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 +1,16 @@
|
|
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
|
+
void print_disjunct_counts(Sentence sent);
|
14
|
+
void print_sentence(FILE *fp, Sentence sent, int w);
|
15
|
+
void print_expression_sizes(Sentence sent);
|
16
|
+
void compute_chosen_words(Sentence sent, Linkage linkage);
|
@@ -0,0 +1,1864 @@
|
|
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
|
+
#define CONTABSZ 8192
|
18
|
+
typedef Connector * connector_table;
|
19
|
+
|
20
|
+
/*
|
21
|
+
typedef struct disjunct_dup_table_s disjunct_dup_table;
|
22
|
+
struct disjunct_dup_table_s
|
23
|
+
{
|
24
|
+
int dup_table_size;
|
25
|
+
Disjunct ** dup_table;
|
26
|
+
};
|
27
|
+
*/
|
28
|
+
|
29
|
+
/* the indiction in a word field that this connector cannot
|
30
|
+
* be used -- is obsolete.
|
31
|
+
*/
|
32
|
+
#define BAD_WORD (MAX_SENTENCE+1)
|
33
|
+
|
34
|
+
typedef struct c_list_s C_list;
|
35
|
+
struct c_list_s
|
36
|
+
{
|
37
|
+
Connector * c;
|
38
|
+
int shallow;
|
39
|
+
C_list * next;
|
40
|
+
};
|
41
|
+
|
42
|
+
typedef struct power_table_s power_table;
|
43
|
+
struct power_table_s
|
44
|
+
{
|
45
|
+
int power_table_size;
|
46
|
+
int l_table_size[MAX_SENTENCE]; /* the sizes of the hash tables */
|
47
|
+
int r_table_size[MAX_SENTENCE];
|
48
|
+
C_list ** l_table[MAX_SENTENCE];
|
49
|
+
C_list ** r_table[MAX_SENTENCE];
|
50
|
+
};
|
51
|
+
|
52
|
+
typedef struct cms_struct Cms;
|
53
|
+
struct cms_struct
|
54
|
+
{
|
55
|
+
Cms * next;
|
56
|
+
const char * name;
|
57
|
+
int count; /* the number of times this is in the multiset */
|
58
|
+
};
|
59
|
+
|
60
|
+
#define CMS_SIZE (2<<10)
|
61
|
+
typedef struct multiset_table_s multiset_table;
|
62
|
+
struct multiset_table_s
|
63
|
+
{
|
64
|
+
Cms * cms_table[CMS_SIZE];
|
65
|
+
};
|
66
|
+
|
67
|
+
typedef struct prune_context_s prune_context;
|
68
|
+
struct prune_context_s
|
69
|
+
{
|
70
|
+
int null_links;
|
71
|
+
char ** deletable;
|
72
|
+
char ** effective_dist;
|
73
|
+
int power_cost;
|
74
|
+
int power_prune_mode; /* either GENTLE or RUTHLESS */
|
75
|
+
int N_changed; /* counts the number of changes
|
76
|
+
of c->word fields in a pass */
|
77
|
+
|
78
|
+
power_table *pt;
|
79
|
+
Sentence sent;
|
80
|
+
};
|
81
|
+
|
82
|
+
/*
|
83
|
+
|
84
|
+
The algorithms in this file prune disjuncts from the disjunct list
|
85
|
+
of the sentence that can be elimininated by a simple checks. The first
|
86
|
+
check works as follows:
|
87
|
+
|
88
|
+
A series of passes are made through the sentence, alternating
|
89
|
+
left-to-right and right-to-left. Consier the left-to-right pass (the
|
90
|
+
other is symmetric). A set S of connectors is maintained (initialized
|
91
|
+
to be empty). Now the disjuncts of the current word are processed.
|
92
|
+
If a given disjunct's left pointing connectors have the property that
|
93
|
+
at least one of them has no connector in S to which it can be matched,
|
94
|
+
then that disjunct is deleted. Now the set S is augmented by the right
|
95
|
+
connectors of the remaining disjuncts of that word. This completes
|
96
|
+
one word. The process continues through the words from left to right.
|
97
|
+
Alternate passes are made until no disjunct is deleted.
|
98
|
+
|
99
|
+
It worries me a little that if there are some really huge disjuncts lists,
|
100
|
+
then this process will probably do nothing. (This fear turns out to be
|
101
|
+
unfounded.)
|
102
|
+
|
103
|
+
Notes: Power pruning will not work if applied before generating the
|
104
|
+
"and" disjuncts. This is because certain of it's tricks don't work.
|
105
|
+
Think about this, and finish this note later....
|
106
|
+
Also, currently I use the standard connector match procedure instead
|
107
|
+
of the pruning one, since I know power pruning will not be used before
|
108
|
+
and generation. Replace this to allow power pruning to work before
|
109
|
+
generating and disjuncts.
|
110
|
+
|
111
|
+
Currently it seems that normal pruning, power pruning, and generation,
|
112
|
+
pruning, and power pruning (after "and" generation) and parsing take
|
113
|
+
about the same amount of time. This is why doing power pruning before
|
114
|
+
"and" generation might be a very good idea.
|
115
|
+
|
116
|
+
New idea: Suppose all the disjuncts of a word have a connector of type
|
117
|
+
c pointing to the right. And further, suppose that there is exactly one
|
118
|
+
word to its right containing that type of connector pointing to the left.
|
119
|
+
Then all the other disjuncts on the latter word can be deleted.
|
120
|
+
(This situation is created by the processing of "either...or", and by
|
121
|
+
the extra disjuncts added to a "," neighboring a conjunction.)
|
122
|
+
|
123
|
+
*/
|
124
|
+
|
125
|
+
/**
|
126
|
+
* This hash function only looks at the leading upper case letters of
|
127
|
+
* the connector string, and the label fields. This ensures that if two
|
128
|
+
* strings match (formally), then they must hash to the same place.
|
129
|
+
*/
|
130
|
+
static inline int hash_S(Connector * c)
|
131
|
+
{
|
132
|
+
int h = connector_hash(c);
|
133
|
+
return (h & (CONTABSZ-1));
|
134
|
+
}
|
135
|
+
|
136
|
+
/**
|
137
|
+
* This is almost identical to match(). Its reason for existance
|
138
|
+
* is the rather subtle fact that with "and" can transform a "Ss"
|
139
|
+
* connector into "Sp". This means that in order for pruning to
|
140
|
+
* work, we must allow a "Ss" connector on word match an "Sp" connector
|
141
|
+
* on a word to its right. This is what this version of match allows.
|
142
|
+
* We assume that a is on a word to the left of b.
|
143
|
+
*/
|
144
|
+
int prune_match(int dist, Connector *a, Connector *b)
|
145
|
+
{
|
146
|
+
const char *s, *t;
|
147
|
+
int x, y;
|
148
|
+
|
149
|
+
if (a->label != b->label) return FALSE;
|
150
|
+
|
151
|
+
x = hash_S(a);
|
152
|
+
y = hash_S(b);
|
153
|
+
if (x != y) return FALSE;
|
154
|
+
|
155
|
+
s = a->string;
|
156
|
+
t = b->string;
|
157
|
+
|
158
|
+
while(s < a->prune_string || t < b->prune_string)
|
159
|
+
{
|
160
|
+
if (*s != *t) return FALSE;
|
161
|
+
s++;
|
162
|
+
t++;
|
163
|
+
}
|
164
|
+
|
165
|
+
/* printf("PM: a=%4s b=%4s ap=%d bp=%d a->ll=%d b->ll=%d dist=%d\n",
|
166
|
+
s, t, x, y, a->length_limit, b->length_limit, dist); */
|
167
|
+
if (dist > a->length_limit || dist > b->length_limit) return FALSE;
|
168
|
+
|
169
|
+
x = a->priority;
|
170
|
+
y = b->priority;
|
171
|
+
|
172
|
+
if ((x == THIN_priority) && (y == THIN_priority))
|
173
|
+
{
|
174
|
+
#if defined(PLURALIZATION)
|
175
|
+
/*
|
176
|
+
if ((*(a->string)=='S') && ((*s=='s') || (*s=='p')) && (*t=='p')) {
|
177
|
+
return TRUE;
|
178
|
+
}
|
179
|
+
*/
|
180
|
+
/*
|
181
|
+
The above is a kludge to stop pruning from killing off disjuncts
|
182
|
+
which (because of pluralization in and) might become valid later.
|
183
|
+
Recall that "and" converts a singular subject into a plural one.
|
184
|
+
The (*s=='p') part is so that "he and I are good" doesn't get killed off.
|
185
|
+
The above hack is subsumed by the following one:
|
186
|
+
*/
|
187
|
+
if ((*(a->string)=='S') && ((*s=='s') || (*s=='p')) &&
|
188
|
+
((*t=='p') || (*t=='s')) &&
|
189
|
+
((s-1 == a->string) || ((s-2 == a->string) && (*(s-1) == 'I')))){
|
190
|
+
return TRUE;
|
191
|
+
}
|
192
|
+
/*
|
193
|
+
This change is to accommodate "nor". In particular we need to
|
194
|
+
prevent "neither John nor I likes dogs" from being killed off.
|
195
|
+
We want to allow this to apply to "are neither a dog nor a cat here"
|
196
|
+
and "is neither a dog nor a cat here". This uses the "SI" connector.
|
197
|
+
The third line above ensures that the connector is either "S" or "SI".
|
198
|
+
*/
|
199
|
+
#endif
|
200
|
+
while ((*s != '\0') && (*t != '\0'))
|
201
|
+
{
|
202
|
+
if ((*s == '*') || (*t == '*') ||
|
203
|
+
((*s == *t) && (*s != '^')))
|
204
|
+
{
|
205
|
+
/* this last case here is rather obscure. It prevents
|
206
|
+
'^' from matching '^'.....Is this necessary?
|
207
|
+
......yes, I think it is. */
|
208
|
+
s++;
|
209
|
+
t++;
|
210
|
+
}
|
211
|
+
else
|
212
|
+
return FALSE;
|
213
|
+
}
|
214
|
+
return TRUE;
|
215
|
+
}
|
216
|
+
else if ((x == UP_priority) && (y == DOWN_priority))
|
217
|
+
{
|
218
|
+
while ((*s!='\0') && (*t!='\0'))
|
219
|
+
{
|
220
|
+
if ((*s == *t) || (*s == '*') || (*t == '^'))
|
221
|
+
{
|
222
|
+
/* that '^' should match on the DOWN_priority
|
223
|
+
node is subtle, but correct */
|
224
|
+
s++;
|
225
|
+
t++;
|
226
|
+
}
|
227
|
+
else
|
228
|
+
return FALSE;
|
229
|
+
}
|
230
|
+
return TRUE;
|
231
|
+
}
|
232
|
+
else if ((y == UP_priority) && (x == DOWN_priority))
|
233
|
+
{
|
234
|
+
while ((*s!='\0') && (*t!='\0'))
|
235
|
+
{
|
236
|
+
if ((*s == *t) || (*t == '*') || (*s == '^'))
|
237
|
+
{
|
238
|
+
s++;
|
239
|
+
t++;
|
240
|
+
}
|
241
|
+
else
|
242
|
+
return FALSE;
|
243
|
+
}
|
244
|
+
return TRUE;
|
245
|
+
}
|
246
|
+
else
|
247
|
+
return FALSE;
|
248
|
+
}
|
249
|
+
|
250
|
+
static void zero_connector_table(connector_table *ct)
|
251
|
+
{
|
252
|
+
memset(ct, 0, sizeof(Connector *) * CONTABSZ);
|
253
|
+
}
|
254
|
+
|
255
|
+
/**
|
256
|
+
* This function puts connector c into the connector table
|
257
|
+
* if one like it isn't already there.
|
258
|
+
*/
|
259
|
+
static void insert_connector(connector_table *ct, Connector * c)
|
260
|
+
{
|
261
|
+
int h;
|
262
|
+
Connector * e;
|
263
|
+
|
264
|
+
h = hash_S(c);
|
265
|
+
|
266
|
+
for (e = ct[h]; e != NULL; e = e->tableNext)
|
267
|
+
{
|
268
|
+
if ((strcmp(c->string, e->string) == 0) &&
|
269
|
+
(c->label == e->label) &&
|
270
|
+
(c->priority == e->priority))
|
271
|
+
return;
|
272
|
+
}
|
273
|
+
c->tableNext = ct[h];
|
274
|
+
ct[h] = c;
|
275
|
+
}
|
276
|
+
|
277
|
+
void prune(Sentence sent)
|
278
|
+
{
|
279
|
+
Connector *e, *f;
|
280
|
+
int w;
|
281
|
+
int N_deleted;
|
282
|
+
Connector *ct[CONTABSZ];
|
283
|
+
Disjunct fake_head, *d, *d1;
|
284
|
+
|
285
|
+
/* XXX why is this here ?? */
|
286
|
+
count_set_effective_distance(sent);
|
287
|
+
|
288
|
+
N_deleted = 1; /* a lie to make it always do at least 2 passes */
|
289
|
+
while(1)
|
290
|
+
{
|
291
|
+
/* Left-to-right pass */
|
292
|
+
zero_connector_table(ct);
|
293
|
+
|
294
|
+
/* For every word */
|
295
|
+
for (w = 0; w < sent->length; w++)
|
296
|
+
{
|
297
|
+
d = &fake_head;
|
298
|
+
d->next = sent->word[w].d;
|
299
|
+
|
300
|
+
/* For every disjunct of word */
|
301
|
+
while ((d1 = d->next))
|
302
|
+
{
|
303
|
+
e = d1->left;
|
304
|
+
|
305
|
+
/* For every left clause of this disjunct */
|
306
|
+
while (e)
|
307
|
+
{
|
308
|
+
int h = hash_S(e);
|
309
|
+
for (f = ct[h]; f != NULL; f = f->tableNext)
|
310
|
+
{
|
311
|
+
if (prune_match(0, f, e)) break;
|
312
|
+
}
|
313
|
+
if (!f) break; /* If f null, not a single match was found */
|
314
|
+
e = e->next;
|
315
|
+
}
|
316
|
+
|
317
|
+
/* We know this disjunct is dead since no match
|
318
|
+
* can be found on a required clause. */
|
319
|
+
if (e)
|
320
|
+
{
|
321
|
+
N_deleted ++;
|
322
|
+
free_connectors(d1->left);
|
323
|
+
free_connectors(d1->right);
|
324
|
+
d->next = d1->next;
|
325
|
+
xfree(d1, sizeof(Disjunct));
|
326
|
+
}
|
327
|
+
else
|
328
|
+
{
|
329
|
+
/* Store surviving disjunct in hash table */
|
330
|
+
for (e = d1->right; e != NULL; e = e->next)
|
331
|
+
{
|
332
|
+
insert_connector(ct, e);
|
333
|
+
}
|
334
|
+
d = d1; /* move on to next disjunct*/
|
335
|
+
}
|
336
|
+
}
|
337
|
+
sent->word[w].d = fake_head.next;
|
338
|
+
}
|
339
|
+
|
340
|
+
if (2 < verbosity)
|
341
|
+
{
|
342
|
+
printf("l->r pass removed %d\n", N_deleted);
|
343
|
+
print_disjunct_counts(sent);
|
344
|
+
}
|
345
|
+
|
346
|
+
/* We did nothing (and this is not the 1st pass) */
|
347
|
+
if (N_deleted == 0) break;
|
348
|
+
|
349
|
+
/* Right-to-left pass */
|
350
|
+
zero_connector_table(ct);
|
351
|
+
N_deleted = 0;
|
352
|
+
|
353
|
+
/* For every word */
|
354
|
+
for (w = sent->length-1; w >= 0; w--)
|
355
|
+
{
|
356
|
+
d = &fake_head;
|
357
|
+
d->next = sent->word[w].d;
|
358
|
+
|
359
|
+
while ((d1 = d->next))
|
360
|
+
{
|
361
|
+
e = d1->right;
|
362
|
+
|
363
|
+
while (e)
|
364
|
+
{
|
365
|
+
int h = hash_S(e);
|
366
|
+
for (f = ct[h]; f != NULL; f = f->tableNext)
|
367
|
+
{
|
368
|
+
if (prune_match(0, e, f)) break;
|
369
|
+
}
|
370
|
+
if (!f) break; /* If f null, not a single match was found */
|
371
|
+
e = e->next;
|
372
|
+
}
|
373
|
+
|
374
|
+
/* We know this disjunct is dead since it can't match
|
375
|
+
* to the right*/
|
376
|
+
if(e)
|
377
|
+
{
|
378
|
+
N_deleted ++;
|
379
|
+
free_connectors(d1->left);
|
380
|
+
free_connectors(d1->right);
|
381
|
+
d->next = d1->next;
|
382
|
+
xfree(d1, sizeof(Disjunct));
|
383
|
+
}
|
384
|
+
else
|
385
|
+
{
|
386
|
+
/* Store surviving disjunct in hash table */
|
387
|
+
for (e = d1->left; e != NULL; e = e->next)
|
388
|
+
{
|
389
|
+
insert_connector(ct, e);
|
390
|
+
}
|
391
|
+
d = d1; /* move on to next disjunct*/
|
392
|
+
}
|
393
|
+
sent->word[w].d = fake_head.next;
|
394
|
+
}
|
395
|
+
}
|
396
|
+
|
397
|
+
if (verbosity > 2)
|
398
|
+
{
|
399
|
+
printf("r->l pass removed %d\n", N_deleted);
|
400
|
+
print_disjunct_counts(sent);
|
401
|
+
}
|
402
|
+
|
403
|
+
/* We made no change on this pass */
|
404
|
+
if (N_deleted == 0) break;
|
405
|
+
N_deleted = 0;
|
406
|
+
}
|
407
|
+
}
|
408
|
+
|
409
|
+
/*
|
410
|
+
The second algorithm eliminates disjuncts that are dominated by
|
411
|
+
another. It works by hashing them all, and checking for domination.
|
412
|
+
*/
|
413
|
+
|
414
|
+
#if FALSE
|
415
|
+
/* ============================================================x */
|
416
|
+
|
417
|
+
/*
|
418
|
+
Consider the idea of deleting a disjunct if it is dominated (in terms of
|
419
|
+
what it can match) by some other disjunct on the same word. This has
|
420
|
+
been implemented below. There are three problems with it:
|
421
|
+
|
422
|
+
(1) It is almost never the case that any disjuncts are eliminated.
|
423
|
+
(The code below has works correctly with fat links, but because
|
424
|
+
all of the fat connectors on a fat disjunct have the same matching
|
425
|
+
string, the only time a disjuct will die is if it is the same
|
426
|
+
as another one. This is captured by the simplistic version below.
|
427
|
+
|
428
|
+
(2) connector_matches_alam may not be exactly correct. I don't
|
429
|
+
think it does the fat link matches properly. (See the comment
|
430
|
+
in and.c for more information about matching fat links.) This is
|
431
|
+
irrelevant because of (1).
|
432
|
+
|
433
|
+
(3) The linkage that is eliminated by this, might just be the one that
|
434
|
+
passes post-processing, as the following example shows.
|
435
|
+
This is pretty silly, and should probably be changed.
|
436
|
+
|
437
|
+
> telling John how our program works would be stupid
|
438
|
+
Accepted (2 linkages, 1 with no P.P. violations)
|
439
|
+
Linkage 1, cost vector = (0, 0, 7)
|
440
|
+
|
441
|
+
+------------------G-----------------+
|
442
|
+
+-----R-----+----CL----+ |
|
443
|
+
+---O---+ | +---D--+---S---+ +--I-+-AI-+
|
444
|
+
| | | | | | | | |
|
445
|
+
telling.g John how our program.n works would be stupid
|
446
|
+
|
447
|
+
///// CLg <---CLg---> CL telling.g
|
448
|
+
(g) telling.g G <---G-----> G would
|
449
|
+
(g) (d) telling.g R <---R-----> R how
|
450
|
+
(g) (d) telling.g O <---O-----> O John
|
451
|
+
(g) (d) how CLe <---CLe---> CL program.n
|
452
|
+
(g) (d) (e) our D <---Ds----> Ds program.n
|
453
|
+
(g) (d) (e) program.n Ss <---Ss----> Ss works
|
454
|
+
(g) would I <---Ix----> Ix be
|
455
|
+
(g) be AI <---AIi---> AIi stupid
|
456
|
+
|
457
|
+
(press return for another)
|
458
|
+
>
|
459
|
+
Linkage 2 (bad), cost vector = (0, 0, 7)
|
460
|
+
|
461
|
+
+------------------G-----------------+
|
462
|
+
+-----R-----+----CL----+ |
|
463
|
+
+---O---+ | +---D--+---S---+ +--I-+-AI-+
|
464
|
+
| | | | | | | | |
|
465
|
+
telling.g John how our program.n works would be stupid
|
466
|
+
|
467
|
+
///// CLg <---CLg---> CL telling.g
|
468
|
+
(g) telling.g G <---G-----> G would
|
469
|
+
(g) (d) telling.g R <---R-----> R how
|
470
|
+
(g) (d) telling.g O <---O-----> O John
|
471
|
+
(g) (d) how CLe <---CLe---> CL program.n
|
472
|
+
(g) (d) (e) our D <---Ds----> Ds program.n
|
473
|
+
(g) (d) (e) program.n Ss <---Ss----> Ss works
|
474
|
+
(g) would I <---Ix----> Ix be
|
475
|
+
(g) be AI <---AI----> AI stupid
|
476
|
+
|
477
|
+
P.P. violations:
|
478
|
+
Special subject rule violated
|
479
|
+
*/
|
480
|
+
|
481
|
+
/**
|
482
|
+
* hash function that takes a string and a seed value i
|
483
|
+
*/
|
484
|
+
static int string_hash(disjunct_dup_table *dt, const char * s, int i)
|
485
|
+
{
|
486
|
+
for(;*s != '\0';s++) i = i + (i<<1) + randtable[(*s + i) & (RTSIZE-1)];
|
487
|
+
return (i & (dt->dup_table_size-1));
|
488
|
+
}
|
489
|
+
|
490
|
+
/**
|
491
|
+
* This returns true if the connector a matches everything that b
|
492
|
+
* matches, and possibly more. (alam=at least as much)
|
493
|
+
*
|
494
|
+
* TRUE for equal connectors.
|
495
|
+
* remains TRUE if multi-match added to the first.
|
496
|
+
* remains TRUE if subsrcripts deleted from the first.
|
497
|
+
*/
|
498
|
+
int connector_matches_alam(Connector * a, Connector * b)
|
499
|
+
{
|
500
|
+
char * s, * t, *u;
|
501
|
+
if (((!a->multi) && b->multi) ||
|
502
|
+
(a->label != b->label) ||
|
503
|
+
(a->priority != b->priority)) return FALSE;
|
504
|
+
s = a->string;
|
505
|
+
t = b->string;
|
506
|
+
|
507
|
+
/* isupper -- connectors cannot be UTF8 at this time */
|
508
|
+
while(isupper(*s) || isupper(*t))
|
509
|
+
{
|
510
|
+
if (*s == *t) {
|
511
|
+
s++;
|
512
|
+
t++;
|
513
|
+
} else return FALSE;
|
514
|
+
}
|
515
|
+
if (a->priority == DOWN_priority) {
|
516
|
+
u = s;
|
517
|
+
s = t;
|
518
|
+
t = u;
|
519
|
+
}
|
520
|
+
while((*s != '\0') && (*t != '\0')) {
|
521
|
+
if ((*s == *t) || (*s == '*') || (*t == '^')) {
|
522
|
+
s++;
|
523
|
+
t++;
|
524
|
+
} else return FALSE;
|
525
|
+
}
|
526
|
+
while ((*s != '\0') && (*s == '*')) s++;
|
527
|
+
return (*s == '\0');
|
528
|
+
}
|
529
|
+
|
530
|
+
|
531
|
+
/**
|
532
|
+
* This hash function that takes a connector and a seed value i.
|
533
|
+
* It only looks at the leading upper case letters of
|
534
|
+
* the string, and the label. This ensures that if two connectors
|
535
|
+
* match, then they must hash to the same place.
|
536
|
+
*/
|
537
|
+
static int conn_hash(Connector * c, int i)
|
538
|
+
{
|
539
|
+
int nb;
|
540
|
+
const char * s;
|
541
|
+
s = c->string;
|
542
|
+
|
543
|
+
i = i + (i<<1) + randtable[(c->label + i) & (RTSIZE-1)];
|
544
|
+
nb = is_utf8_upper(s);
|
545
|
+
while(nb)
|
546
|
+
{
|
547
|
+
i = i + (i<<1) + randtable[(*s + i) & (RTSIZE-1)];
|
548
|
+
s += nb;
|
549
|
+
nb = is_utf8_upper(s);
|
550
|
+
}
|
551
|
+
return i;
|
552
|
+
}
|
553
|
+
|
554
|
+
static inline int pconnector_hash(disjunct_dup_table *dt, Connector * c, int i)
|
555
|
+
{
|
556
|
+
i = conn_hash(c, i);
|
557
|
+
return (i & (ct->dup_table_size-1));
|
558
|
+
}
|
559
|
+
|
560
|
+
/**
|
561
|
+
* This is a hash function for disjuncts
|
562
|
+
*/
|
563
|
+
static int hash_disjunct(disjunct_dup_table *dt, Disjunct * d)
|
564
|
+
{
|
565
|
+
int i;
|
566
|
+
Connector *e;
|
567
|
+
i = 0;
|
568
|
+
for (e = d->left ; e != NULL; e = e->next)
|
569
|
+
{
|
570
|
+
i = pconnector_hash(dt, e, i);
|
571
|
+
}
|
572
|
+
for (e = d->right ; e != NULL; e = e->next)
|
573
|
+
{
|
574
|
+
i = pconnector_hash(dt, e, i);
|
575
|
+
}
|
576
|
+
return string_hash(dt, d->string, i);
|
577
|
+
}
|
578
|
+
|
579
|
+
/**
|
580
|
+
* Returns TRUE if disjunct d1 can match anything that d2 can
|
581
|
+
* if this happens, it constitutes a proof that there is absolutely
|
582
|
+
* no use for d2.
|
583
|
+
*/
|
584
|
+
static int disjunct_matches_alam(Disjunct * d1, Disjunct * d2)
|
585
|
+
{
|
586
|
+
Connector *e1, *e2;
|
587
|
+
if (d1->cost > d2->cost) return FALSE;
|
588
|
+
e1 = d1->left;
|
589
|
+
e2 = d2->left;
|
590
|
+
while((e1!=NULL) && (e2!=NULL)) {
|
591
|
+
if (!connector_matches_alam(e1,e2)) break;
|
592
|
+
e1 = e1->next;
|
593
|
+
e2 = e2->next;
|
594
|
+
}
|
595
|
+
if ((e1!=NULL) || (e2!=NULL)) return FALSE;
|
596
|
+
e1 = d1->right;
|
597
|
+
e2 = d2->right;
|
598
|
+
while((e1!=NULL) && (e2!=NULL)) {
|
599
|
+
if (!connector_matches_alam(e1,e2)) break;
|
600
|
+
e1 = e1->next;
|
601
|
+
e2 = e2->next;
|
602
|
+
}
|
603
|
+
if ((e1!=NULL) || (e2!=NULL)) return FALSE;
|
604
|
+
return (strcmp(d1->string, d2->string) == 0);
|
605
|
+
}
|
606
|
+
|
607
|
+
/**
|
608
|
+
* Takes the list of disjuncts pointed to by d, eliminates all
|
609
|
+
* duplicates, and returns a pointer to a new list.
|
610
|
+
* It frees the disjuncts that are eliminated.
|
611
|
+
*/
|
612
|
+
Disjunct * eliminate_duplicate_disjuncts(Disjunct * d)
|
613
|
+
{
|
614
|
+
int i, h, count;
|
615
|
+
Disjunct *dn, *dx, *dxn, *front;
|
616
|
+
count = 0;
|
617
|
+
disjunct_dup_table *dt;
|
618
|
+
|
619
|
+
dt = disjunct_dup_table_new(next_power_of_two_up(2 * count_disjuncts(d)));
|
620
|
+
|
621
|
+
for (;d!=NULL; d = dn)
|
622
|
+
{
|
623
|
+
dn = d->next;
|
624
|
+
h = hash_disjunct(d);
|
625
|
+
|
626
|
+
front = NULL;
|
627
|
+
for (dx = dt->dup_table[h]; dx != NULL; dx = dxn)
|
628
|
+
{
|
629
|
+
dxn = dx->next;
|
630
|
+
if (disjunct_matches_alam(dx,d))
|
631
|
+
{
|
632
|
+
/* we know that d should be killed */
|
633
|
+
d->next = NULL;
|
634
|
+
free_disjuncts(d);
|
635
|
+
count++;
|
636
|
+
front = catenate_disjuncts(front, dx);
|
637
|
+
break;
|
638
|
+
} else if (disjunct_matches_alam(d,dx)) {
|
639
|
+
/* we know that dx should be killed off */
|
640
|
+
dx->next = NULL;
|
641
|
+
free_disjuncts(dx);
|
642
|
+
count++;
|
643
|
+
} else {
|
644
|
+
/* neither should be killed off */
|
645
|
+
dx->next = front;
|
646
|
+
front = dx;
|
647
|
+
}
|
648
|
+
}
|
649
|
+
if (dx == NULL) {
|
650
|
+
/* we put d in the table */
|
651
|
+
d->next = front;
|
652
|
+
front = d;
|
653
|
+
}
|
654
|
+
dt->dup_table[h] = front;
|
655
|
+
}
|
656
|
+
|
657
|
+
/* d is now NULL */
|
658
|
+
for (i = 0; i < dt->dup_table_size; i++)
|
659
|
+
{
|
660
|
+
for (dx = dt->dup_table[i]; dx != NULL; dx = dxn)
|
661
|
+
{
|
662
|
+
dxn = dx->next;
|
663
|
+
dx->next = d;
|
664
|
+
d = dx;
|
665
|
+
}
|
666
|
+
}
|
667
|
+
|
668
|
+
if ((verbosity > 2) && (count != 0)) printf("killed %d duplicates\n", count);
|
669
|
+
|
670
|
+
disjunct_dup_table_delete(dt);
|
671
|
+
return d;
|
672
|
+
}
|
673
|
+
|
674
|
+
/* ============================================================x */
|
675
|
+
#endif
|
676
|
+
|
677
|
+
/* ================================================================= */
|
678
|
+
/**
|
679
|
+
* Here is expression pruning. This is done even before the expressions
|
680
|
+
* are turned into lists of disjuncts.
|
681
|
+
*
|
682
|
+
* This uses many of the same data structures and functions that are used
|
683
|
+
* by prune.
|
684
|
+
*
|
685
|
+
* The purge operations remove all irrelevant stuff from the expression,
|
686
|
+
* and free the purged stuff. A connector is deemed irrelevant if its
|
687
|
+
* string pointer has been set to NULL. The passes through the sentence
|
688
|
+
* have the job of doing this.
|
689
|
+
*
|
690
|
+
* If an OR or AND type expression node has one child, we can replace it
|
691
|
+
* by its child. This, of course, is not really necessary, except for
|
692
|
+
* performance(?)
|
693
|
+
*/
|
694
|
+
|
695
|
+
static Exp* purge_Exp(Exp *);
|
696
|
+
|
697
|
+
/**
|
698
|
+
* Get rid of the elements with null expressions
|
699
|
+
*/
|
700
|
+
static E_list * or_purge_E_list(E_list * l)
|
701
|
+
{
|
702
|
+
E_list * el;
|
703
|
+
if (l == NULL) return NULL;
|
704
|
+
if ((l->e = purge_Exp(l->e)) == NULL)
|
705
|
+
{
|
706
|
+
el = or_purge_E_list(l->next);
|
707
|
+
xfree((char *)l, sizeof(E_list));
|
708
|
+
return el;
|
709
|
+
}
|
710
|
+
l->next = or_purge_E_list(l->next);
|
711
|
+
return l;
|
712
|
+
}
|
713
|
+
|
714
|
+
/**
|
715
|
+
* Returns 0 iff the length of the disjunct list is 0.
|
716
|
+
* If this is the case, it frees the structure rooted at l.
|
717
|
+
*/
|
718
|
+
static int and_purge_E_list(E_list * l)
|
719
|
+
{
|
720
|
+
if (l == NULL) return 1;
|
721
|
+
if ((l->e = purge_Exp(l->e)) == NULL)
|
722
|
+
{
|
723
|
+
free_E_list(l->next);
|
724
|
+
xfree((char *)l, sizeof(E_list));
|
725
|
+
return 0;
|
726
|
+
}
|
727
|
+
if (and_purge_E_list(l->next) == 0)
|
728
|
+
{
|
729
|
+
free_Exp(l->e);
|
730
|
+
xfree((char *)l, sizeof(E_list));
|
731
|
+
return 0;
|
732
|
+
}
|
733
|
+
return 1;
|
734
|
+
}
|
735
|
+
|
736
|
+
/**
|
737
|
+
* Must be called with a non-null expression.
|
738
|
+
* Return NULL iff the expression has no disjuncts.
|
739
|
+
*/
|
740
|
+
static Exp* purge_Exp(Exp *e)
|
741
|
+
{
|
742
|
+
if (e->type == CONNECTOR_type)
|
743
|
+
{
|
744
|
+
if (e->u.string == NULL)
|
745
|
+
{
|
746
|
+
xfree((char *)e, sizeof(Exp));
|
747
|
+
return NULL;
|
748
|
+
}
|
749
|
+
else
|
750
|
+
{
|
751
|
+
return e;
|
752
|
+
}
|
753
|
+
}
|
754
|
+
if (e->type == AND_type)
|
755
|
+
{
|
756
|
+
if (and_purge_E_list(e->u.l) == 0)
|
757
|
+
{
|
758
|
+
xfree((char *)e, sizeof(Exp));
|
759
|
+
return NULL;
|
760
|
+
}
|
761
|
+
}
|
762
|
+
else
|
763
|
+
{
|
764
|
+
e->u.l = or_purge_E_list(e->u.l);
|
765
|
+
if (e->u.l == NULL)
|
766
|
+
{
|
767
|
+
xfree((char *)e, sizeof(Exp));
|
768
|
+
return NULL;
|
769
|
+
}
|
770
|
+
}
|
771
|
+
|
772
|
+
/* This code makes it kill off nodes that have just one child
|
773
|
+
(1) It's going to give an insignificant speed-up
|
774
|
+
(2) Costs have not been handled correctly here.
|
775
|
+
The code is excised for these reasons.
|
776
|
+
*/
|
777
|
+
/*
|
778
|
+
if ((e->u.l != NULL) && (e->u.l->next == NULL))
|
779
|
+
{
|
780
|
+
ne = e->u.l->e;
|
781
|
+
xfree((char *) e->u.l, sizeof(E_list));
|
782
|
+
xfree((char *) e, sizeof(Exp));
|
783
|
+
return ne;
|
784
|
+
}
|
785
|
+
*/
|
786
|
+
return e;
|
787
|
+
}
|
788
|
+
|
789
|
+
/**
|
790
|
+
* Returns TRUE if c can match anything in the set S.
|
791
|
+
*/
|
792
|
+
static inline int matches_S(connector_table *ct, Connector * c, int dir)
|
793
|
+
{
|
794
|
+
Connector * e;
|
795
|
+
int h = hash_S(c);
|
796
|
+
|
797
|
+
if (dir == '-')
|
798
|
+
{
|
799
|
+
for (e = ct[h]; e != NULL; e = e->tableNext)
|
800
|
+
{
|
801
|
+
if (prune_match(0, e, c)) return TRUE;
|
802
|
+
}
|
803
|
+
return FALSE;
|
804
|
+
}
|
805
|
+
else
|
806
|
+
{
|
807
|
+
for (e = ct[h]; e != NULL; e = e->tableNext)
|
808
|
+
{
|
809
|
+
if (prune_match(0, c, e)) return TRUE;
|
810
|
+
}
|
811
|
+
return FALSE;
|
812
|
+
}
|
813
|
+
}
|
814
|
+
|
815
|
+
/**
|
816
|
+
* Mark as dead all of the dir-pointing connectors
|
817
|
+
* in e that are not matched by anything in the current set.
|
818
|
+
* Returns the number of connectors so marked.
|
819
|
+
*/
|
820
|
+
static int mark_dead_connectors(connector_table *ct, Exp * e, int dir)
|
821
|
+
{
|
822
|
+
int count;
|
823
|
+
count = 0;
|
824
|
+
if (e->type == CONNECTOR_type)
|
825
|
+
{
|
826
|
+
if (e->dir == dir)
|
827
|
+
{
|
828
|
+
Connector dummy;
|
829
|
+
init_connector(&dummy);
|
830
|
+
dummy.label = NORMAL_LABEL;
|
831
|
+
dummy.priority = THIN_priority;
|
832
|
+
dummy.string = e->u.string;
|
833
|
+
if (!matches_S(ct, &dummy, dir))
|
834
|
+
{
|
835
|
+
e->u.string = NULL;
|
836
|
+
count++;
|
837
|
+
}
|
838
|
+
}
|
839
|
+
}
|
840
|
+
else
|
841
|
+
{
|
842
|
+
E_list *l;
|
843
|
+
for (l = e->u.l; l != NULL; l = l->next)
|
844
|
+
{
|
845
|
+
count += mark_dead_connectors(ct, l->e, dir);
|
846
|
+
}
|
847
|
+
}
|
848
|
+
return count;
|
849
|
+
}
|
850
|
+
|
851
|
+
/**
|
852
|
+
* Put into the set S all of the dir-pointing connectors still in e.
|
853
|
+
* Return a list of allocated dummy connectors; these will need to be
|
854
|
+
* freed.
|
855
|
+
*/
|
856
|
+
static Connector * insert_connectors(connector_table *ct, Exp * e,
|
857
|
+
Connector *alloc_list, int dir)
|
858
|
+
{
|
859
|
+
if (e->type == CONNECTOR_type)
|
860
|
+
{
|
861
|
+
if (e->dir == dir)
|
862
|
+
{
|
863
|
+
Connector *dummy = connector_new();
|
864
|
+
dummy->string = e->u.string;
|
865
|
+
insert_connector(ct, dummy);
|
866
|
+
dummy->next = alloc_list;
|
867
|
+
alloc_list = dummy;
|
868
|
+
}
|
869
|
+
}
|
870
|
+
else
|
871
|
+
{
|
872
|
+
E_list *l;
|
873
|
+
for (l=e->u.l; l!=NULL; l=l->next)
|
874
|
+
{
|
875
|
+
alloc_list = insert_connectors(ct, l->e, alloc_list, dir);
|
876
|
+
}
|
877
|
+
}
|
878
|
+
return alloc_list;
|
879
|
+
}
|
880
|
+
|
881
|
+
/**
|
882
|
+
* This removes the expressions that are empty from the list corresponding
|
883
|
+
* to word w of the sentence.
|
884
|
+
*/
|
885
|
+
static void clean_up_expressions(Sentence sent, int w)
|
886
|
+
{
|
887
|
+
X_node head_node, *d, *d1;
|
888
|
+
d = &head_node;
|
889
|
+
d->next = sent->word[w].x;
|
890
|
+
while (d->next != NULL)
|
891
|
+
{
|
892
|
+
if (d->next->exp == NULL)
|
893
|
+
{
|
894
|
+
d1 = d->next;
|
895
|
+
d->next = d1->next;
|
896
|
+
xfree((char *)d1, sizeof(X_node));
|
897
|
+
}
|
898
|
+
else
|
899
|
+
{
|
900
|
+
d = d->next;
|
901
|
+
}
|
902
|
+
}
|
903
|
+
sent->word[w].x = head_node.next;
|
904
|
+
}
|
905
|
+
|
906
|
+
void expression_prune(Sentence sent)
|
907
|
+
{
|
908
|
+
int N_deleted;
|
909
|
+
X_node * x;
|
910
|
+
int w;
|
911
|
+
Connector *ct[CONTABSZ];
|
912
|
+
Connector *dummy_list = NULL;
|
913
|
+
|
914
|
+
zero_connector_table(ct);
|
915
|
+
|
916
|
+
N_deleted = 1; /* a lie to make it always do at least 2 passes */
|
917
|
+
|
918
|
+
while(1)
|
919
|
+
{
|
920
|
+
/* Left-to-right pass */
|
921
|
+
/* For every word */
|
922
|
+
for (w = 0; w < sent->length; w++)
|
923
|
+
{
|
924
|
+
/* For every expression in word */
|
925
|
+
for (x = sent->word[w].x; x != NULL; x = x->next)
|
926
|
+
{
|
927
|
+
/* printf("before marking: "); print_expression(x->exp); printf("\n"); */
|
928
|
+
N_deleted += mark_dead_connectors(ct, x->exp, '-');
|
929
|
+
/* printf(" after marking: "); print_expression(x->exp); printf("\n"); */
|
930
|
+
}
|
931
|
+
for (x = sent->word[w].x; x != NULL; x = x->next)
|
932
|
+
{
|
933
|
+
/* printf("before purging: "); print_expression(x->exp); printf("\n"); */
|
934
|
+
x->exp = purge_Exp(x->exp);
|
935
|
+
/* printf("after purging: "); print_expression(x->exp); printf("\n"); */
|
936
|
+
}
|
937
|
+
|
938
|
+
/* gets rid of X_nodes with NULL exp */
|
939
|
+
clean_up_expressions(sent, w);
|
940
|
+
for (x = sent->word[w].x; x != NULL; x = x->next)
|
941
|
+
{
|
942
|
+
dummy_list = insert_connectors(ct, x->exp, dummy_list, '+');
|
943
|
+
}
|
944
|
+
}
|
945
|
+
|
946
|
+
if (verbosity > 2)
|
947
|
+
{
|
948
|
+
printf("l->r pass removed %d\n", N_deleted);
|
949
|
+
print_expression_sizes(sent);
|
950
|
+
}
|
951
|
+
|
952
|
+
/* Free the allocated dummy connectors */
|
953
|
+
free_connectors(dummy_list);
|
954
|
+
dummy_list = NULL;
|
955
|
+
zero_connector_table(ct);
|
956
|
+
|
957
|
+
if (N_deleted == 0) break;
|
958
|
+
|
959
|
+
/* Right-to-left pass */
|
960
|
+
N_deleted = 0;
|
961
|
+
for (w = sent->length-1; w >= 0; w--)
|
962
|
+
{
|
963
|
+
for (x = sent->word[w].x; x != NULL; x = x->next)
|
964
|
+
{
|
965
|
+
/* printf("before marking: "); print_expression(x->exp); printf("\n"); */
|
966
|
+
N_deleted += mark_dead_connectors(ct, x->exp, '+');
|
967
|
+
/* printf("after marking: "); print_expression(x->exp); printf("\n"); */
|
968
|
+
}
|
969
|
+
for (x = sent->word[w].x; x != NULL; x = x->next)
|
970
|
+
{
|
971
|
+
/* printf("before perging: "); print_expression(x->exp); printf("\n"); */
|
972
|
+
x->exp = purge_Exp(x->exp);
|
973
|
+
/* printf("after perging: "); print_expression(x->exp); printf("\n"); */
|
974
|
+
}
|
975
|
+
clean_up_expressions(sent, w); /* gets rid of X_nodes with NULL exp */
|
976
|
+
for (x = sent->word[w].x; x != NULL; x = x->next)
|
977
|
+
{
|
978
|
+
dummy_list = insert_connectors(ct, x->exp, dummy_list, '-');
|
979
|
+
}
|
980
|
+
}
|
981
|
+
|
982
|
+
if (verbosity > 2)
|
983
|
+
{
|
984
|
+
printf("r->l pass removed %d\n", N_deleted);
|
985
|
+
print_expression_sizes(sent);
|
986
|
+
}
|
987
|
+
|
988
|
+
/* Free the allocated dummy connectors */
|
989
|
+
free_connectors(dummy_list);
|
990
|
+
dummy_list = NULL;
|
991
|
+
zero_connector_table(ct);
|
992
|
+
if (N_deleted == 0) break;
|
993
|
+
N_deleted = 0;
|
994
|
+
}
|
995
|
+
}
|
996
|
+
|
997
|
+
|
998
|
+
|
999
|
+
/*
|
1000
|
+
Here is what you've been waiting for: POWER-PRUNE
|
1001
|
+
|
1002
|
+
The kinds of constraints it checks for are the following:
|
1003
|
+
|
1004
|
+
1) successive connectors on the same disjunct have to go to
|
1005
|
+
nearer and nearer words.
|
1006
|
+
|
1007
|
+
2) two deep connectors cannot attach to eachother
|
1008
|
+
(A connectors is deep if it is not the first in its list, it
|
1009
|
+
is shallow if it is the first in its list, it is deepest if it
|
1010
|
+
is the last on its list.)
|
1011
|
+
|
1012
|
+
3) on two adjacent words, a pair of connectors can be used
|
1013
|
+
only if they're the deepest ones on their disjuncts
|
1014
|
+
|
1015
|
+
4) on two non-adjacent words, a pair of connectors can be used only
|
1016
|
+
if not [both of them are the deepest].
|
1017
|
+
|
1018
|
+
The data structure consists of a pair of hash tables on every word.
|
1019
|
+
Each bucket of a hash table has a list of pointers to connectors.
|
1020
|
+
These nodes also store if the chosen connector is shallow.
|
1021
|
+
*/
|
1022
|
+
/*
|
1023
|
+
As with normal pruning, we make alternate left->right and right->left
|
1024
|
+
passes. In the R->L pass, when we're on a word w, we make use of
|
1025
|
+
all the left-pointing hash tables on the words to the right of w.
|
1026
|
+
After the pruning on this word, we build the left-pointing hash table
|
1027
|
+
this word. This guarantees idempotence of the pass -- after doing an
|
1028
|
+
L->R, doing another would change nothing.
|
1029
|
+
|
1030
|
+
Each connector has an integer c_word field. This refers to the closest
|
1031
|
+
word that it could be connected to. These are initially determined by
|
1032
|
+
how deep the connector is. For example, a deepest connector can connect
|
1033
|
+
to the neighboring word, so its c_word field is w+1 (w-1 if this is a left
|
1034
|
+
pointing connector). It's neighboring shallow connector has a c_word
|
1035
|
+
value of w+2, etc.
|
1036
|
+
|
1037
|
+
The pruning process adjusts these c_word values as it goes along,
|
1038
|
+
accumulating information about any way of linking this sentence.
|
1039
|
+
The pruning process stops only after no disjunct is deleted and no
|
1040
|
+
c_word values change.
|
1041
|
+
|
1042
|
+
The difference between RUTHLESS and GENTLE power pruning is simply
|
1043
|
+
that GENTLE uses the deletable region array, and RUTHLESS does not.
|
1044
|
+
So we can get the effect of these two different methods simply by
|
1045
|
+
always unsuring that deletable[][] has been defined. With nothing
|
1046
|
+
deletable, this is equivalent to RUTHLESS. --DS, 7/97
|
1047
|
+
*/
|
1048
|
+
|
1049
|
+
/**
|
1050
|
+
* returns the number of connectors in the left lists of the disjuncts.
|
1051
|
+
*/
|
1052
|
+
static int left_connector_count(Disjunct * d)
|
1053
|
+
{
|
1054
|
+
Connector *c;
|
1055
|
+
int i=0;
|
1056
|
+
for (;d!=NULL; d=d->next) {
|
1057
|
+
for (c = d->left; c!=NULL; c = c->next) i++;
|
1058
|
+
}
|
1059
|
+
return i;
|
1060
|
+
}
|
1061
|
+
|
1062
|
+
static int right_connector_count(Disjunct * d)
|
1063
|
+
{
|
1064
|
+
Connector *c;
|
1065
|
+
int i=0;
|
1066
|
+
for (;d!=NULL; d=d->next) {
|
1067
|
+
for (c = d->right; c!=NULL; c = c->next) i++;
|
1068
|
+
}
|
1069
|
+
return i;
|
1070
|
+
}
|
1071
|
+
|
1072
|
+
static void free_C_list(C_list * t)
|
1073
|
+
{
|
1074
|
+
C_list *xt;
|
1075
|
+
for (; t!=NULL; t=xt) {
|
1076
|
+
xt = t->next;
|
1077
|
+
xfree((char *)t, sizeof(C_list));
|
1078
|
+
}
|
1079
|
+
}
|
1080
|
+
|
1081
|
+
/**
|
1082
|
+
* free all of the hash tables and C_lists
|
1083
|
+
*/
|
1084
|
+
static void power_table_delete(power_table *pt)
|
1085
|
+
{
|
1086
|
+
int w;
|
1087
|
+
int i;
|
1088
|
+
|
1089
|
+
for (w = 0; w < pt->power_table_size; w++)
|
1090
|
+
{
|
1091
|
+
for (i = 0; i < pt->l_table_size[w]; i++)
|
1092
|
+
{
|
1093
|
+
free_C_list(pt->l_table[w][i]);
|
1094
|
+
}
|
1095
|
+
xfree((char *)pt->l_table[w], pt->l_table_size[w] * sizeof (C_list *));
|
1096
|
+
|
1097
|
+
for (i = 0; i < pt->r_table_size[w]; i++)
|
1098
|
+
{
|
1099
|
+
free_C_list(pt->r_table[w][i]);
|
1100
|
+
}
|
1101
|
+
xfree((char *)pt->r_table[w], pt->r_table_size[w] * sizeof (C_list *));
|
1102
|
+
}
|
1103
|
+
free(pt);
|
1104
|
+
}
|
1105
|
+
|
1106
|
+
/**
|
1107
|
+
* The disjunct d (whose left or right pointer points to c) is put
|
1108
|
+
* into the appropriate hash table
|
1109
|
+
*/
|
1110
|
+
static void put_into_power_table(int size, C_list ** t, Connector * c, int shal)
|
1111
|
+
{
|
1112
|
+
int h;
|
1113
|
+
C_list * m;
|
1114
|
+
h = connector_hash(c) & (size-1);
|
1115
|
+
m = (C_list *) xalloc (sizeof(C_list));
|
1116
|
+
m->next = t[h];
|
1117
|
+
t[h] = m;
|
1118
|
+
m->c = c;
|
1119
|
+
m->shallow = shal;
|
1120
|
+
}
|
1121
|
+
|
1122
|
+
static int set_dist_fields(Connector * c, int w, int delta)
|
1123
|
+
{
|
1124
|
+
int i;
|
1125
|
+
if (c==NULL) return w;
|
1126
|
+
i = set_dist_fields(c->next, w, delta) + delta;
|
1127
|
+
c->word = i;
|
1128
|
+
return i;
|
1129
|
+
}
|
1130
|
+
|
1131
|
+
/**
|
1132
|
+
* Allocates and builds the initial power hash tables
|
1133
|
+
*/
|
1134
|
+
static power_table * power_table_new(Sentence sent)
|
1135
|
+
{
|
1136
|
+
power_table *pt;
|
1137
|
+
int w, len, size, i;
|
1138
|
+
C_list ** t;
|
1139
|
+
Disjunct * d, * xd, * head;
|
1140
|
+
Connector * c;
|
1141
|
+
|
1142
|
+
pt = (power_table *) malloc (sizeof(power_table));
|
1143
|
+
pt->power_table_size = sent->length;
|
1144
|
+
|
1145
|
+
/* first we initialize the word fields of the connectors, and
|
1146
|
+
eliminate those disjuncts with illegal connectors */
|
1147
|
+
for (w=0; w<sent->length; w++)
|
1148
|
+
{
|
1149
|
+
head = NULL;
|
1150
|
+
for (d=sent->word[w].d; d!=NULL; d=xd) {
|
1151
|
+
xd = d->next;
|
1152
|
+
if ((set_dist_fields(d->left, w, -1) < 0) ||
|
1153
|
+
(set_dist_fields(d->right, w, 1) >= sent->length)) {
|
1154
|
+
d->next = NULL;
|
1155
|
+
free_disjuncts(d);
|
1156
|
+
} else {
|
1157
|
+
d->next = head;
|
1158
|
+
head = d;
|
1159
|
+
}
|
1160
|
+
}
|
1161
|
+
sent->word[w].d = head;
|
1162
|
+
}
|
1163
|
+
|
1164
|
+
for (w=0; w<sent->length; w++)
|
1165
|
+
{
|
1166
|
+
len = left_connector_count(sent->word[w].d);
|
1167
|
+
size = next_power_of_two_up(len);
|
1168
|
+
pt->l_table_size[w] = size;
|
1169
|
+
t = pt->l_table[w] = (C_list **) xalloc(size * sizeof(C_list *));
|
1170
|
+
for (i=0; i<size; i++) t[i] = NULL;
|
1171
|
+
|
1172
|
+
for (d=sent->word[w].d; d!=NULL; d=d->next) {
|
1173
|
+
c = d->left;
|
1174
|
+
if (c != NULL) {
|
1175
|
+
put_into_power_table(size, t, c, TRUE);
|
1176
|
+
for (c=c->next; c!=NULL; c=c->next){
|
1177
|
+
put_into_power_table(size, t, c, FALSE);
|
1178
|
+
}
|
1179
|
+
}
|
1180
|
+
}
|
1181
|
+
|
1182
|
+
len = right_connector_count(sent->word[w].d);
|
1183
|
+
size = next_power_of_two_up(len);
|
1184
|
+
pt->r_table_size[w] = size;
|
1185
|
+
t = pt->r_table[w] = (C_list **) xalloc(size * sizeof(C_list *));
|
1186
|
+
for (i=0; i<size; i++) t[i] = NULL;
|
1187
|
+
|
1188
|
+
for (d=sent->word[w].d; d!=NULL; d=d->next) {
|
1189
|
+
c = d->right;
|
1190
|
+
if (c != NULL) {
|
1191
|
+
put_into_power_table(size, t, c, TRUE);
|
1192
|
+
for (c=c->next; c!=NULL; c=c->next){
|
1193
|
+
put_into_power_table(size, t, c, FALSE);
|
1194
|
+
}
|
1195
|
+
}
|
1196
|
+
}
|
1197
|
+
}
|
1198
|
+
|
1199
|
+
return pt;
|
1200
|
+
}
|
1201
|
+
|
1202
|
+
/**
|
1203
|
+
* This runs through all the connectors in this table, and eliminates those
|
1204
|
+
* who are obsolete. The word fields of an obsolete one has been set to
|
1205
|
+
* BAD_WORD.
|
1206
|
+
*/
|
1207
|
+
static void clean_table(int size, C_list ** t)
|
1208
|
+
{
|
1209
|
+
int i;
|
1210
|
+
C_list * m, * xm, * head;
|
1211
|
+
for (i=0; i<size; i++) {
|
1212
|
+
head = NULL;
|
1213
|
+
for (m=t[i]; m!=NULL; m=xm) {
|
1214
|
+
xm = m->next;
|
1215
|
+
if (m->c->word != BAD_WORD) {
|
1216
|
+
m->next = head;
|
1217
|
+
head = m;
|
1218
|
+
} else {
|
1219
|
+
xfree((char *) m, sizeof(C_list));
|
1220
|
+
}
|
1221
|
+
}
|
1222
|
+
t[i] = head;
|
1223
|
+
}
|
1224
|
+
}
|
1225
|
+
|
1226
|
+
/**
|
1227
|
+
* This takes two connectors (and whether these are shallow or not)
|
1228
|
+
* (and the two words that these came from) and returns TRUE if it is
|
1229
|
+
* possible for these two to match based on local considerations.
|
1230
|
+
*/
|
1231
|
+
static int possible_connection(prune_context *pc,
|
1232
|
+
Connector *lc, Connector *rc,
|
1233
|
+
int lshallow, int rshallow,
|
1234
|
+
int lword, int rword)
|
1235
|
+
{
|
1236
|
+
if ((!lshallow) && (!rshallow)) return FALSE;
|
1237
|
+
/* two deep connectors can't work */
|
1238
|
+
if ((lc->word > rword) || (rc->word < lword)) return FALSE;
|
1239
|
+
/* word range constraints */
|
1240
|
+
|
1241
|
+
assert(lword < rword, "Bad word order in possible connection.");
|
1242
|
+
|
1243
|
+
/* Now, notice that the only differences between the following two
|
1244
|
+
cases is that (1) ruthless uses match and gentle uses prune_match.
|
1245
|
+
and (2) ruthless doesn't use deletable[][]. This latter fact is
|
1246
|
+
irrelevant, since deletable[][] is now guaranteed to have been
|
1247
|
+
created. */
|
1248
|
+
|
1249
|
+
if (pc->power_prune_mode == RUTHLESS) {
|
1250
|
+
if (lword == rword-1) {
|
1251
|
+
if (!((lc->next == NULL) && (rc->next == NULL))) return FALSE;
|
1252
|
+
} else {
|
1253
|
+
if ((!pc->null_links) &&
|
1254
|
+
(lc->next == NULL) && (rc->next == NULL) && (!lc->multi) && (!rc->multi)) {
|
1255
|
+
return FALSE;
|
1256
|
+
}
|
1257
|
+
}
|
1258
|
+
return do_match(pc->sent, lc, rc, lword, rword);
|
1259
|
+
} else {
|
1260
|
+
if (lword == rword-1) {
|
1261
|
+
if (!((lc->next == NULL) && (rc->next == NULL))) return FALSE;
|
1262
|
+
} else {
|
1263
|
+
if ((!pc->null_links) &&
|
1264
|
+
(lc->next == NULL) && (rc->next == NULL) && (!lc->multi) && (!rc->multi) &&
|
1265
|
+
!pc->deletable[lword][rword]) {
|
1266
|
+
return FALSE;
|
1267
|
+
}
|
1268
|
+
}
|
1269
|
+
return prune_match(pc->effective_dist[lword][rword], lc, rc);
|
1270
|
+
}
|
1271
|
+
}
|
1272
|
+
|
1273
|
+
|
1274
|
+
/**
|
1275
|
+
* This returns TRUE if the right table of word w contains
|
1276
|
+
* a connector that can match to c. shallow tells if c is shallow.
|
1277
|
+
*/
|
1278
|
+
static int right_table_search(prune_context *pc, int w, Connector *c, int shallow, int word_c)
|
1279
|
+
{
|
1280
|
+
int size, h;
|
1281
|
+
C_list *cl;
|
1282
|
+
power_table *pt;
|
1283
|
+
|
1284
|
+
pt = pc->pt;
|
1285
|
+
size = pt->r_table_size[w];
|
1286
|
+
h = connector_hash(c) & (size-1);
|
1287
|
+
for (cl = pt->r_table[w][h]; cl != NULL; cl = cl->next)
|
1288
|
+
{
|
1289
|
+
if (possible_connection(pc, cl->c, c, cl->shallow, shallow, w, word_c))
|
1290
|
+
{
|
1291
|
+
return TRUE;
|
1292
|
+
}
|
1293
|
+
}
|
1294
|
+
return FALSE;
|
1295
|
+
}
|
1296
|
+
|
1297
|
+
/**
|
1298
|
+
* This returns TRUE if the right table of word w contains
|
1299
|
+
* a connector that can match to c. shallows tells if c is shallow
|
1300
|
+
*/
|
1301
|
+
static int left_table_search(prune_context *pc, int w, Connector *c, int shallow, int word_c)
|
1302
|
+
{
|
1303
|
+
int size, h;
|
1304
|
+
C_list *cl;
|
1305
|
+
power_table *pt;
|
1306
|
+
|
1307
|
+
pt = pc->pt;
|
1308
|
+
size = pt->l_table_size[w];
|
1309
|
+
h = connector_hash(c) & (size-1);
|
1310
|
+
for (cl = pt->l_table[w][h]; cl != NULL; cl = cl->next)
|
1311
|
+
{
|
1312
|
+
if (possible_connection(pc, c, cl->c, shallow, cl->shallow, word_c, w))
|
1313
|
+
{
|
1314
|
+
return TRUE;
|
1315
|
+
}
|
1316
|
+
}
|
1317
|
+
return FALSE;
|
1318
|
+
}
|
1319
|
+
|
1320
|
+
#if NOT_USED_NOW
|
1321
|
+
static int ok_cwords(Sentence sent, Connector *c)
|
1322
|
+
{
|
1323
|
+
for (; c != NULL; c=c->next) {
|
1324
|
+
if (c->word == BAD_WORD) return FALSE;
|
1325
|
+
if (c->word >= sent->length) return FALSE;
|
1326
|
+
}
|
1327
|
+
return TRUE;
|
1328
|
+
}
|
1329
|
+
#endif
|
1330
|
+
|
1331
|
+
/**
|
1332
|
+
* take this connector list, and try to match it with the words
|
1333
|
+
* w-1, w-2, w-3...Returns the word to which the first connector of the
|
1334
|
+
* list could possibly be matched. If c is NULL, returns w. If there
|
1335
|
+
* is no way to match this list, it returns a negative number.
|
1336
|
+
* If it does find a way to match it, it updates the c->word fields
|
1337
|
+
* correctly.
|
1338
|
+
*/
|
1339
|
+
static int left_connector_list_update(prune_context *pc, Connector *c, int word_c, int w, int shallow)
|
1340
|
+
{
|
1341
|
+
int n;
|
1342
|
+
int foundmatch;
|
1343
|
+
|
1344
|
+
if (c==NULL) return w;
|
1345
|
+
n = left_connector_list_update(pc, c->next, word_c, w, FALSE) - 1;
|
1346
|
+
if (((int) c->word) < n) n = c->word;
|
1347
|
+
|
1348
|
+
/* n is now the rightmost word we need to check */
|
1349
|
+
foundmatch = FALSE;
|
1350
|
+
for (; (n >= 0) && ((w-n) < MAX_SENTENCE); n--) {
|
1351
|
+
pc->power_cost++;
|
1352
|
+
if (right_table_search(pc, n, c, shallow, word_c)) {
|
1353
|
+
foundmatch = TRUE;
|
1354
|
+
break;
|
1355
|
+
}
|
1356
|
+
}
|
1357
|
+
if (n < ((int) c->word)) {
|
1358
|
+
c->word = n;
|
1359
|
+
pc->N_changed++;
|
1360
|
+
}
|
1361
|
+
return (foundmatch ? n : -1);
|
1362
|
+
}
|
1363
|
+
|
1364
|
+
/**
|
1365
|
+
* take this connector list, and try to match it with the words
|
1366
|
+
* w+1, w+2, w+3...Returns the word to which the first connector of the
|
1367
|
+
* list could possibly be matched. If c is NULL, returns w. If there
|
1368
|
+
* is no way to match this list, it returns a number greater than N_words-1
|
1369
|
+
* If it does find a way to match it, it updates the c->word fields
|
1370
|
+
* correctly.
|
1371
|
+
*/
|
1372
|
+
static int right_connector_list_update(prune_context *pc, Sentence sent, Connector *c,
|
1373
|
+
int word_c, int w, int shallow)
|
1374
|
+
{
|
1375
|
+
int n;
|
1376
|
+
int foundmatch;
|
1377
|
+
|
1378
|
+
if (c==NULL) return w;
|
1379
|
+
n = right_connector_list_update(pc, sent, c->next, word_c, w, FALSE) + 1;
|
1380
|
+
if (c->word > n) n = c->word;
|
1381
|
+
|
1382
|
+
/* n is now the leftmost word we need to check */
|
1383
|
+
foundmatch = FALSE;
|
1384
|
+
for (; (n < sent->length) && ((n-w) < MAX_SENTENCE); n++) {
|
1385
|
+
pc->power_cost++;
|
1386
|
+
if (left_table_search(pc, n, c, shallow, word_c)) {
|
1387
|
+
foundmatch = TRUE;
|
1388
|
+
break;
|
1389
|
+
}
|
1390
|
+
}
|
1391
|
+
if (n > c->word) {
|
1392
|
+
c->word = n;
|
1393
|
+
pc->N_changed++;
|
1394
|
+
}
|
1395
|
+
return (foundmatch ? n : sent->length);
|
1396
|
+
}
|
1397
|
+
|
1398
|
+
/** The return value is the number of disjuncts deleted */
|
1399
|
+
int power_prune(Sentence sent, int mode, Parse_Options opts)
|
1400
|
+
{
|
1401
|
+
power_table *pt;
|
1402
|
+
prune_context *pc;
|
1403
|
+
Disjunct *d, *free_later, *dx, *nd;
|
1404
|
+
Connector *c;
|
1405
|
+
int w, N_deleted, total_deleted;
|
1406
|
+
|
1407
|
+
pc = (prune_context *) malloc (sizeof(prune_context));
|
1408
|
+
pc->power_cost = 0;
|
1409
|
+
pc->power_prune_mode = mode;
|
1410
|
+
pc->null_links = (opts->min_null_count > 0);
|
1411
|
+
pc->N_changed = 1; /* forces it always to make at least two passes */
|
1412
|
+
pc->deletable = sent->deletable;
|
1413
|
+
pc->effective_dist = sent->effective_dist;
|
1414
|
+
pc->sent = sent;
|
1415
|
+
|
1416
|
+
count_set_effective_distance(sent);
|
1417
|
+
|
1418
|
+
pt = power_table_new(sent);
|
1419
|
+
pc->pt = pt;
|
1420
|
+
|
1421
|
+
free_later = NULL;
|
1422
|
+
N_deleted = 0;
|
1423
|
+
|
1424
|
+
total_deleted = 0;
|
1425
|
+
|
1426
|
+
while (1)
|
1427
|
+
{
|
1428
|
+
/* left-to-right pass */
|
1429
|
+
for (w = 0; w < sent->length; w++) {
|
1430
|
+
if (parse_options_resources_exhausted(opts)) break;
|
1431
|
+
for (d = sent->word[w].d; d != NULL; d = d->next) {
|
1432
|
+
if (d->left == NULL) continue;
|
1433
|
+
if (left_connector_list_update(pc, d->left, w, w, TRUE) < 0) {
|
1434
|
+
for (c=d->left ;c!=NULL; c = c->next) c->word = BAD_WORD;
|
1435
|
+
for (c=d->right ;c!=NULL; c = c->next) c->word = BAD_WORD;
|
1436
|
+
N_deleted++;
|
1437
|
+
total_deleted++;
|
1438
|
+
}
|
1439
|
+
}
|
1440
|
+
|
1441
|
+
clean_table(pt->r_table_size[w], pt->r_table[w]);
|
1442
|
+
nd = NULL;
|
1443
|
+
for (d = sent->word[w].d; d != NULL; d = dx) {
|
1444
|
+
dx = d->next;
|
1445
|
+
if ((d->left != NULL) && (d->left->word == BAD_WORD)) {
|
1446
|
+
d->next = free_later;
|
1447
|
+
free_later = d;
|
1448
|
+
} else {
|
1449
|
+
d->next = nd;
|
1450
|
+
nd = d;
|
1451
|
+
}
|
1452
|
+
}
|
1453
|
+
sent->word[w].d = nd;
|
1454
|
+
}
|
1455
|
+
if (verbosity > 2) {
|
1456
|
+
printf("l->r pass changed %d and deleted %d\n",pc->N_changed,N_deleted);
|
1457
|
+
}
|
1458
|
+
|
1459
|
+
if (pc->N_changed == 0) break;
|
1460
|
+
|
1461
|
+
pc->N_changed = N_deleted = 0;
|
1462
|
+
/* right-to-left pass */
|
1463
|
+
|
1464
|
+
for (w = sent->length-1; w >= 0; w--) {
|
1465
|
+
if (parse_options_resources_exhausted(opts)) break;
|
1466
|
+
for (d = sent->word[w].d; d != NULL; d = d->next) {
|
1467
|
+
if (d->right == NULL) continue;
|
1468
|
+
if (right_connector_list_update(pc, sent, d->right,w,w,TRUE) >= sent->length){
|
1469
|
+
for (c=d->right;c!=NULL; c = c->next) c->word = BAD_WORD;
|
1470
|
+
for (c=d->left ;c!=NULL; c = c->next) c->word = BAD_WORD;
|
1471
|
+
N_deleted++;
|
1472
|
+
total_deleted++;
|
1473
|
+
}
|
1474
|
+
}
|
1475
|
+
clean_table(pt->l_table_size[w], pt->l_table[w]);
|
1476
|
+
nd = NULL;
|
1477
|
+
for (d = sent->word[w].d; d != NULL; d = dx) {
|
1478
|
+
dx = d->next;
|
1479
|
+
if ((d->right != NULL) && (d->right->word == BAD_WORD)) {
|
1480
|
+
d->next = free_later;
|
1481
|
+
free_later = d;
|
1482
|
+
} else {
|
1483
|
+
d->next = nd;
|
1484
|
+
nd = d;
|
1485
|
+
}
|
1486
|
+
}
|
1487
|
+
sent->word[w].d = nd;
|
1488
|
+
}
|
1489
|
+
|
1490
|
+
if (verbosity > 2) {
|
1491
|
+
printf("r->l pass changed %d and deleted %d\n", pc->N_changed,N_deleted);
|
1492
|
+
}
|
1493
|
+
|
1494
|
+
if (pc->N_changed == 0) break;
|
1495
|
+
pc->N_changed = N_deleted = 0;
|
1496
|
+
}
|
1497
|
+
free_disjuncts(free_later);
|
1498
|
+
power_table_delete(pt);
|
1499
|
+
pt = NULL;
|
1500
|
+
pc->pt = NULL;
|
1501
|
+
|
1502
|
+
if (verbosity > 2) printf("%d power prune cost:\n", pc->power_cost);
|
1503
|
+
|
1504
|
+
if (mode == RUTHLESS) {
|
1505
|
+
print_time(opts, "power pruned (ruthless)");
|
1506
|
+
} else {
|
1507
|
+
print_time(opts, "power pruned (gentle)");
|
1508
|
+
}
|
1509
|
+
|
1510
|
+
if (verbosity > 2) {
|
1511
|
+
if (mode == RUTHLESS) {
|
1512
|
+
printf("\nAfter power_pruning (ruthless):\n");
|
1513
|
+
} else {
|
1514
|
+
printf("\nAfter power_pruning (gentle):\n");
|
1515
|
+
}
|
1516
|
+
print_disjunct_counts(sent);
|
1517
|
+
}
|
1518
|
+
|
1519
|
+
free(pc);
|
1520
|
+
return total_deleted;
|
1521
|
+
}
|
1522
|
+
|
1523
|
+
/* ===================================================================
|
1524
|
+
PP Pruning
|
1525
|
+
|
1526
|
+
The "contains one" post-processing constraints give us a new way to
|
1527
|
+
prune. Suppose there's a rule that says "a group that contains foo
|
1528
|
+
must contain a bar or a baz." Here foo, bar, and baz are connector
|
1529
|
+
types. foo is the trigger link, bar and baz are called the criterion
|
1530
|
+
links. If, after considering the disjuncts we find that there is is
|
1531
|
+
a foo, but neither a bar, nor a baz, then we can eliminte the disjuct
|
1532
|
+
containing bar.
|
1533
|
+
|
1534
|
+
Things are actually a bit more complex, because of the matching rules
|
1535
|
+
and subscripts. The problem is that post-processing deals with link
|
1536
|
+
names, while at this point all we have to work with is connector
|
1537
|
+
names. Consider the foo part. Consider a connector C. When does
|
1538
|
+
foo match C for our purposes? It matches it if every possible link
|
1539
|
+
name L (that can result from C being at one end of that link) results
|
1540
|
+
in post_process_match(foo,L) being true. Suppose foo contains a "*".
|
1541
|
+
Then there is no C that has this property. This is because the *s in
|
1542
|
+
C may be replaced by some other subscripts in the construction of L.
|
1543
|
+
And the non-* chars in L will not post_process_match the * in foo.
|
1544
|
+
|
1545
|
+
So let's assume that foo has no *. Now the result we want is simply
|
1546
|
+
given by post_process_match(foo, C). Proof: L is the same as C but
|
1547
|
+
with some *s replaced by some other letters. Since foo contains no *
|
1548
|
+
the replacement in C of some * by some other letter could change
|
1549
|
+
post_process_match from FALSE to TRUE, but not vice versa. Therefore
|
1550
|
+
it's conservative to use this test.
|
1551
|
+
|
1552
|
+
For the criterion parts, we need to determine if there is a
|
1553
|
+
collection of connectors C1, C2,... such that by combining them you
|
1554
|
+
can get a link name L that post_process_matches bar or baz. Here's a
|
1555
|
+
way to do this. Say bar="Xabc". Then we see if there are connector
|
1556
|
+
names that post_process_match "Xa##", "X#b#", and "X##c". They must
|
1557
|
+
all be there in order for it to be possible to create a link name
|
1558
|
+
"Xabc". A "*" in the criterion part is a little different. In this
|
1559
|
+
case we can simply skip the * (treat it like an upper case letter)
|
1560
|
+
for this purpose. So if bar="X*ab" then we look for "X*#b" and
|
1561
|
+
"X*a#". (The * in this case could be treated the same as another
|
1562
|
+
subscript without breaking it.) Note also that it's only necessary
|
1563
|
+
to find a way to match one of the many criterion links that may be in
|
1564
|
+
the rule. If none can match, then we can delete the disjunct
|
1565
|
+
containing C.
|
1566
|
+
|
1567
|
+
Here's how we're going to implement this. We'll maintain a multiset
|
1568
|
+
of connector names. We'll represent them in a hash table, where the
|
1569
|
+
hash function uses only the upper case letters of the connector name.
|
1570
|
+
We'll insert all the connectors into the multiset. The multiset will
|
1571
|
+
support the operation of deletion (probably simplest to just
|
1572
|
+
decrement the count). Here's the algorithm.
|
1573
|
+
|
1574
|
+
Insert all the connectors into M.
|
1575
|
+
|
1576
|
+
While the previous pass caused a count to go to 0 do:
|
1577
|
+
For each connector C do
|
1578
|
+
For each rule R do
|
1579
|
+
if C is a trigger for R and the criterion links
|
1580
|
+
of the rule cannot be satisfied by the connectors in
|
1581
|
+
M, Then:
|
1582
|
+
We delete C's disjunct. But before we do,
|
1583
|
+
we remove all the connectors of this disjunct
|
1584
|
+
from the multiset. Keep tabs on whether or not
|
1585
|
+
any of the counts went to 0.
|
1586
|
+
|
1587
|
+
|
1588
|
+
|
1589
|
+
Efficiency hacks to be added later:
|
1590
|
+
Note for a given rule can become less and less satisfiable.
|
1591
|
+
That is, rule_satisfiable(r) for a given rule r can change from
|
1592
|
+
TRUE to FALSE, but not vice versa. So once it's FALSE, we can just
|
1593
|
+
remember that.
|
1594
|
+
|
1595
|
+
Consider the effect of a pass p on the set of rules that are
|
1596
|
+
satisfiable. Suppose this set does not change. Then pass p+1
|
1597
|
+
will do nothing. This is true even if pass p caused some
|
1598
|
+
disjuncts to be deleted. (This observation will only obviate
|
1599
|
+
the need for the last pass.)
|
1600
|
+
|
1601
|
+
*/
|
1602
|
+
|
1603
|
+
static multiset_table * cms_table_new(void)
|
1604
|
+
{
|
1605
|
+
multiset_table *mt;
|
1606
|
+
int i;
|
1607
|
+
|
1608
|
+
mt = (multiset_table *) malloc(sizeof(multiset_table));
|
1609
|
+
|
1610
|
+
for (i=0; i<CMS_SIZE; i++) {
|
1611
|
+
mt->cms_table[i] = NULL;
|
1612
|
+
}
|
1613
|
+
return mt;
|
1614
|
+
}
|
1615
|
+
|
1616
|
+
static void cms_table_delete(multiset_table *mt)
|
1617
|
+
{
|
1618
|
+
Cms * cms, *xcms;
|
1619
|
+
int i;
|
1620
|
+
for (i=0; i<CMS_SIZE; i++)
|
1621
|
+
{
|
1622
|
+
for (cms = mt->cms_table[i]; cms != NULL; cms = xcms)
|
1623
|
+
{
|
1624
|
+
xcms = cms->next;
|
1625
|
+
xfree(cms, sizeof(Cms));
|
1626
|
+
}
|
1627
|
+
}
|
1628
|
+
free(mt);
|
1629
|
+
}
|
1630
|
+
|
1631
|
+
static int cms_hash(const char * s)
|
1632
|
+
{
|
1633
|
+
unsigned int i = 5381;
|
1634
|
+
while (isupper((int) *s)) /* connector names are not yet UTF8-capable */
|
1635
|
+
{
|
1636
|
+
i = ((i << 5) + i) + *s;
|
1637
|
+
s++;
|
1638
|
+
}
|
1639
|
+
return (i & (CMS_SIZE-1));
|
1640
|
+
}
|
1641
|
+
|
1642
|
+
/**
|
1643
|
+
* This returns TRUE if there is a connector name C in the table
|
1644
|
+
* such that post_process_match(pp_match_name, C) is TRUE
|
1645
|
+
*/
|
1646
|
+
static int match_in_cms_table(multiset_table *cmt, const char * pp_match_name)
|
1647
|
+
{
|
1648
|
+
Cms * cms;
|
1649
|
+
for (cms = cmt->cms_table[cms_hash(pp_match_name)]; cms != NULL; cms = cms->next)
|
1650
|
+
{
|
1651
|
+
if(post_process_match(pp_match_name, cms->name)) return TRUE;
|
1652
|
+
}
|
1653
|
+
return FALSE;
|
1654
|
+
}
|
1655
|
+
|
1656
|
+
static Cms * lookup_in_cms_table(multiset_table *cmt, const char * str)
|
1657
|
+
{
|
1658
|
+
Cms * cms;
|
1659
|
+
for (cms = cmt->cms_table[cms_hash(str)]; cms != NULL; cms = cms->next)
|
1660
|
+
{
|
1661
|
+
if(strcmp(str, cms->name) == 0) return cms;
|
1662
|
+
}
|
1663
|
+
return NULL;
|
1664
|
+
}
|
1665
|
+
|
1666
|
+
static void insert_in_cms_table(multiset_table *cmt, const char * str)
|
1667
|
+
{
|
1668
|
+
Cms * cms;
|
1669
|
+
int h;
|
1670
|
+
cms = lookup_in_cms_table(cmt, str);
|
1671
|
+
if (cms != NULL) {
|
1672
|
+
cms->count++;
|
1673
|
+
} else {
|
1674
|
+
cms = (Cms *) xalloc(sizeof(Cms));
|
1675
|
+
cms->name = str; /* don't copy the string...just keep a pointer to it.
|
1676
|
+
we won't free these later */
|
1677
|
+
cms->count = 1;
|
1678
|
+
h = cms_hash(str);
|
1679
|
+
cms->next = cmt->cms_table[h];
|
1680
|
+
cmt->cms_table[h] = cms;
|
1681
|
+
}
|
1682
|
+
}
|
1683
|
+
|
1684
|
+
/**
|
1685
|
+
* Delete the given string from the table. Return TRUE if
|
1686
|
+
* this caused a count to go to 0, return FALSE otherwise.
|
1687
|
+
*/
|
1688
|
+
static int delete_from_cms_table(multiset_table *cmt, const char * str)
|
1689
|
+
{
|
1690
|
+
Cms * cms;
|
1691
|
+
cms = lookup_in_cms_table(cmt, str);
|
1692
|
+
if (cms != NULL && cms->count > 0) {
|
1693
|
+
cms->count--;
|
1694
|
+
return (cms->count == 0);
|
1695
|
+
}
|
1696
|
+
return FALSE;
|
1697
|
+
}
|
1698
|
+
|
1699
|
+
static int rule_satisfiable(multiset_table *cmt, pp_linkset *ls)
|
1700
|
+
{
|
1701
|
+
int hashval;
|
1702
|
+
const char * t;
|
1703
|
+
char name[20], *s;
|
1704
|
+
pp_linkset_node *p;
|
1705
|
+
int bad, n_subscripts;
|
1706
|
+
|
1707
|
+
for (hashval = 0; hashval < ls->hash_table_size; hashval++)
|
1708
|
+
{
|
1709
|
+
for (p = ls->hash_table[hashval]; p!=NULL; p=p->next)
|
1710
|
+
{
|
1711
|
+
/* ok, we've got our hands on one of the criterion links */
|
1712
|
+
strncpy(name, p->str, sizeof(name)-1);
|
1713
|
+
/* could actually use the string in place because we change it back */
|
1714
|
+
name[sizeof(name)-1] = '\0';
|
1715
|
+
/* now we want to see if we can satisfy this criterion link */
|
1716
|
+
/* with a collection of the links in the cms table */
|
1717
|
+
|
1718
|
+
for (s = name; isupper((int)*s); s++) {}
|
1719
|
+
for (;*s != '\0'; s++) if (*s != '*') *s = '#';
|
1720
|
+
for (s = name, t = p->str; isupper((int) *s); s++, t++) {}
|
1721
|
+
|
1722
|
+
/* s and t remain in lockstep */
|
1723
|
+
bad = 0;
|
1724
|
+
n_subscripts = 0;
|
1725
|
+
for (;*s != '\0' && bad==0; s++, t++) {
|
1726
|
+
if (*s == '*') continue;
|
1727
|
+
n_subscripts++;
|
1728
|
+
/* after the upper case part, and is not a * so must be a regular subscript */
|
1729
|
+
*s = *t;
|
1730
|
+
if (!match_in_cms_table(cmt, name)) bad++;
|
1731
|
+
*s = '#';
|
1732
|
+
}
|
1733
|
+
|
1734
|
+
if (n_subscripts == 0) {
|
1735
|
+
/* now we handle the special case which occurs if there
|
1736
|
+
were 0 subscripts */
|
1737
|
+
if (!match_in_cms_table(cmt, name)) bad++;
|
1738
|
+
}
|
1739
|
+
|
1740
|
+
/* now if bad==0 this criterion link does the job
|
1741
|
+
to satisfy the needs of the trigger link */
|
1742
|
+
|
1743
|
+
if (bad == 0) return TRUE;
|
1744
|
+
}
|
1745
|
+
}
|
1746
|
+
return FALSE;
|
1747
|
+
}
|
1748
|
+
|
1749
|
+
static int pp_prune(Sentence sent, Parse_Options opts)
|
1750
|
+
{
|
1751
|
+
pp_knowledge * knowledge;
|
1752
|
+
pp_rule rule;
|
1753
|
+
const char * selector;
|
1754
|
+
pp_linkset * link_set;
|
1755
|
+
int i, w, dir;
|
1756
|
+
Disjunct *d;
|
1757
|
+
Connector *c;
|
1758
|
+
int change, total_deleted, N_deleted, deleteme;
|
1759
|
+
multiset_table *cmt;
|
1760
|
+
|
1761
|
+
if (sent->dict->postprocessor == NULL) return 0;
|
1762
|
+
|
1763
|
+
knowledge = sent->dict->postprocessor->knowledge;
|
1764
|
+
|
1765
|
+
cmt = cms_table_new();
|
1766
|
+
|
1767
|
+
for (w = 0; w < sent->length; w++) {
|
1768
|
+
for (d = sent->word[w].d; d != NULL; d = d->next) {
|
1769
|
+
d->marked = TRUE;
|
1770
|
+
for (dir=0; dir < 2; dir++) {
|
1771
|
+
for (c = ( (dir)?(d->left):(d->right) ); c!=NULL; c=c->next) {
|
1772
|
+
insert_in_cms_table(cmt, c->string);
|
1773
|
+
}
|
1774
|
+
}
|
1775
|
+
}
|
1776
|
+
}
|
1777
|
+
|
1778
|
+
total_deleted = 0;
|
1779
|
+
change = 1;
|
1780
|
+
while (change > 0) {
|
1781
|
+
change = 0;
|
1782
|
+
N_deleted = 0;
|
1783
|
+
for (w = 0; w < sent->length; w++) {
|
1784
|
+
for (d = sent->word[w].d; d != NULL; d = d->next) {
|
1785
|
+
if (!d->marked) continue;
|
1786
|
+
deleteme = FALSE;
|
1787
|
+
for (dir=0; dir < 2; dir++) {
|
1788
|
+
for (c = ( (dir)?(d->left):(d->right) ); c!=NULL; c=c->next) {
|
1789
|
+
for (i=0; i<knowledge->n_contains_one_rules; i++) {
|
1790
|
+
|
1791
|
+
rule = knowledge->contains_one_rules[i]; /* the ith rule */
|
1792
|
+
selector = rule.selector; /* selector string for this rule */
|
1793
|
+
link_set = rule.link_set; /* the set of criterion links */
|
1794
|
+
|
1795
|
+
if (strchr(selector, '*') != NULL) continue; /* If it has a * forget it */
|
1796
|
+
|
1797
|
+
if (!post_process_match(selector, c->string)) continue;
|
1798
|
+
|
1799
|
+
/*
|
1800
|
+
printf("pp_prune: trigger ok. selector = %s c->string = %s\n", selector, c->string);
|
1801
|
+
*/
|
1802
|
+
|
1803
|
+
/* We know c matches the trigger link of the rule. */
|
1804
|
+
/* Now check the criterion links */
|
1805
|
+
|
1806
|
+
if (!rule_satisfiable(cmt, link_set)) {
|
1807
|
+
deleteme = TRUE;
|
1808
|
+
}
|
1809
|
+
if (deleteme) break;
|
1810
|
+
}
|
1811
|
+
if (deleteme) break;
|
1812
|
+
}
|
1813
|
+
if (deleteme) break;
|
1814
|
+
}
|
1815
|
+
|
1816
|
+
if (deleteme) { /* now we delete this disjunct */
|
1817
|
+
N_deleted++;
|
1818
|
+
total_deleted++;
|
1819
|
+
d->marked = FALSE; /* mark for deletion later */
|
1820
|
+
for (dir=0; dir < 2; dir++) {
|
1821
|
+
for (c = ( (dir)?(d->left):(d->right) ); c!=NULL; c=c->next) {
|
1822
|
+
change += delete_from_cms_table(cmt, c->string);
|
1823
|
+
}
|
1824
|
+
}
|
1825
|
+
}
|
1826
|
+
}
|
1827
|
+
}
|
1828
|
+
|
1829
|
+
if (verbosity > 2) {
|
1830
|
+
printf("pp_prune pass deleted %d\n", N_deleted);
|
1831
|
+
}
|
1832
|
+
|
1833
|
+
}
|
1834
|
+
delete_unmarked_disjuncts(sent);
|
1835
|
+
cms_table_delete(cmt);
|
1836
|
+
|
1837
|
+
if (verbosity > 2) {
|
1838
|
+
printf("\nAfter pp_pruning:\n");
|
1839
|
+
print_disjunct_counts(sent);
|
1840
|
+
}
|
1841
|
+
|
1842
|
+
print_time(opts, "pp pruning");
|
1843
|
+
|
1844
|
+
return total_deleted;
|
1845
|
+
}
|
1846
|
+
|
1847
|
+
|
1848
|
+
/**
|
1849
|
+
* Do the following pruning steps until nothing happens:
|
1850
|
+
* power pp power pp power pp....
|
1851
|
+
* Make sure you do them both at least once.
|
1852
|
+
*/
|
1853
|
+
void pp_and_power_prune(Sentence sent, int mode, Parse_Options opts)
|
1854
|
+
{
|
1855
|
+
power_prune(sent, mode, opts);
|
1856
|
+
|
1857
|
+
for (;;) {
|
1858
|
+
if (parse_options_resources_exhausted(opts)) break;
|
1859
|
+
if (pp_prune(sent, opts) == 0) break;
|
1860
|
+
if (parse_options_resources_exhausted(opts)) break;
|
1861
|
+
if (power_prune(sent, mode, opts) == 0) break;
|
1862
|
+
}
|
1863
|
+
}
|
1864
|
+
|