|
+
/**
|
926
|
+
* This is basically a "map" function for build_fat_link_substitutions.
|
927
|
+
* It's applied to the disjuncts for all regular words of the sentence.
|
928
|
+
*/
|
929
|
+
Disjunct * explode_disjunct_list(Sentence sent, Disjunct *d)
|
930
|
+
{
|
931
|
+
Disjunct *d1;
|
932
|
+
|
933
|
+
d1 = NULL;
|
934
|
+
|
935
|
+
for (; d!=NULL; d = d->next) {
|
936
|
+
d1 = catenate_disjuncts(d1, build_fat_link_substitutions(sent, d));
|
937
|
+
}
|
938
|
+
return d1;
|
939
|
+
}
|
940
|
+
|
941
|
+
/**
|
942
|
+
* Builds and returns a disjunct list for the comma. These are the
|
943
|
+
* disjuncts that are used when "," operates in conjunction with "and".
|
944
|
+
* Does not deal with the ", and" issue, nor the other uses
|
945
|
+
* of comma.
|
946
|
+
*/
|
947
|
+
Disjunct * build_COMMA_disjunct_list(Sentence sent)
|
948
|
+
{
|
949
|
+
int lab;
|
950
|
+
Disjunct *d1, *d2, *d, work_disjunct, *wd;
|
951
|
+
Connector work_connector1, work_connector2, *c1, *c2;
|
952
|
+
Connector work_connector3, *c3;
|
953
|
+
c1 = init_connector(&work_connector1);
|
954
|
+
c2 = init_connector(&work_connector2);
|
955
|
+
c3 = init_connector(&work_connector3);
|
956
|
+
wd = &work_disjunct;
|
957
|
+
|
958
|
+
d1 = NULL; /* where we put the list we're building */
|
959
|
+
|
960
|
+
c1->next = NULL;
|
961
|
+
c2->next = c3;
|
962
|
+
c3->next = NULL;
|
963
|
+
c1->priority = c3->priority = DOWN_priority;
|
964
|
+
c2->priority = UP_priority;
|
965
|
+
c1->multi = c2->multi = c3->multi = FALSE;
|
966
|
+
wd->left = c1;
|
967
|
+
wd->right = c2;
|
968
|
+
wd->string = ","; /* *** fix this later?? */
|
969
|
+
wd->next = NULL;
|
970
|
+
wd->cost = 0;
|
971
|
+
for (lab = 0; lab < sent->and_data.LT_size; lab++) {
|
972
|
+
for (d = sent->and_data.label_table[lab]; d!=NULL; d=d->next) {
|
973
|
+
c1->string = c2->string = c3->string = d->string;
|
974
|
+
c1->label = c2->label = c3->label = lab;
|
975
|
+
d2 = copy_disjunct(wd);
|
976
|
+
d2->next = d1;
|
977
|
+
d1 = d2;
|
978
|
+
}
|
979
|
+
}
|
980
|
+
return d1;
|
981
|
+
}
|
982
|
+
|
983
|
+
/**
|
984
|
+
* Builds and returns a disjunct list for "and", "or" and "nor"
|
985
|
+
* for each disjunct in the label_table, we build three disjuncts
|
986
|
+
* this means that "Danny and Tycho and Billy" will be parsable in
|
987
|
+
* two ways. I don't know an easy way to avoid this
|
988
|
+
* the string is either "and", or "or", or "nor" at the moment.
|
989
|
+
*/
|
990
|
+
Disjunct * build_AND_disjunct_list(Sentence sent, char * s)
|
991
|
+
{
|
992
|
+
int lab;
|
993
|
+
Disjunct *d_list, *d1, *d3, *d, *d_copy;
|
994
|
+
Connector *c1, *c2, *c3;
|
995
|
+
|
996
|
+
d_list = NULL; /* where we put the list we're building */
|
997
|
+
|
998
|
+
for (lab = 0; lab < sent->and_data.LT_size; lab++) {
|
999
|
+
for (d = sent->and_data.label_table[lab]; d!=NULL; d=d->next) {
|
1000
|
+
d1 = build_fat_link_substitutions(sent, d);
|
1001
|
+
d_copy = copy_disjunct(d); /* also include the thing itself! */
|
1002
|
+
d_copy->next = d1;
|
1003
|
+
d1 = d_copy;
|
1004
|
+
for(;d1 != NULL; d1 = d3) {
|
1005
|
+
d3 = d1->next;
|
1006
|
+
|
1007
|
+
c1 = connector_new();
|
1008
|
+
c2 = connector_new();
|
1009
|
+
c1->priority = c2->priority = DOWN_priority;
|
1010
|
+
connector_set_string(c1, d->string);
|
1011
|
+
connector_set_string(c2, d->string);
|
1012
|
+
c1->label = c2->label = lab;
|
1013
|
+
|
1014
|
+
d1->string = s;
|
1015
|
+
|
1016
|
+
if (d1->right == NULL) {
|
1017
|
+
d1->right = c2;
|
1018
|
+
} else {
|
1019
|
+
for (c3=d1->right; c3->next != NULL; c3 = c3->next)
|
1020
|
+
;
|
1021
|
+
c3->next = c2;
|
1022
|
+
}
|
1023
|
+
if (d1->left == NULL) {
|
1024
|
+
d1->left = c1;
|
1025
|
+
} else {
|
1026
|
+
for (c3=d1->left; c3->next != NULL; c3 = c3->next)
|
1027
|
+
;
|
1028
|
+
c3->next = c1;
|
1029
|
+
}
|
1030
|
+
d1->next = d_list;
|
1031
|
+
d_list = d1;
|
1032
|
+
}
|
1033
|
+
}
|
1034
|
+
}
|
1035
|
+
#if defined(PLURALIZATION)
|
1036
|
+
/* here is where "and" makes singular into plural. */
|
1037
|
+
/* must accommodate "he and I are good", "Davy and I are good"
|
1038
|
+
"Danny and Davy are good", and reject all of these with "is"
|
1039
|
+
instead of "are".
|
1040
|
+
|
1041
|
+
The SI connectors must also be modified to accommodate "are John
|
1042
|
+
and Dave here", but kill "is John and Dave here"
|
1043
|
+
*/
|
1044
|
+
if (strcmp(s, "and") == 0)
|
1045
|
+
{
|
1046
|
+
for (d1 = d_list; d1 != NULL; d1 = d1->next)
|
1047
|
+
{
|
1048
|
+
for (c1 = d1->right; c1 != NULL; c1 = c1->next)
|
1049
|
+
{
|
1050
|
+
if ((c1->string[0] == 'S') &&
|
1051
|
+
((c1->string[1] == '^') ||
|
1052
|
+
(c1->string[1] == 's') ||
|
1053
|
+
(c1->string[1] == 'p') ||
|
1054
|
+
(c1->string[1] == '\0')))
|
1055
|
+
{
|
1056
|
+
connector_set_string(c1, "Sp");
|
1057
|
+
}
|
1058
|
+
}
|
1059
|
+
for (c1 = d1->left; c1 != NULL; c1 = c1->next)
|
1060
|
+
{
|
1061
|
+
if ((c1->string[0] == 'S') && (c1->string[1] == 'I') &&
|
1062
|
+
((c1->string[2] == '^') ||
|
1063
|
+
(c1->string[2] == 's') ||
|
1064
|
+
(c1->string[2] == 'p') ||
|
1065
|
+
(c1->string[2] == '\0')))
|
1066
|
+
{
|
1067
|
+
connector_set_string(c1, "SIp");
|
1068
|
+
}
|
1069
|
+
}
|
1070
|
+
}
|
1071
|
+
}
|
1072
|
+
/*
|
1073
|
+
"a cat or a dog is here" vs "a cat or a dog are here"
|
1074
|
+
The first seems right, the second seems wrong. I'll stick with this.
|
1075
|
+
|
1076
|
+
That is, "or" has the property that if both parts are the same in
|
1077
|
+
number, we use that but if they differ, we use plural.
|
1078
|
+
|
1079
|
+
The connectors on "I" must be handled specially. We accept
|
1080
|
+
"I or the dogs are here" but reject "I or the dogs is here"
|
1081
|
+
*/
|
1082
|
+
|
1083
|
+
/* the code here still does now work "right", rejecting "is John or I invited"
|
1084
|
+
and accepting "I or my friend know what happened"
|
1085
|
+
|
1086
|
+
The more generous code for "nor" has been used instead
|
1087
|
+
*/
|
1088
|
+
/*
|
1089
|
+
else if (strcmp(s, "or") == 0) {
|
1090
|
+
for (d1 = d_list; d1!=NULL; d1=d1->next) {
|
1091
|
+
for (c1=d1->right; c1!=NULL; c1=c1->next) {
|
1092
|
+
if (c1->string[0] == 'S') {
|
1093
|
+
if (c1->string[1]=='^') {
|
1094
|
+
if (c1->string[2]=='a') {
|
1095
|
+
connector_set_string(c1, "Ss");
|
1096
|
+
} else {
|
1097
|
+
connector_set_string(c1, "Sp");
|
1098
|
+
}
|
1099
|
+
} else if ((c1->string[1]=='p') && (c1->string[2]=='a')){
|
1100
|
+
connector_set_string(c1, "Sp");
|
1101
|
+
}
|
1102
|
+
}
|
1103
|
+
}
|
1104
|
+
for (c1=d1->left; c1!=NULL; c1=c1->next) {
|
1105
|
+
if ((c1->string[0] == 'S') && (c1->string[1] == 'I')) {
|
1106
|
+
if (c1->string[2]=='^') {
|
1107
|
+
if (c1->string[3]=='a') {
|
1108
|
+
connector_set_string(c1, "Ss");
|
1109
|
+
} else {
|
1110
|
+
connector_set_string(c1, "Sp");
|
1111
|
+
}
|
1112
|
+
} else if ((c1->string[2]=='p') && (c1->string[3]=='a')){
|
1113
|
+
connector_set_string(c1, "Sp");
|
1114
|
+
}
|
1115
|
+
}
|
1116
|
+
}
|
1117
|
+
}
|
1118
|
+
}
|
1119
|
+
*/
|
1120
|
+
/*
|
1121
|
+
It appears that the "nor" of two things can be either singular or
|
1122
|
+
plural. "neither she nor John likes dogs"
|
1123
|
+
"neither she nor John like dogs"
|
1124
|
+
|
1125
|
+
*/
|
1126
|
+
else if ((strcmp(s,"nor")==0) || (strcmp(s,"or")==0)) {
|
1127
|
+
for (d1 = d_list; d1!=NULL; d1=d1->next) {
|
1128
|
+
for (c1=d1->right; c1!=NULL; c1=c1->next) {
|
1129
|
+
if ((c1->string[0] == 'S') &&
|
1130
|
+
((c1->string[1]=='^') ||
|
1131
|
+
(c1->string[1]=='s') ||
|
1132
|
+
(c1->string[1]=='p'))) {
|
1133
|
+
connector_set_string(c1, "S");
|
1134
|
+
}
|
1135
|
+
}
|
1136
|
+
for (c1=d1->left; c1!=NULL; c1=c1->next) {
|
1137
|
+
if ((c1->string[0] == 'S') && (c1->string[1] == 'I') &&
|
1138
|
+
((c1->string[2]=='^') ||
|
1139
|
+
(c1->string[2]=='s') ||
|
1140
|
+
(c1->string[2]=='p'))) {
|
1141
|
+
connector_set_string(c1, "SI");
|
1142
|
+
}
|
1143
|
+
}
|
1144
|
+
}
|
1145
|
+
}
|
1146
|
+
|
1147
|
+
#endif
|
1148
|
+
return d_list;
|
1149
|
+
}
|
1150
|
+
|
1151
|
+
|
1152
|
+
/* The following routines' purpose is to eliminate all but the
|
1153
|
+
canonical linkage (of a collection of linkages that are identical
|
1154
|
+
except for fat links). An example of the problem is
|
1155
|
+
"I went to a talk and ate lunch". Without the canonical checker
|
1156
|
+
this has two linkages with identical structure.
|
1157
|
+
|
1158
|
+
We restrict our attention to a collection of linkages that are all
|
1159
|
+
isomorphic. Consider the set of all disjuncts that are used on one
|
1160
|
+
word (over the collection of linkages). This set is closed under GCD,
|
1161
|
+
since two linkages could both be used in that position, then so could
|
1162
|
+
their GCD. The GCD has been constructed and put in the label table.
|
1163
|
+
|
1164
|
+
The canonical linkage is the one in which the minimal disjunct that
|
1165
|
+
ever occurrs in a position is used in that position. It is easy to
|
1166
|
+
prove that a disjunct is not canonical -- just find one of it's fat
|
1167
|
+
disjuncts that can be replaced by a smaller one. If this can not be
|
1168
|
+
done, then the linkage is canonical.
|
1169
|
+
|
1170
|
+
The algorithm uses link_array[] and chosen_disjuncts[] as input to
|
1171
|
+
describe the linkage, and also uses the label_table.
|
1172
|
+
|
1173
|
+
(1) find all the words with fat disjuncts
|
1174
|
+
(2) scan all links and build, for each fat disjucnt used,
|
1175
|
+
an "image" structure that contains what this disjunct must
|
1176
|
+
connect to in the rest of the linkage.
|
1177
|
+
(3) For each fat disjunct, run through the label_table for disjuncts
|
1178
|
+
with the same label, considering only those with strictly more
|
1179
|
+
restricted match sets (this uses the string fields of the disjuncts
|
1180
|
+
from the table).
|
1181
|
+
(4) For each that passes this test, we see if it can replace the chosen
|
1182
|
+
disjunct. This is performed by examining how this disjunct
|
1183
|
+
compares with the image structure for this word.
|
1184
|
+
*/
|
1185
|
+
|
1186
|
+
struct Image_node_struct {
|
1187
|
+
Image_node * next;
|
1188
|
+
Connector * c; /* the connector the place on the disjunct must match */
|
1189
|
+
int place; /* Indicates the place in the fat disjunct where this
|
1190
|
+
connector must connect. If 0 then this is a fat
|
1191
|
+
connector. If >0 then go place to the right, if
|
1192
|
+
<0 then go -place to the left. */
|
1193
|
+
};
|
1194
|
+
|
1195
|
+
/**
|
1196
|
+
* Fill in the has_fat_down array. Uses link_array[].
|
1197
|
+
* Returns TRUE if there exists at least one word with a
|
1198
|
+
* fat down label.
|
1199
|
+
*/
|
1200
|
+
int set_has_fat_down(Sentence sent)
|
1201
|
+
{
|
1202
|
+
int link, w, N_fat;
|
1203
|
+
Parse_info pi = sent->parse_info;
|
1204
|
+
|
1205
|
+
N_fat = 0;
|
1206
|
+
|
1207
|
+
for (w = 0; w < pi->N_words; w++)
|
1208
|
+
{
|
1209
|
+
pi->has_fat_down[w] = FALSE;
|
1210
|
+
}
|
1211
|
+
|
1212
|
+
for (link = 0; link < pi->N_links; link++)
|
1213
|
+
{
|
1214
|
+
if (pi->link_array[link].lc->priority == DOWN_priority)
|
1215
|
+
{
|
1216
|
+
N_fat ++;
|
1217
|
+
pi->has_fat_down[pi->link_array[link].l] = TRUE;
|
1218
|
+
}
|
1219
|
+
else if (pi->link_array[link].rc->priority == DOWN_priority)
|
1220
|
+
{
|
1221
|
+
N_fat ++;
|
1222
|
+
pi->has_fat_down[pi->link_array[link].r] = TRUE;
|
1223
|
+
}
|
1224
|
+
}
|
1225
|
+
return (N_fat > 0);
|
1226
|
+
}
|
1227
|
+
|
1228
|
+
static void free_image_array(Parse_info pi)
|
1229
|
+
{
|
1230
|
+
int w;
|
1231
|
+
Image_node * in, * inx;
|
1232
|
+
for (w = 0; w < pi->N_words; w++)
|
1233
|
+
{
|
1234
|
+
for (in = pi->image_array[w]; in != NULL; in = inx)
|
1235
|
+
{
|
1236
|
+
inx = in->next;
|
1237
|
+
xfree((char *)in, sizeof(Image_node));
|
1238
|
+
}
|
1239
|
+
pi->image_array[w] = NULL;
|
1240
|
+
}
|
1241
|
+
}
|
1242
|
+
|
1243
|
+
/**
|
1244
|
+
* Uses link_array, chosen_disjuncts, and down_label to construct
|
1245
|
+
* image_array
|
1246
|
+
*/
|
1247
|
+
static void build_image_array(Sentence sent)
|
1248
|
+
{
|
1249
|
+
int link, end, word;
|
1250
|
+
Connector * this_end_con, *other_end_con, * upcon, * updiscon, *clist;
|
1251
|
+
Disjunct * dis, * updis;
|
1252
|
+
Image_node * in;
|
1253
|
+
Parse_info pi = sent->parse_info;
|
1254
|
+
|
1255
|
+
for (word=0; word<pi->N_words; word++)
|
1256
|
+
{
|
1257
|
+
pi->image_array[word] = NULL;
|
1258
|
+
}
|
1259
|
+
|
1260
|
+
for (end = -1; end <= 1; end += 2)
|
1261
|
+
{
|
1262
|
+
for (link = 0; link < pi->N_links; link++)
|
1263
|
+
{
|
1264
|
+
if (end < 0)
|
1265
|
+
{
|
1266
|
+
word = pi->link_array[link].l;
|
1267
|
+
if (!pi->has_fat_down[word]) continue;
|
1268
|
+
this_end_con = pi->link_array[link].lc;
|
1269
|
+
other_end_con = pi->link_array[link].rc;
|
1270
|
+
dis = pi->chosen_disjuncts[word];
|
1271
|
+
clist = dis->right;
|
1272
|
+
}
|
1273
|
+
else
|
1274
|
+
{
|
1275
|
+
word = pi->link_array[link].r;
|
1276
|
+
if (!pi->has_fat_down[word]) continue;
|
1277
|
+
this_end_con = pi->link_array[link].rc;
|
1278
|
+
other_end_con = pi->link_array[link].lc;
|
1279
|
+
dis = pi->chosen_disjuncts[word];
|
1280
|
+
clist = dis->left;
|
1281
|
+
}
|
1282
|
+
|
1283
|
+
if (this_end_con->priority == DOWN_priority) continue;
|
1284
|
+
if ((this_end_con->label != NORMAL_LABEL) &&
|
1285
|
+
(this_end_con->label < 0)) continue;
|
1286
|
+
/* no need to construct an image node for down links,
|
1287
|
+
or commas links or either/neither links */
|
1288
|
+
|
1289
|
+
in = (Image_node *) xalloc(sizeof(Image_node));
|
1290
|
+
in->next = pi->image_array[word];
|
1291
|
+
pi->image_array[word] = in;
|
1292
|
+
in->c = other_end_con;
|
1293
|
+
|
1294
|
+
/* the rest of this code is for computing in->place */
|
1295
|
+
if (this_end_con->priority == UP_priority)
|
1296
|
+
{
|
1297
|
+
in->place = 0;
|
1298
|
+
}
|
1299
|
+
else
|
1300
|
+
{
|
1301
|
+
in->place = 1;
|
1302
|
+
if ((dis->left != NULL) &&
|
1303
|
+
(dis->left->priority == UP_priority))
|
1304
|
+
{
|
1305
|
+
upcon = dis->left;
|
1306
|
+
}
|
1307
|
+
else if ((dis->right != NULL) &&
|
1308
|
+
(dis->right->priority == UP_priority))
|
1309
|
+
{
|
1310
|
+
upcon = dis->right;
|
1311
|
+
}
|
1312
|
+
else
|
1313
|
+
{
|
1314
|
+
upcon = NULL;
|
1315
|
+
}
|
1316
|
+
if (upcon != NULL)
|
1317
|
+
{
|
1318
|
+
/* add on extra for a fat up link */
|
1319
|
+
updis = sent->and_data.label_table[upcon->label];
|
1320
|
+
if (end > 0)
|
1321
|
+
{
|
1322
|
+
updiscon = updis->left;
|
1323
|
+
}
|
1324
|
+
else
|
1325
|
+
{
|
1326
|
+
updiscon = updis->right;
|
1327
|
+
}
|
1328
|
+
for (;updiscon != NULL; updiscon = updiscon->next)
|
1329
|
+
{
|
1330
|
+
in->place ++;
|
1331
|
+
}
|
1332
|
+
}
|
1333
|
+
for (; clist != this_end_con; clist = clist->next)
|
1334
|
+
{
|
1335
|
+
if (clist->label < 0) in->place++;
|
1336
|
+
}
|
1337
|
+
in->place = in->place * (-end);
|
1338
|
+
}
|
1339
|
+
}
|
1340
|
+
}
|
1341
|
+
}
|
1342
|
+
|
1343
|
+
/**
|
1344
|
+
* returns TRUE if string s represents a strictly smaller match set
|
1345
|
+
* than does t
|
1346
|
+
*/
|
1347
|
+
static int strictly_smaller(const char * s, const char * t)
|
1348
|
+
{
|
1349
|
+
int strictness;
|
1350
|
+
strictness = 0;
|
1351
|
+
for (;(*s!='\0') && (*t!='\0'); s++,t++) {
|
1352
|
+
if (*s == *t) continue;
|
1353
|
+
if ((*t == '*') || (*s == '^')) {
|
1354
|
+
strictness++;
|
1355
|
+
} else {
|
1356
|
+
return FALSE;
|
1357
|
+
}
|
1358
|
+
}
|
1359
|
+
assert(! ((*s!='\0') || (*t!='\0')), "s and t should be the same length!");
|
1360
|
+
return (strictness > 0);
|
1361
|
+
}
|
1362
|
+
|
1363
|
+
/**
|
1364
|
+
* dis points to a disjunct in the label_table. label is the label
|
1365
|
+
* of a different set of disjuncts. These can be derived from the label
|
1366
|
+
* of dis. Find the specific disjunct of in label_table[label]
|
1367
|
+
* which corresponds to dis.
|
1368
|
+
*/
|
1369
|
+
static Disjunct * find_subdisjunct(Sentence sent, Disjunct * dis, int label)
|
1370
|
+
{
|
1371
|
+
Disjunct * d;
|
1372
|
+
Connector * cx, *cy;
|
1373
|
+
for (d=sent->and_data.label_table[label]; d!=NULL; d=d->next)
|
1374
|
+
{
|
1375
|
+
for (cx=d->left, cy=dis->left; cx!=NULL; cx=cx->next,cy=cy->next)
|
1376
|
+
{
|
1377
|
+
/* if ((cx->string != cy->string) || */
|
1378
|
+
if ((strcmp(connector_get_string(cx),
|
1379
|
+
connector_get_string(cy)) != 0) ||
|
1380
|
+
(cx->multi != cy->multi)) break;/* have to check multi? */
|
1381
|
+
}
|
1382
|
+
if (cx!=NULL) continue;
|
1383
|
+
for (cx=d->right, cy=dis->right; cx!=NULL; cx=cx->next,cy=cy->next)
|
1384
|
+
{
|
1385
|
+
/* if ((cx->string != cy->string) || */
|
1386
|
+
if ((strcmp(connector_get_string(cx),
|
1387
|
+
connector_get_string(cy)) != 0) ||
|
1388
|
+
(cx->multi != cy->multi)) break;
|
1389
|
+
}
|
1390
|
+
if (cx==NULL) break;
|
1391
|
+
}
|
1392
|
+
assert(d!=NULL, "Never found subdisjunct");
|
1393
|
+
return d;
|
1394
|
+
}
|
1395
|
+
|
1396
|
+
/**
|
1397
|
+
* is_canonical_linkage --
|
1398
|
+
* This uses link_array[], chosen_disjuncts[], has_fat_down[].
|
1399
|
+
* It assumes that there is a fat link in the current linkage.
|
1400
|
+
* See the comments above for more information about how it works
|
1401
|
+
*/
|
1402
|
+
int is_canonical_linkage(Sentence sent)
|
1403
|
+
{
|
1404
|
+
int w, d_label=0, place;
|
1405
|
+
Connector *d_c, *c, dummy_connector, *upcon;
|
1406
|
+
Disjunct *dis, *chosen_d;
|
1407
|
+
Image_node * in;
|
1408
|
+
Parse_info pi = sent->parse_info;
|
1409
|
+
|
1410
|
+
init_connector(&dummy_connector);
|
1411
|
+
dummy_connector.priority = UP_priority;
|
1412
|
+
|
1413
|
+
build_image_array(sent);
|
1414
|
+
|
1415
|
+
for (w=0; w<pi->N_words; w++)
|
1416
|
+
{
|
1417
|
+
if (!pi->has_fat_down[w]) continue;
|
1418
|
+
chosen_d = pi->chosen_disjuncts[w];
|
1419
|
+
|
1420
|
+
/* there must be a down connector in both the left and right list */
|
1421
|
+
for (d_c = chosen_d->left; d_c!=NULL; d_c=d_c->next)
|
1422
|
+
{
|
1423
|
+
if (d_c->priority == DOWN_priority)
|
1424
|
+
{
|
1425
|
+
d_label = d_c->label;
|
1426
|
+
break;
|
1427
|
+
}
|
1428
|
+
}
|
1429
|
+
assert(d_c != NULL, "Should have found the down link.");
|
1430
|
+
|
1431
|
+
if ((chosen_d->left != NULL) &&
|
1432
|
+
(chosen_d->left->priority == UP_priority)) {
|
1433
|
+
upcon = chosen_d->left;
|
1434
|
+
} else if ((chosen_d->right != NULL) &&
|
1435
|
+
(chosen_d->right->priority == UP_priority)) {
|
1436
|
+
upcon = chosen_d->right;
|
1437
|
+
} else {
|
1438
|
+
upcon = NULL;
|
1439
|
+
}
|
1440
|
+
|
1441
|
+
/* check that the disjunct on w is minimal (canonical) */
|
1442
|
+
|
1443
|
+
for (dis=sent->and_data.label_table[d_label]; dis!=NULL; dis=dis->next)
|
1444
|
+
{
|
1445
|
+
/* now, reject a disjunct if it's not strictly below the old */
|
1446
|
+
if(!strictly_smaller(dis->string,
|
1447
|
+
connector_get_string(d_c))) continue;
|
1448
|
+
|
1449
|
+
/* Now, it has to match the image connectors */
|
1450
|
+
for (in = pi->image_array[w]; in != NULL; in = in->next)
|
1451
|
+
{
|
1452
|
+
place = in->place;
|
1453
|
+
if (place == 0)
|
1454
|
+
{
|
1455
|
+
assert(upcon != NULL, "Should have found an up link");
|
1456
|
+
dummy_connector.label = upcon->label;
|
1457
|
+
|
1458
|
+
/* now we have to compute the string of the
|
1459
|
+
disjunct with upcon->label that corresponds
|
1460
|
+
to dis */
|
1461
|
+
if (upcon->label == d_label)
|
1462
|
+
{
|
1463
|
+
connector_set_string(&dummy_connector, dis->string);
|
1464
|
+
} else {
|
1465
|
+
connector_set_string(&dummy_connector,
|
1466
|
+
find_subdisjunct(sent, dis, upcon->label)->string);
|
1467
|
+
}
|
1468
|
+
|
1469
|
+
/* I hope using x_match here is right */
|
1470
|
+
if (!x_match(sent, &dummy_connector, in->c)) break;
|
1471
|
+
} else if (place > 0) {
|
1472
|
+
for (c=dis->right; place > 1; place--) {
|
1473
|
+
c = c->next;
|
1474
|
+
}
|
1475
|
+
if (!x_match(sent, c, in->c)) break; /* Ditto above comment --DS 07/97*/
|
1476
|
+
} else {
|
1477
|
+
for (c=dis->left; place < -1; place++) {
|
1478
|
+
c = c->next;
|
1479
|
+
}
|
1480
|
+
if (!x_match(sent, c, in->c)) break; /* Ditto Ditto */
|
1481
|
+
}
|
1482
|
+
}
|
1483
|
+
|
1484
|
+
if (in == NULL) break;
|
1485
|
+
}
|
1486
|
+
if (dis != NULL) break;
|
1487
|
+
/* there is a better disjunct that the one we're using, so this
|
1488
|
+
word is bad, so we're done */
|
1489
|
+
}
|
1490
|
+
free_image_array(pi);
|
1491
|
+
return (w == pi->N_words);
|
1492
|
+
}
|
1493
|
+
|
1494
|
+
/**
|
1495
|
+
* This takes as input link_array[], sublinkage->link[]->l and
|
1496
|
+
* sublinkage->link[]->r (and also has_fat_down[word], which has been
|
1497
|
+
* computed in a prior call to is_canonical()), and from these
|
1498
|
+
* computes sublinkage->link[].lc and .rc. We assume these have
|
1499
|
+
* been initialized with the values from link_array. We also assume
|
1500
|
+
* that there are fat links.
|
1501
|
+
*/
|
1502
|
+
void compute_pp_link_array_connectors(Sentence sent, Sublinkage *sublinkage)
|
1503
|
+
{
|
1504
|
+
int link, end, word, place;
|
1505
|
+
Connector * this_end_con, * upcon, * updiscon, *clist, *con, *mycon;
|
1506
|
+
Disjunct * dis, * updis, *mydis;
|
1507
|
+
Parse_info pi = sent->parse_info;
|
1508
|
+
|
1509
|
+
for (end = -1; end <= 1; end += 2)
|
1510
|
+
{
|
1511
|
+
for (link=0; link<pi->N_links; link++)
|
1512
|
+
{
|
1513
|
+
if (sublinkage->link[link]->l == -1) continue;
|
1514
|
+
if (end < 0)
|
1515
|
+
{
|
1516
|
+
word = pi->link_array[link].l;
|
1517
|
+
if (!pi->has_fat_down[word]) continue;
|
1518
|
+
this_end_con = pi->link_array[link].lc;
|
1519
|
+
dis = pi->chosen_disjuncts[word];
|
1520
|
+
mydis = pi->chosen_disjuncts[sublinkage->link[link]->l];
|
1521
|
+
clist = dis->right;
|
1522
|
+
}
|
1523
|
+
else
|
1524
|
+
{
|
1525
|
+
word = pi->link_array[link].r;
|
1526
|
+
if (!pi->has_fat_down[word]) continue;
|
1527
|
+
this_end_con = pi->link_array[link].rc;
|
1528
|
+
dis = pi->chosen_disjuncts[word];
|
1529
|
+
mydis = pi->chosen_disjuncts[sublinkage->link[link]->r];
|
1530
|
+
clist = dis->left;
|
1531
|
+
}
|
1532
|
+
|
1533
|
+
if (this_end_con->label != NORMAL_LABEL) continue;
|
1534
|
+
/* no need to construct a connector for up links,
|
1535
|
+
or commas links or either/neither links */
|
1536
|
+
|
1537
|
+
/* Now compute the place */
|
1538
|
+
place = 0;
|
1539
|
+
if ((dis->left != NULL) &&
|
1540
|
+
(dis->left->priority == UP_priority)) {
|
1541
|
+
upcon = dis->left;
|
1542
|
+
} else if ((dis->right != NULL) &&
|
1543
|
+
(dis->right->priority == UP_priority)) {
|
1544
|
+
upcon = dis->right;
|
1545
|
+
} else {
|
1546
|
+
upcon = NULL;
|
1547
|
+
}
|
1548
|
+
if (upcon != NULL) { /* add on extra for a fat up link */
|
1549
|
+
updis = sent->and_data.label_table[upcon->label];
|
1550
|
+
if (end > 0) {
|
1551
|
+
updiscon = updis->left;
|
1552
|
+
} else {
|
1553
|
+
updiscon = updis->right;
|
1554
|
+
}
|
1555
|
+
for (;updiscon != NULL; updiscon = updiscon->next) {
|
1556
|
+
place ++;
|
1557
|
+
}
|
1558
|
+
}
|
1559
|
+
for (; clist != this_end_con; clist = clist->next) {
|
1560
|
+
if (clist->label < 0) place++;
|
1561
|
+
}
|
1562
|
+
/* place has just been computed */
|
1563
|
+
|
1564
|
+
/* now find the right disjunct in the table */
|
1565
|
+
if ((mydis->left != NULL) &&
|
1566
|
+
(mydis->left->priority == UP_priority)) {
|
1567
|
+
mycon = mydis->left;
|
1568
|
+
} else if ((mydis->right != NULL) &&
|
1569
|
+
(mydis->right->priority == UP_priority)) {
|
1570
|
+
mycon = mydis->right;
|
1571
|
+
} else {
|
1572
|
+
printf("word = %d\n", word);
|
1573
|
+
printf("fat link: [%d, %d]\n",
|
1574
|
+
pi->link_array[link].l, pi->link_array[link].r);
|
1575
|
+
printf("thin link: [%d, %d]\n",
|
1576
|
+
sublinkage->link[link]->l, sublinkage->link[link]->r);
|
1577
|
+
assert(FALSE, "There should be a fat UP link here");
|
1578
|
+
}
|
1579
|
+
|
1580
|
+
for (dis=sent->and_data.label_table[mycon->label];
|
1581
|
+
dis != NULL; dis=dis->next) {
|
1582
|
+
if (dis->string == connector_get_string(mycon)) break;
|
1583
|
+
}
|
1584
|
+
assert(dis!=NULL, "Should have found this connector string");
|
1585
|
+
/* the disjunct in the table has just been found */
|
1586
|
+
|
1587
|
+
if (end < 0)
|
1588
|
+
{
|
1589
|
+
for (con = dis->right; place > 0; place--, con=con->next) {}
|
1590
|
+
/* sublinkage->link[link]->lc = con; OLD CODE */
|
1591
|
+
exfree_connectors(sublinkage->link[link]->lc);
|
1592
|
+
sublinkage->link[link]->lc = excopy_connectors(con);
|
1593
|
+
}
|
1594
|
+
else
|
1595
|
+
{
|
1596
|
+
for (con = dis->left; place > 0; place--, con=con->next) {}
|
1597
|
+
/* sublinkage->link[link]->rc = con; OLD CODE */
|
1598
|
+
exfree_connectors(sublinkage->link[link]->rc);
|
1599
|
+
sublinkage->link[link]->rc = excopy_connectors(con);
|
1600
|
+
}
|
1601
|
+
}
|
1602
|
+
}
|
1603
|
+
}
|