From d9cc1a6986188a97e09e4c8cef46c34eee5f9cd2 Mon Sep 17 00:00:00 2001 From: Chris Dyer Date: Sun, 10 Nov 2013 00:58:44 -0500 Subject: guard against direct includes of tr1 --- decoder/Makefile.am | 4 - decoder/apply_models.cc | 11 +- decoder/cdec_ff.cc | 11 - decoder/decoder.cc | 10 +- decoder/dwarf.cc | 3209 ------------------------------------ decoder/dwarf.h | 286 ---- decoder/earley_composer.cc | 11 +- decoder/factored_lexicon_helper.cc | 1 + decoder/factored_lexicon_helper.h | 3 +- decoder/ff_dwarf.cc | 894 ---------- decoder/ff_dwarf.h | 100 -- decoder/ff_lm.cc | 101 -- decoder/ff_lm.h | 22 - decoder/ff_source_syntax.cc | 6 + decoder/ff_source_syntax2.cc | 4 +- decoder/ff_source_syntax2_p.cc | 10 +- decoder/ff_source_syntax_p.cc | 9 +- decoder/ff_wordalign.cc | 9 +- decoder/ff_wordalign.h | 13 +- decoder/ff_wordset.cc | 52 +- decoder/ff_wordset.h | 73 +- decoder/grammar.cc | 11 +- decoder/hg_intersect.cc | 8 +- decoder/kbest.h | 11 +- decoder/maxtrans_blunsom.cc | 11 +- decoder/phrasebased_translator.cc | 11 +- decoder/scfg_translator.cc | 10 +- decoder/sentence_metadata.h | 4 +- 28 files changed, 165 insertions(+), 4740 deletions(-) delete mode 100644 decoder/dwarf.cc delete mode 100644 decoder/dwarf.h delete mode 100644 decoder/ff_dwarf.cc delete mode 100644 decoder/ff_dwarf.h (limited to 'decoder') diff --git a/decoder/Makefile.am b/decoder/Makefile.am index 914faaea..39a13ad8 100644 --- a/decoder/Makefile.am +++ b/decoder/Makefile.am @@ -41,7 +41,6 @@ libcdec_a_SOURCES = \ cfg_options.h \ csplit.h \ decoder.h \ - dwarf.h \ earley_composer.h \ exp_semiring.h \ factored_lexicon_helper.h \ @@ -51,7 +50,6 @@ libcdec_a_SOURCES = \ ff_charset.h \ ff_context.h \ ff_csplit.h \ - ff_dwarf.h \ ff_external.h \ ff_factory.h \ ff_klm.h \ @@ -103,8 +101,6 @@ libcdec_a_SOURCES = \ maxtrans_blunsom.cc \ cdec_ff.cc \ cfg.cc \ - dwarf.cc \ - ff_dwarf.cc \ ff_external.cc \ rule_lexer.cc \ fst_translator.cc \ diff --git a/decoder/apply_models.cc b/decoder/apply_models.cc index 330de9e2..2e093d6a 100644 --- a/decoder/apply_models.cc +++ b/decoder/apply_models.cc @@ -8,8 +8,14 @@ #include #include -#include -#include +#ifdef HAVE_CXX11 +# include +# include +#else +# include +# include +namespace std { using std::tr1::unordered_map; using std::tr1::unordered_set; } +#endif #include @@ -23,7 +29,6 @@ #define FAST_CP_2 3 using namespace std; -using namespace std::tr1; struct Candidate; typedef SmallVectorInt JVector; diff --git a/decoder/cdec_ff.cc b/decoder/cdec_ff.cc index e7b31f50..09597e87 100644 --- a/decoder/cdec_ff.cc +++ b/decoder/cdec_ff.cc @@ -29,12 +29,8 @@ #include "ff_register.h" #include "ff_charset.h" #include "ff_wordset.h" -#include "ff_dwarf.h" #include "ff_external.h" -#ifdef HAVE_GLC -#include -#endif void register_feature_functions() { static bool registered = false; @@ -51,9 +47,6 @@ void register_feature_functions() { RegisterFF(); //TODO: use for all features the new Register which requires static FF::usage(false,false) give name -#ifdef HAVE_RANDLM - ff_registry.Register("RandLM", new FFFactory); -#endif ff_registry.Register("SpanFeatures", new FFFactory()); ff_registry.Register("NgramFeatures", new FFFactory()); ff_registry.Register("RuleContextFeatures", new FFFactory()); @@ -98,10 +91,6 @@ void register_feature_functions() { ff_registry.Register("WordPairFeatures", new FFFactory); ff_registry.Register("SourcePathFeatures", new FFFactory); ff_registry.Register("WordSet", new FFFactory); - ff_registry.Register("Dwarf", new FFFactory); ff_registry.Register("External", new FFFactory); -#ifdef HAVE_GLC - ff_registry.Register("ContextCRF", new FFFactory); -#endif } diff --git a/decoder/decoder.cc b/decoder/decoder.cc index 31e6dc46..2c0e07b7 100644 --- a/decoder/decoder.cc +++ b/decoder/decoder.cc @@ -1,6 +1,11 @@ #include "decoder.h" -#include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_map; } +#endif #include #include #include @@ -61,7 +66,6 @@ static const double kMINUS_EPSILON = -1e-6; // don't be too strict using namespace std; -using namespace std::tr1; namespace po = boost::program_options; static bool verbose_feature_functions=true; @@ -90,7 +94,7 @@ struct ELengthWeightFunction { } }; inline void ShowBanner() { - cerr << "cdec v1.0 (c) 2009-2011 by Chris Dyer\n"; + cerr << "cdec (c) 2009--2013 by Chris Dyer\n"; } inline string str(char const* name,po::variables_map const& conf) { diff --git a/decoder/dwarf.cc b/decoder/dwarf.cc deleted file mode 100644 index fb0404a6..00000000 --- a/decoder/dwarf.cc +++ /dev/null @@ -1,3209 +0,0 @@ -#include "dwarf.h" -#include "tdict.h" -#include "wordid.h" -#include "lattice.h" -#include "ff_dwarf.h" -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -using namespace std; -using namespace std::tr1; -using namespace boost::tuples; -using namespace boost; - -Alignment::Alignment() { - //unordered_map,int> XX; - _I=0; - _J=0; - kSOS = TD::Convert(""); - kEOS = TD::Convert(""); - kUNK = TD::Convert("**UNKNOWN**"); - SourceFWAntsIdxs = new int*[MAX_ARITY]; - SourceFWAntsAbsIdxs = new int*[MAX_ARITY]; - TargetFWAntsIdxs = new int*[MAX_ARITY]; - SourceAntsIdxs = new int*[MAX_ARITY]; - TargetAntsIdxs = new int*[MAX_ARITY]; - AntsAl = new int*[MAX_ARITY]; - for (int idx=0; idx=0); - if (start==-1) start = _tSpan[j][0]; - if (_tSpan[j][0]==MINIMUM_INIT) return -1; - for (int idx=start; idx<=_tSpan[j][1]; idx++) { - if (_matrix[j][idx]) return idx; - } - return -1; -} - -int Alignment::sourceOf(int i, int start) { - assert(i>=0); - if (start==-1) start = _sSpan[i][0]; - if (_sSpan[i][0]==MINIMUM_INIT) return -1; - for (int idx=start; idx<=_sSpan[i][1]; idx++) { - if (_matrix[idx][i]) return idx; - } - return -1; -} - -void Alignment::clearAls(int prevJ, int prevI) { - for (int j=0; j<=prevJ; j++) { - for (int i=0; i neither, 1 -> leftFirst, 2 -> rightFirst, 3 -> dontCare - if (DEBUG) cerr << "DominanceSource(" << fw1 << "," << fw2 << ")" << endl; - //cerr << TD::Convert(_f[fw1]) << "," << TD::Convert(_f[fw2]) << endl; - //cerr << AsString() << endl; - int dom = 0; - curr_al.push_back(fw1); curr_al.push_back(fw2); - if (doms_hash.find(curr_al)==doms_hash.end()) { - int* block = blockSource(fw1,fw2); - //cerr << "block = " << block[0] << "," << block[1] << "," << block[2] << "," << block[3] << endl; - if (block[0]==fw1) { - int tfw10 = _tSpan[fw1][0]; - int tfw11 = _tSpan[fw1][1]; - //cerr << "tfw = " << tfw10 << "," << tfw11 << endl; - if (tfw11<0) { - dom+=1; - } else { - if ((block[2]==tfw10 || block[3]==tfw11)) dom+=1; - } - } - if (block[1]==fw2) { - int tfw20 = _tSpan[fw2][0]; - int tfw21 = _tSpan[fw2][1]; - //cerr << "tfw = " << tfw20 << "," << tfw21 << endl; - if (tfw21<0) { - dom+=2; - } else { - if ((block[2]==tfw20 || block[3]==tfw21)) dom+=2; - } - } - delete block; - doms_hash.insert(pair,int>(curr_al,dom)); - } else { - dom = doms_hash[curr_al]; - } - if (DEBUG) cerr << " dom = " << dom << endl; - curr_al.pop_back(); curr_al.pop_back(); - return dom; -} - -vector Alignment::DominanceSource4Sampler(int fw1, int fw2) { - if (DEBUG) cerr << "DominanceSource4Sampler(" << fw1 << "," << fw2 << ")" << endl; - int dom = 0; - int* block = blockSource(fw1,fw2); - //cerr << "block = " << block[0] << "," << block[1] << "," << block[2] << "," << block[3] << endl; - if (block[0]==fw1) { - int tfw10 = _tSpan[fw1][0]; - int tfw11 = _tSpan[fw1][1]; - //cerr << "tfw = " << tfw10 << "," << tfw11 << endl; - if (tfw11<0) { - dom+=1; - } else { - if ((block[2]==tfw10 || block[3]==tfw11)) dom+=1; - } - } - if (block[1]==fw2) { - int tfw20 = _tSpan[fw2][0]; - int tfw21 = _tSpan[fw2][1]; - //cerr << "tfw = " << tfw20 << "," << tfw21 << endl; - if (tfw21<0) { - dom+=2; - } else { - if ((block[2]==tfw20 || block[3]==tfw21)) dom+=2; - } - } - if (DEBUG) cerr << "doms = " << dom << endl; - vector ret; - ret.push_back(dom); ret.push_back(block[0]); ret.push_back(block[1]); - ret.push_back(block[2]); ret.push_back(block[3]); - delete block; - return ret; -} - -int Alignment::DominanceTarget(int fw1, int fw2) { - int dom = 0; - curr_al.push_back(fw1); curr_al.push_back(fw2); - if (domt_hash.find(curr_al)==domt_hash.end()) { - int* block = blockTarget(fw1,fw2); - if (block[2]==fw1) { - int sfw10 = _sSpan[fw1][0]; - int sfw11 = _sSpan[fw1][1]; - if (sfw11<0) { - dom+=1; - } else { - if (block[0]==sfw10 || block[1]==sfw11) dom+=1; - } - } - if (block[3]==fw2) { - int sfw20 = _sSpan[fw2][0]; - int sfw21 = _sSpan[fw2][0]; - if (sfw21<0) { - dom+=2; - } else { - if (block[0]==sfw20 || block[1]==sfw21) dom+=2; - } - } - delete block; - domt_hash.insert(pair,int>(curr_al,dom)); - } else { - dom = domt_hash[curr_al]; - } - curr_al.pop_back(); curr_al.pop_back(); - return dom; -} - -vector Alignment::DominanceTarget4Sampler(int fw1, int fw2) { - int dom = 0; - int* block = blockTarget(fw1,fw2); - if (block[2]==fw1) { - int sfw10 = _sSpan[fw1][0]; - int sfw11 = _sSpan[fw1][1]; - if (sfw11<0) { - dom+=1; - } else { - if (block[0]==sfw10 || block[1]==sfw11) dom+=1; - } - } - if (block[3]==fw2) { - int sfw20 = _sSpan[fw2][0]; - int sfw21 = _sSpan[fw2][0]; - if (sfw21<0) { - dom+=2; - } else { - if (block[0]==sfw20 || block[1]==sfw21) dom+=2; - } - } - vector ret; - ret.push_back(dom); ret.push_back(block[0]); ret.push_back(block[1]); - ret.push_back(block[2]); ret.push_back(block[3]); - delete block; - return ret; -} - -void Alignment::OrientationSource(int fw, int* oril, int* orir, bool Lcompute, bool Rcompute) { - OrientationSource(fw,fw,oril,orir,Lcompute,Rcompute); -} - -vector Alignment::OrientationSourceLeft4Sampler(int fw) { - return OrientationSourceLeft4Sampler(fw,fw); -} - -vector Alignment::OrientationSourceLeft4Sampler(int fw0, int fw1) { - if (DEBUG) cerr << "OrientationSourceLeft4Sampler(" << fw0 << "," << fw1 << ")" << endl; - int oril = 0; - int N0=fw0-1; - while (N0>=0) { - if (minTSpan(N0)!=MINIMUM_INIT) break; - N0--; - } - int N1=fw1+1; - while (N1<_J) { - if (minTSpan(N1)!=MINIMUM_INIT) break; - N1++; - } - if (minTSpan(fw0)==MINIMUM_INIT && minTSpan(fw1)==MINIMUM_INIT) { - fw0 = N1; fw1 = N0; - } - if (DEBUG) cerr << "fw0=" << fw0 << ", fw1=" << fw1 << ", N0=" << N0 << ", N1=" << N1 << endl; - if (maxTSpan(N0)=0) { - if (minTSpan(N0)!=MINIMUM_INIT) break; - N0--; - } - int N1=fw1+1; - while (N1<_J) { - if (minTSpan(N1)!=MINIMUM_INIT) break; - N1++; - } - if (minTSpan(fw0)==MINIMUM_INIT && minTSpan(fw1)==MINIMUM_INIT) { - fw0 = N1; fw1 = N0; - } - if (DEBUG) cerr << "fw0=" << fw0 << ", fw1=" << fw1 << ", N0=" << N0 << ", N1=" << N1 << endl; - if (maxTSpan(N1)=0) { - if (minTSpan(N0)!=MINIMUM_INIT) break; - N0--; - } - int N1=fw1+1; - while (N1<_J) { - if (minTSpan(N1)!=MINIMUM_INIT) break; - N1++; - } - if (minTSpan(fw0)==MINIMUM_INIT && minTSpan(fw1)==MINIMUM_INIT) { - fw0 = N1; fw1 = N0; - //cerr << "minTSpan(fw)==MINIMUM_INIT, thus fw0=" << fw0 << ", fw1=" << fw1 << endl; - } - if (DEBUG) cerr << "fw0=" << fw0 << ", fw1=" << fw1 << ", N0=" << N0 << ", N1=" << N1 << endl; - if (maxTSpan(N0) Alignment::OrientationTargetLeft4Sampler(int fw) { - return OrientationTargetLeft4Sampler(fw,fw); -} - -vector Alignment::OrientationTargetLeft4Sampler(int fw0, int fw1) { - if (DEBUG) cerr << "OrientationTargetLeft4Sampler " << fw0 << "," << fw1 << endl; - int oril=0; - int N0=fw0-1; - while (N0>=0) { - if (minSSpan(N0)!=MINIMUM_INIT) break; - N0--; - } - int N1=fw1+1; - while (N1<_I) { - if (minSSpan(N1)!=MINIMUM_INIT) break; - N1++; - } - if (minSSpan(fw0)==MINIMUM_INIT && minSSpan(fw1)==MINIMUM_INIT) { - fw0=N1; fw1=N0; - } - if (maxSSpan(N0)=0) { - if (minSSpan(N0)!=MINIMUM_INIT) break; - N0--; - } - int N1=fw1+1; - while (N1<_I) { - if (minSSpan(N1)!=MINIMUM_INIT) break; - N1++; - } - if (minSSpan(fw0)==MINIMUM_INIT && minSSpan(fw1)==MINIMUM_INIT) { - fw0=N1; fw1=N0; - } - if (maxSSpan(N1)=0) { - if (minSSpan(N0)!=MINIMUM_INIT) break; - N0--; - } - int N1=fw1+1; - while (N1<_I) { - if (minSSpan(N1)!=MINIMUM_INIT) break; - N1++; - } - if (minSSpan(fw0)==MINIMUM_INIT && minSSpan(fw1)==MINIMUM_INIT) { - fw0=N1; fw1=N0; - } - if (DEBUG) { - cerr << "fw0:" << fw0 << ", fw1:" << fw1 << ", N0:" << N0 << ", N1:" << N1 << endl ; - cerr << "minSSpan(N0)=" << minSSpan(N0) << " maxSSpan(N0)=" << maxSSpan(N0); - cerr << " minSSpan(fw0)="<< minSSpan(fw0) << " maxSSpan(fw0)=" << maxSSpan(fw0) << endl; - cerr << "minSSpan(fw1)=" << minSSpan(fw1) << " maxSSpan(fw1)=" << maxSSpan(fw1); - cerr << " minSSpan(N1)="<< minSSpan(N1) << " maxSSpan(N1)=" << maxSSpan(N1) << endl; - } - if (maxSSpan(N0)=0; j--) - if (_tSpan[j][0]!=MINIMUM_INIT) return j; - return -1; -} - -int Alignment::firstTargetAligned(int start) { - for (int i=start; i<_I; i++) - if (_sSpan[i][0]!=MINIMUM_INIT) return i; - return -1; -} - -int Alignment::lastTargetAligned(int end) { - for (int i=end; i>=0; i--) - if (_sSpan[i][0]!=MINIMUM_INIT) return i; - return -1; -} - -void Alignment::BorderingSFWsOnly() { -// removes the record of all function word alignments, except those at the borders -// the number of alignments kept may be more than two -// i.e. where the leftmost / the rightmost alignments are unaligned. -// In such cases, this function continues keeping function word alignments until the -// first (or last) alignment words. - if (SourceFWIdxs[0]>2) { - int firstCut = 1; - for (int j=2; j<=SourceFWIdxs[0]; j++) { - if (SourceFWIdxs[3*j-2]>fas) break; - firstCut=j; - } - int lastCut = SourceFWIdxs[0]; - for (int j=SourceFWIdxs[0]-1; j>=0; j--) { - if (SourceFWIdxs[3*j-2]=lastCut) return; - int delta = 0; - for (int j=lastCut; j<=SourceFWIdxs[0]; j++) { - delta++; - SourceFWIdxs[3*(firstCut+delta)-2]=SourceFWIdxs[3*j-2]; - SourceFWIdxs[3*(firstCut+delta)-1]=SourceFWIdxs[3*j-1]; - SourceFWIdxs[3*(firstCut+delta)] =SourceFWIdxs[3*j]; - } - SourceFWIdxs[0]=firstCut+delta; - } -} - -void Alignment::BorderingTFWsOnly() { -// similar to BorderingSFWsOnly() except this looks at the source side. - if (TargetFWIdxs[0]>2) { - int firstCut = 1; - for (int j=2; j<=TargetFWIdxs[0]; j++) { - if (TargetFWIdxs[3*j-2]>fat) break; - firstCut=j; - } - int lastCut = TargetFWIdxs[0]; - for (int j=TargetFWIdxs[0]-1; j>=0; j--) { - if (TargetFWIdxs[3*j-2]=lastCut) return; - int delta = 0; - for (int j=lastCut; j<=TargetFWIdxs[0]; j++) { - delta++; - TargetFWIdxs[3*(firstCut+delta)-2]=TargetFWIdxs[3*j-2]; - TargetFWIdxs[3*(firstCut+delta)-1]=TargetFWIdxs[3*j-1]; - TargetFWIdxs[3*(firstCut+delta)] =TargetFWIdxs[3*j]; - } - TargetFWIdxs[0]=firstCut+delta; - } -} - -void Alignment::FillFWIdxsState(int* state, int fas, int las, int fat, int lat) { - if (DEBUG) cerr << "FillFWIdxsState ("<< fas <<","<< las<<"," << fat <<"," << lat << ")" << endl; - if (fas==las) las+=1; - if (fat==lat) lat+=1; - for (int idx=0; idx<12; idx++) state[idx]=-1; - if (SourceFWIdxs[0]<=2) { - if (SourceFWIdxs[0]>=1) {state[0]=SourceFWIdxs[1]; state[1]=SourceFWIdxs[2]; state[2]=SourceFWIdxs[3];} - if (SourceFWIdxs[0]==2) {state[3]=SourceFWIdxs[4]; state[4]=SourceFWIdxs[5]; state[5]=SourceFWIdxs[6];} - } else { - if (SourceFWIdxs[1]>fas) { - state[0]=SourceFWIdxs[1]; state[1]=SourceFWIdxs[2]; state[2]=SourceFWIdxs[3]; - } else { - ostringstream issf; ostringstream isse; - for (int idx=1; idx<=SourceFWIdxs[0]; idx++) { - if (SourceFWIdxs[3*idx-2]>las) break; - if (idx>1) { issf << " "; isse << " ";}; - issf << TD::Convert(SourceFWIdxs[3*idx-1]); - isse << TD::Convert(SourceFWIdxs[3*idx]); - state[0]=SourceFWIdxs[3*idx-2]; - if (state[0]>=fas) break; - } - if (state[0]>=0) { - state[1]=TD::Convert(issf.str())*-1; state[2]=TD::Convert(isse.str()); //multiplying source with -1 as marker - } - } - if (SourceFWIdxs[SourceFWIdxs[0]*3-2]==las) { - state[3]=SourceFWIdxs[SourceFWIdxs[0]*3-2]; - state[4]=SourceFWIdxs[SourceFWIdxs[0]*3-1]; - state[5]=SourceFWIdxs[SourceFWIdxs[0]*3]; - } else { - int lastCut = SourceFWIdxs[0]; - for (int j=lastCut-1; j>=state[0]+1; j--) { - if (SourceFWIdxs[3*j-2]==state[0]) break; - if (SourceFWIdxs[3*j-2]lastCut) { issf << " "; isse << " ";}; - issf << TD::Convert(SourceFWIdxs[3*idx-1]); - isse << TD::Convert(SourceFWIdxs[3*idx]); - } - if (state[3]>=0) { - //multiplying source with -1 as compound marker - state[4]=TD::Convert(issf.str())*-1; state[5]=TD::Convert(isse.str()); - } - } - } - if (TargetFWIdxs[0]<=2) { - if (TargetFWIdxs[0]>=1) {state[6]=TargetFWIdxs[1]; state[7]=TargetFWIdxs[2]; state[8]=TargetFWIdxs[3];} - if (TargetFWIdxs[0]==2) {state[9]=TargetFWIdxs[4]; state[10]=TargetFWIdxs[5]; state[11]=TargetFWIdxs[6];} - } else { - if (TargetFWIdxs[1]>fat) { //shouldn't come here if SetTargetBorderingFW is invoked - state[6]=TargetFWIdxs[1]; state[7]=TargetFWIdxs[2]; state[8]=TargetFWIdxs[3]; - } else { - ostringstream issf; ostringstream isse; - for (int idx=1; idx<=TargetFWIdxs[0]; idx++) { - if (TargetFWIdxs[3*idx-2]>fat) break; - if (idx>1) { issf << " "; isse << " ";}; - issf << TD::Convert(TargetFWIdxs[3*idx-1]); - isse << TD::Convert(TargetFWIdxs[3*idx]); - state[6]=TargetFWIdxs[3*idx-2]; - } - state[7]=TD::Convert(issf.str()); state[8]=TD::Convert(isse.str())*-1; - //multiplying target with -1 as compound marker - } - if (TargetFWIdxs[TargetFWIdxs[0]*3-2]==lat) { - state[9]=TargetFWIdxs[TargetFWIdxs[0]*3-2]; - state[10]=TargetFWIdxs[TargetFWIdxs[0]*3-1]; - state[11]=TargetFWIdxs[TargetFWIdxs[0]*3]; - } else { - int lastCut = TargetFWIdxs[0]; - for (int j=lastCut-1; j>=1; j--) { - if (TargetFWIdxs[3*j-2]<=state[9]) break; - if (TargetFWIdxs[3*j-2]lastCut) issf << " "; isse << " ";; - issf << TD::Convert(TargetFWIdxs[3*idx-1]); - isse << TD::Convert(TargetFWIdxs[3*idx]); - } - state[10]=TD::Convert(issf.str()); state[11]=TD::Convert(isse.str())*-1; - } - } -} - -void Alignment::simplifyBackward(vector*blocks, int* block, const vector& danglings) { -// given a *block*, see whether its target span contains any index inside *danglings*. -// if yes, break it; otherwise, keep it. put the result(s) to *blocks* - if (DEBUG) cerr << "simplifyBackward[" << block[0] << "," << block[1] << "," << block[2] << "," << block[3] << "]" << endl; - if (DEBUG) for (int i=0; idanglings[i_dangling]) { - if (i_dangling+1 >= danglings.size()) break; - i_dangling++; - } - while (danglings[i_dangling]==currIdx) { - i_dangling++; - currIdx++; - } - /*if (i_dangling>=danglings.size() && currIdx) { - blocks->push_back(block); - if (DEBUG) cerr << "pushing(1) " << block[0] << "," << block[1] << "," << block[2] << "," << block[3] << endl; - return; - } - if (block[3]push_back(block); - if (DEBUG) cerr << "pushing(2) " << block[0] << "," << block[1] << "," << block[2] << "," << block[3] << endl; - return; - }*/ - if (DEBUG) cerr << "i_dangling = " << i_dangling << endl; - int anchorIdx = danglings[i_dangling]; - if (i_dangling+1>=danglings.size() || anchorIdx>block[3]+1) anchorIdx=block[3]+1; - if (DEBUG) cerr << "anchorIdx = " << anchorIdx << ", currIdx = " << currIdx << endl; - do { - while(currIdx treat this as one alignment each - // record all function word alignments first, important because it may be unaligned - // return true if it's truly simple (no function word alignment involves); false, otherwise - if (DEBUG) cerr << "begin simplify" << endl; - reset(0,0); reset(_J-1,_I-1); // remove the phrase boundary alignments, NEED TO CHECK AGAIN !!! - if (SourceFWIdxs[0]+TargetFWIdxs[0]==0) { // return singleton - if (DEBUG) cerr << "no function words" << endl; - for (int idx=0; idx<12; idx++) ret[idx]=-1; - ret[12]=1; ret[13]=0; ret[14]=0; // 0-0 - FillFWIdxsState(ret,0,0,0,0); - return; - } - curr_al.insert(curr_al.begin(),curr_al.size()); - curr_al.push_back(SourceFWIdxs[0]); - for (int i=1; i<=SourceFWIdxs[0]; i++) curr_al.push_back(SourceFWIdxs[3*i-2]); - curr_al.push_back(TargetFWIdxs[0]); - for (int i=1; i<=TargetFWIdxs[0]; i++) curr_al.push_back(TargetFWIdxs[3*i-2]); - vector el; - if (simplify_hash.find(curr_al)==simplify_hash.end()) { - if (DEBUG) { - cerr << "SourceFWIdxs:" << SourceFWIdxs[0] << endl; - for (int i=1; i<=SourceFWIdxs[0]; i++) - cerr << SourceFWIdxs[3*i-2] << "," << SourceFWIdxs[3*i-1] << "," << SourceFWIdxs[3*i] << endl; - cerr << "TargetFWIdxs:" << TargetFWIdxs[0] << endl; - for (int i=1; i<=TargetFWIdxs[0]; i++) { - cerr << TargetFWIdxs[3*i-2] << "," << TargetFWIdxs[3*i-1] << "," << TargetFWIdxs[3*i] << endl; - } - } - - vector< int* > blocks; // each element contains s1,s2,t1,t2 - int currIdx = 1; // start from 1 to avoid considering phrase start - std::set FWIdxs; - std::vector DanglingTargetFWIdxs; - for (int i=1; i<= SourceFWIdxs[0]; i++) FWIdxs.insert(SourceFWIdxs[3*i-2]); - for (int i=1; i<= TargetFWIdxs[0]; i++) { - int source = sourceOf(TargetFWIdxs[3*i-2]); - if (source>=0) { - do { - FWIdxs.insert(source); - source = sourceOf(TargetFWIdxs[3*i-2],source+1); - } while(source >=0); - } else { - int *block = new int[4]; - block[0]=-1; block[1]=-1; block[2]=TargetFWIdxs[3*i-2]; block[3]=TargetFWIdxs[3*i-2]; - blocks.push_back(block); - if (DEBUG) cerr << "pushing[1] " << block[0] << "," << block[1] << "," << block[2] << "," << block[3] << endl; - DanglingTargetFWIdxs.push_back(TargetFWIdxs[3*i-2]); - } - } - if (DEBUG) - for (std::set::const_iterator iter=FWIdxs.begin(); iter!=FWIdxs.end(); iter++) { - cerr << "FWIdxs=" << *iter << endl; - } - std::set::const_iterator currFWIdx = FWIdxs.begin(); - if (currFWIdx == FWIdxs.end()) { - int* block = new int[4]; - block[0]=1; block[1]=_J-2; block[2]=1; block[3]=_I-2; // no need to consider phrase boundaries - simplifyBackward(&blocks,block,DanglingTargetFWIdxs); - } else { - int anchorIdx = *currFWIdx; // also used to denote _J+1 - do { - // add alignments whose source from currIdx to currFWIdx-1 - while (currIdx=0) target = targetOf(anchorIdx,target+1); - } while (target>=0); - // advance indexes - currIdx = anchorIdx+1; - anchorIdx = getJ()-1; // was minus 2 - if (++currFWIdx!=FWIdxs.end()) anchorIdx = *currFWIdx; - } while (currIdx<=getJ()-2); - } - - - vector source_block_mapper(getJ(),-1); - vector target_block_mapper(getI(),-1); - for (int i = 0; i=0) source_block_mapper[blocks[i][0]]=1; - if (blocks[i][2]>=0) target_block_mapper[blocks[i][2]]=1; - } - int curr = 1; - int prev = -1; - for (int idx=0; idx0) { - source_block_mapper[idx]=curr++; - prev = curr; - } else { - source_block_mapper[idx]=prev; - } - } - curr = 1; - for (int idx=0; idx0) { - target_block_mapper[idx]=curr++; - prev = curr; - } else { - target_block_mapper[idx]=prev; - } - } - - //assert(blocks.size()<=50); - if (DEBUG) cerr << "resulting alignment:" << endl; - for (int i = 0; i, vector > (curr_al,el)); - if (DEBUG) cerr << "inserted" << endl; - } else { - el = simplify_hash[curr_al]; - } - if (DEBUG) { - cerr << "pull key:el = "; - for (int ii=0; ii1) quickSort(num,1,num[0]); -} - -void Alignment::quickSort(int arr[], int left, int right) { - int i = left, j = right; - int tmp1,tmp2,tmp3; - int mid = (left + right) / 2; - int pivot = arr[3*mid-2]; - - /* partition */ - while (i <= j) { - while (arr[3*i-2] < pivot) i++; - while (arr[3*j-2] > pivot) j--; - if (i <= j) { - tmp1 = arr[3*i-2]; tmp2 = arr[3*i-1]; tmp3 = arr[3*i]; - arr[3*i-2] = arr[3*j-2]; arr[3*i-1] = arr[3*j-1]; arr[3*i] = arr[3*j]; - arr[3*j-2] = tmp1; arr[3*j-1] = tmp2; arr[3*j] = tmp3; - i++; - j--; - } - }; - - /* recursion */ - if (left < j) quickSort(arr, left, j); - if (i < right) quickSort(arr, i, right); -} - -double Alignment::ScoreOrientation(const CountTable& table, int offset, int ori, WordID cond1, WordID cond2) { - string source = TD::Convert(cond1); - string sourceidx; - if (table.mode == 1) { - sourceidx = source; - int slashidx = sourceidx.find_last_of("/"); - source = sourceidx.substr(0,slashidx); - string idx = sourceidx.substr(slashidx+1); - if (DEBUG) cerr << " sourceidx = " << sourceidx << ", idx = " << idx << endl; - if (idx == "X") { - if (DEBUG) cerr << " idx == X, returning 0" << endl; - return 0; - } - } - string target = TD::Convert(cond2); - if (DEBUG) cerr << "sourceidx='" << sourceidx << "', source='" << source << "', target='" << target << "'" << endl; - double count = table.ultimate[offset+ori-1]; - double total = table.ultimate[offset+5]; - double alpha = 0.1; - double prob = count/total; - if (DEBUG) cerr << "level0 " << count << "/" << total << "=" << prob << endl; - - WordID key_id = (table.mode!=1) ? cond1 : TD::Convert(source); - map::const_iterator it = table.model.find(key_id); - bool stop = (it==table.model.end()); - if (!stop) { - stop=true; - if (it->second[offset+5]>=0) { - count = it->second[offset+ori-1] + alpha * prob; - total = it->second[offset+5] + alpha; - prob = count/total; - stop = false; - if (DEBUG) cerr << "level1 " << count << "/" << total << "=" << prob << endl; - } - } - if (stop) return prob; - - string key = source + " " + target; - it = table.model.find(TD::Convert(key)); - stop = (it==table.model.end()); - if (!stop) { - stop = true; - if (it->second[offset+5]>=0) { - count = it->second[offset+ori-1] + alpha * prob; - total = it->second[offset+5] + alpha; - prob = count/total; - stop = false; - if (DEBUG) cerr << "level2 " << count << "/" << total << "=" << prob << endl; - } - } - - if (stop || table.mode!=1) return prob; - - key = sourceidx + " " + target; - it = table.model.find(TD::Convert(key)); - if (it!=table.model.end()) { - if (it->second[offset+5]>=0) { - count = it->second[offset+ori-1] + alpha * prob; - total = it->second[offset+5] + alpha; - prob = count/total; - if (DEBUG) cerr << "level3 " << count << "/" << total << "=" << prob << endl; - } - } - - return prob; -} - -void Alignment::ScoreOrientation(const CountTable& table, int offset, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, - double alpha1, double beta1) { - if (DEBUG) cerr << "ScoreOrientation:" << TD::Convert(cond1) << "," << TD::Convert(cond2) << ", alpha1 = " << alpha1 << ", beta1 = " << beta1 << endl; - double ret = ScoreOrientation(table,offset,ori,cond1,cond2); - if (isBonus) { - if (table.mode == 0) *bonus += log(ret); else *bonus += ret; - } else { - if (table.mode == 0) *cost += log(ret); else *cost += ret; - } -} - -double Alignment::ScoreOrientationLeft(const CountTable& table, int ori, WordID cond1, WordID cond2) { - double ret = ScoreOrientation(table,0,ori,cond1,cond2); - if (table.mode == 0) return log(ret); - return ret; -} - -double Alignment::ScoreOrientationLeftBackward(const CountTable& table, int ori, WordID cond1, WordID cond2) { - double ret = ScoreOrientation(table,12,ori,cond1,cond2); - if (table.mode == 0) return log(ret); - return ret; -} - -double Alignment::ScoreOrientationRight(const CountTable& table, int ori, WordID cond1, WordID cond2) { - double ret = ScoreOrientation(table,6,ori,cond1,cond2); - if (table.mode == 0) return log(ret); - return ret; -} - -double Alignment::ScoreOrientationRightBackward(const CountTable& table, int ori, WordID cond1, WordID cond2) { - double ret = ScoreOrientation(table,18,ori,cond1,cond2); - if (table.mode == 0) return log(ret); - return ret; -} - -void Alignment::ScoreOrientationLeft(const CountTable& table, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, double alpha1, double beta1) { - if (DEBUG) cerr << "ScoreOrientationLeft(" << isBonus << ")" << endl; - ScoreOrientation(table,0,ori,cond1,cond2,isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha1,beta1); -} - -void Alignment::ScoreOrientationLeftBackward(const CountTable& table, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, double alpha1, double beta1) { - if (DEBUG) cerr << "ScoreOrientationLeftBackward" << endl; - ScoreOrientation(table,12,ori,cond1,cond2,isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha1,beta1); -} - -void Alignment::ScoreOrientationRight(const CountTable& table, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, double alpha1, double beta1) { - if (DEBUG) cerr << "ScoreOrientationRight(" << isBonus << ")" << endl; - ScoreOrientation(table,6,ori,cond1,cond2,isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha1,beta1); -} - -void Alignment::ScoreOrientationRightBackward(const CountTable& table, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, double alpha1, double beta1) { - if (DEBUG) cerr << "ScoreOrientationRightBackward" << endl; - ScoreOrientation(table,18,ori,cond1,cond2,isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha1,beta1); -} - -void Alignment::computeOrientationSourceBackwardPos(const CountTable& table, double *cost, double *bonus, - double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, int maxfwidx, int maxdepth1, int maxdepth2) { - if (DEBUG) cerr << "computeOrientationSourceBackward" << endl; - int oril, orir; - for (int idx=1; idx<=SourceFWRuleIdxs[0]; idx++) { - if (DEBUG) cerr << "considering SourceFWRuleIdxs[" << idx << "]: " << SourceFWRuleIdxs[3*idx-2] << endl; - if (!(SourceFWRuleAbsIdxs[idx]<=maxdepth1 || maxfwidx-SourceFWRuleAbsIdxs[idx]+1<=maxdepth2)) continue; - int* fwblock = blockSource(SourceFWRuleIdxs[3*idx-2],SourceFWRuleIdxs[3*idx-2]); - bool aligned = (fwblock[2]!=MINIMUM_INIT); - if (aligned) { - OrientationTarget(fwblock[2],fwblock[3],&oril,&orir); - } else { - OrientationSource(SourceFWRuleIdxs[3*idx-2],&oril,&orir); - } - if (DEBUG) cerr << "oril = " << oril << ", orir = " << orir << endl; - bool isBonus = false; // fas -> first aligned source word, las -> last aligned source word - if ((aligned && fwblock[2]<=fat)|| - (!aligned && SourceFWRuleIdxs[3*idx-2]<=fas)) isBonus=true; - if (SourceFWRuleAbsIdxs[idx]<=maxdepth1) { - ostringstream nusource; - nusource << TD::Convert(SourceFWRuleIdxs[3*idx-1]) << "/" << SourceFWRuleAbsIdxs[idx]; - ScoreOrientationLeftBackward(table,oril,TD::Convert(nusource.str()),SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - if (maxfwidx-SourceFWRuleAbsIdxs[idx]+1<=maxdepth2) { - ostringstream nusource; - nusource << TD::Convert(SourceFWRuleIdxs[3*idx-1]) << "/" << ((maxfwidx-SourceFWRuleAbsIdxs[idx]+1)*-1); - ScoreOrientationLeftBackward(table,oril,TD::Convert(nusource.str()),SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - isBonus = false; - if ((aligned && lat<=fwblock[3])|| - (!aligned && las<=SourceFWRuleIdxs[3*idx-2])) isBonus=true; - if (SourceFWRuleAbsIdxs[idx]<=maxdepth1) { - ostringstream nusource; - nusource << TD::Convert(SourceFWRuleIdxs[3*idx-1]) << "/" << SourceFWRuleAbsIdxs[idx]; - ScoreOrientationRightBackward(table,orir,SourceFWRuleIdxs[3*idx-1],SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - if (maxfwidx-SourceFWRuleAbsIdxs[idx]+1<=maxdepth2) { - ostringstream nusource; - nusource << TD::Convert(SourceFWRuleIdxs[3*idx-1]) << "/" << ((maxfwidx-SourceFWRuleAbsIdxs[idx]+1)*-1); - ScoreOrientationRightBackward(table,orir,SourceFWRuleIdxs[3*idx-1],SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - delete fwblock; - } - for (int i_ant=0; i_ant<_Arity; i_ant++) { - // antfas -> first aligned source word antecedent-wise - // antlas -> last aligned source word antecedent-wise - int antfat = firstTargetAligned(TargetAntsIdxs[i_ant][1]); - int antlat = lastTargetAligned(TargetAntsIdxs[i_ant][TargetAntsIdxs[i_ant][0]]); - int antfas = firstSourceAligned(SourceAntsIdxs[i_ant][1]); - int antlas = lastSourceAligned(SourceAntsIdxs[i_ant][SourceAntsIdxs[i_ant][0]]); - assert(antfat <= antlat); - assert(antfas <= antlas); - for (int idx=1; idx<=SourceFWAntsIdxs[i_ant][0]; idx++) { - if (DEBUG) - cerr << "considering SourceFWAntsIdxs[" << i_ant << "][" << idx << "]: " << SourceFWAntsIdxs[i_ant][3*idx-2] << endl; - if (!(SourceFWAntsAbsIdxs[i_ant][idx]<=maxdepth1 || maxfwidx-SourceFWAntsAbsIdxs[i_ant][idx]+1<=maxdepth2)) continue; - int* fwblock = blockSource(SourceFWAntsIdxs[i_ant][3*idx-2],SourceFWAntsIdxs[i_ant][3*idx-2]); - //bool aligned = (minTSpan(SourceFWAntsIdxs[i_ant][3*idx-2])!=MINIMUM_INIT); - bool aligned = (fwblock[2]!=MINIMUM_INIT); - bool Lcompute = true; bool Rcompute = true; - if (DEBUG) { - cerr << " aligned = " << aligned << endl; - cerr << " fwblock = " << fwblock[0] << "," << fwblock[1] << "," << fwblock[2] << "," << fwblock[3] << endl; - cerr << " antfas=" << antfas << ", antlas=" << antlas << ", antfat=" << antfat << ", antlat=" << antlat << endl; - } - if (aligned) { - if (DEBUG) cerr << "laligned" << endl; - if (antfat first aligned source word, las -> last aligned source word - if (SourceFWRuleIdxs[3*idx-2]<=fas) isBonus=true; - if (!isBonus) // this is unnecessary because fas <= las assertion - if (minTSpan(SourceFWRuleIdxs[3*idx-2])==MINIMUM_INIT && las<=SourceFWRuleIdxs[3*idx-2]) isBonus=true; - if (maxdepth1>0) { - oss << source << "/"; - if (SourceFWRuleAbsIdxs[idx]<=maxdepth1) - oss << SourceFWRuleAbsIdxs[idx]; - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationLeft(table,oril,sourceID,SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - if (maxdepth2>0) { - oss << source << "/"; - if (maxfwidx-SourceFWRuleAbsIdxs[idx]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWRuleAbsIdxs[idx]+1)*-1); - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationLeft(table,oril,sourceID,SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - isBonus = false; - if (las<=SourceFWRuleIdxs[3*idx-2]) isBonus=true; - if (!isBonus) // this is unnecessary becuase fas <= las assertion - if (minTSpan(SourceFWRuleIdxs[3*idx-2])==MINIMUM_INIT && SourceFWRuleIdxs[3*idx-2]<=fas) isBonus=true; - if (maxdepth1>0) { - oss << source << "/"; - if (SourceFWRuleAbsIdxs[idx]<=maxdepth1) - oss << SourceFWRuleAbsIdxs[idx]; - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationRight(table,orir,sourceID,SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - if (maxdepth2>0) { - oss << source << "/"; - if (maxfwidx-SourceFWRuleAbsIdxs[idx]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWRuleAbsIdxs[idx]+1)*-1); - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationRight(table,orir,sourceID,SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - - } - for (int i_ant=0; i_ant<_Arity; i_ant++) { - for (int idx=1; idx<=SourceFWAntsIdxs[i_ant][0]; idx++) { - if (DEBUG) - cerr << "considering SourceFWAntsIdxs[" << i_ant << "][" << idx << "]: " << SourceFWAntsIdxs[i_ant][3*idx-2] << endl; - //if (!((SourceFWAntsAbsIdxs[i_ant][idx]<=maxdepth1)||(maxfwidx-SourceFWAntsAbsIdxs[i_ant][idx]+1<=maxdepth2))) continue; - // antfas -> first aligned source word antecedent-wise - // antlas -> last aligned source word antecedent-wise - int antfas = firstSourceAligned(SourceAntsIdxs[i_ant][1]); - int antlas = lastSourceAligned(SourceAntsIdxs[i_ant][SourceAntsIdxs[i_ant][0]]); - if (DEBUG) cerr << " SourceFWAntsAbsIdxs[i_ant][3*idx-1]=" << SourceFWAntsAbsIdxs[i_ant][3*idx-1] << endl; - string source = TD::Convert(SourceFWAntsIdxs[i_ant][3*idx-1]); - assert(antfas <= antlas); - bool aligned = (minTSpan(SourceFWAntsIdxs[i_ant][3*idx-2])!=MINIMUM_INIT); - bool Lcompute = true;bool Rcompute = true; - if ((aligned && antfas0) { - oss << source << "/"; - if (SourceFWAntsAbsIdxs[i_ant][idx]<=maxdepth1) - oss << SourceFWAntsAbsIdxs[i_ant][idx]; - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationLeft(table,oril,sourceID,SourceFWAntsIdxs[i_ant][3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - if (maxdepth2>0) { - oss << source << "/"; - if (maxfwidx-SourceFWAntsAbsIdxs[i_ant][idx]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWAntsAbsIdxs[i_ant][idx]+1)*-1); - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationLeft(table,oril,sourceID,SourceFWAntsIdxs[i_ant][3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - } - isBonus = false; - if (Rcompute) { - if (las<=SourceFWAntsIdxs[i_ant][3*idx-2]) isBonus = true; - //if (!isBonus) // this is unnecessary - // if (!aligned && SourceFWAntsIdxs[i_ant][3*idx-2]<=fas) isBonus=true; - if (maxdepth1>0) { - oss << source << "/"; - if (SourceFWAntsAbsIdxs[i_ant][idx]<=maxdepth1) - oss << SourceFWAntsAbsIdxs[i_ant][idx]; - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationRight(table,orir,sourceID,SourceFWAntsIdxs[i_ant][3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - if (maxdepth2>0) { - oss << source << "/"; - if (maxfwidx-SourceFWAntsAbsIdxs[i_ant][idx]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWAntsAbsIdxs[i_ant][idx]+1)*-1); - else - oss << "X"; - sourceID = TD::Convert(oss.str()); - oss.str(""); - ScoreOrientationRight(table,orir,sourceID,SourceFWAntsIdxs[i_ant][3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - } - } - } -} - -void Alignment::computeOrientationSource(const CountTable& table, double *cost, double *bonus, - double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus) { -// a bit complex due to imperfect state (TO DO!!!) -// 1. there are cases where function word alignments come from antecedents, which orientation -// (either its left or its right) has been computed earlier. -// 2. some orientation will go as bonus - if (DEBUG) cerr << "computeOrientationSource" << endl; - int oril, orir; - for (int idx=1; idx<=SourceFWRuleIdxs[0]; idx++) { - if (DEBUG) cerr << "considering SourceFWRuleIdxs[" << idx << "]: " << SourceFWRuleIdxs[3*idx-2] << endl; - OrientationSource(SourceFWRuleIdxs[3*idx-2],&oril,&orir); - bool isBonus = false; // fas -> first aligned source word, las -> last aligned source word - if (SourceFWRuleIdxs[3*idx-2]<=fas) isBonus=true; - if (!isBonus) // this is unnecessary because fas <= las assertion - if (minTSpan(SourceFWRuleIdxs[3*idx-2])==MINIMUM_INIT && las<=SourceFWRuleIdxs[3*idx-2]) isBonus=true; - ScoreOrientationLeft(table,oril,SourceFWRuleIdxs[3*idx-1],SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - isBonus = false; - if (las<=SourceFWRuleIdxs[3*idx-2]) isBonus=true; - if (!isBonus) // this is unnecessary becuase fas <= las assertion - if (minTSpan(SourceFWRuleIdxs[3*idx-2])==MINIMUM_INIT && SourceFWRuleIdxs[3*idx-2]<=fas) isBonus=true; - ScoreOrientationRight(table,orir,SourceFWRuleIdxs[3*idx-1],SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - for (int i_ant=0; i_ant<_Arity; i_ant++) { - for (int idx=1; idx<=SourceFWAntsIdxs[i_ant][0]; idx++) { - if (DEBUG) - cerr << "considering SourceFWAntsIdxs[" << i_ant << "][" << idx << "]: " << SourceFWAntsIdxs[i_ant][3*idx-2] << endl; - // antfas -> first aligned source word antecedent-wise - // antlas -> last aligned source word antecedent-wise - int antfas = firstSourceAligned(SourceAntsIdxs[i_ant][1]); - int antlas = lastSourceAligned(SourceAntsIdxs[i_ant][SourceAntsIdxs[i_ant][0]]); - assert(antfas <= antlas); - bool aligned = (minTSpan(SourceFWAntsIdxs[i_ant][3*idx-2])!=MINIMUM_INIT); - bool Lcompute = true;bool Rcompute = true; - if ((aligned && antfas& tags) { - if (DEBUG) cerr << "computeOrientationSourceGen" << endl; - int oril, orir; - for (int idx=1; idx<=SourceFWRuleIdxs[0]; idx++) { - if (DEBUG) cerr << "considering SourceFWRuleIdxs[" << idx << "]: " << SourceFWRuleIdxs[3*idx-2] << endl; - OrientationSource(SourceFWRuleIdxs[3*idx-2],&oril,&orir); - bool isBonus = false; // fas -> first aligned source word, las -> last aligned source word - if (SourceFWRuleIdxs[3*idx-2]<=fas) isBonus=true; - if (!isBonus) // this is unnecessary because fas <= las assertion - if (minTSpan(SourceFWRuleIdxs[3*idx-2])==MINIMUM_INIT && las<=SourceFWRuleIdxs[3*idx-2]) isBonus=true; - ScoreOrientationLeft(table,oril,generalize(SourceFWRuleIdxs[3*idx-1],tags),SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - isBonus = false; - if (las<=SourceFWRuleIdxs[3*idx-2]) isBonus=true; - if (!isBonus) // this is unnecessary becuase fas <= las assertion - if (minTSpan(SourceFWRuleIdxs[3*idx-2])==MINIMUM_INIT && SourceFWRuleIdxs[3*idx-2]<=fas) isBonus=true; - ScoreOrientationRight(table,orir,generalize(SourceFWRuleIdxs[3*idx-1],tags),SourceFWRuleIdxs[3*idx], - isBonus,cost,bonus,bo1,bo1_bonus,bo2,bo2_bonus,alpha_oris,beta_oris); - } - for (int i_ant=0; i_ant<_Arity; i_ant++) { - for (int idx=1; idx<=SourceFWAntsIdxs[i_ant][0]; idx++) { - if (DEBUG) - cerr << "considering SourceFWAntsIdxs[" << i_ant << "][" << idx << "]: " << SourceFWAntsIdxs[i_ant][3*idx-2] << endl; - // antfas -> first aligned source word antecedent-wise - // antlas -> last aligned source word antecedent-wise - int antfas = firstSourceAligned(SourceAntsIdxs[i_ant][1]); - int antlas = lastSourceAligned(SourceAntsIdxs[i_ant][SourceAntsIdxs[i_ant][0]]); - assert(antfas <= antlas); - bool aligned = (minTSpan(SourceFWAntsIdxs[i_ant][3*idx-2])!=MINIMUM_INIT); - bool Lcompute = true;bool Rcompute = true; - if ((aligned && antfas0) { - if (lfw>=0) { - int dom = DominanceSource(0,SourceFWIdxs[1]); - if (DEBUG) cerr << " --> lfw = " << lfw << "-" << TD::Convert(lfw) << endl; - if (DEBUG) cerr << " --> rfw = " << rfw << "-" << TD::Convert(rfw) << endl; - ScoreDominance(table,dom,lfw,SourceFWIdxs[2],lfw,SourceFWIdxs[3],bonus,bo1_bonus,bo2_bonus,true,alpha_doms,beta_doms); - } - if (rfw>=0) { - int dom = DominanceSource(SourceFWIdxs[3*SourceFWIdxs[0]-2],_J-1); - ScoreDominance(table,dom,SourceFWIdxs[3*SourceFWIdxs[0]-1],rfw,SourceFWIdxs[3*SourceFWIdxs[0]], - rfw,bonus,bo1_bonus,bo2_bonus,true,alpha_doms,beta_doms); - } - } -} - -void Alignment::computeDominanceSourcePos(const CountTable& table, WordID lfw, WordID rfw, - double *cost, double *bonus, double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, int maxfwidx, int maxdepth1, int maxdepth2) { - if (DEBUG) cerr << "computeDominanceSourcePos" << endl; - if (DEBUG) cerr << " initial cost=" << *cost << ", initial bonus=" << *bonus << endl; - ostringstream oss; - for (int idx=2; idx<=SourceFWIdxs[0]; idx++) { - if (DEBUG) { - cerr << "PrevSourceFWIdxs :" << SourceFWIdxs[3*(idx-1)-2] << "," << SourceFWIdxs[3*(idx-1)-1] - << "," << SourceFWIdxs[3*(idx-1)] << endl; - cerr << "CurrSourceFWIdxs :" << SourceFWIdxs[3*(idx)-2] << "," << SourceFWIdxs[3*(idx)-1] - << "," << SourceFWIdxs[3*(idx)] << endl; - } - //if (!((SourceFWAbsIdxs[3*(idx-1)-2]<=maxdepth1 && SourceFWAbsIdxs[3*idx-2]<=maxdepth1) || - // (maxfwidx-SourceFWAbsIdxs[3*(idx-1)-2]+1<=maxdepth2 && maxfwidx-SourceFWAbsIdxs[3*idx-2]+1<=maxdepth2))) continue; - bool compute = true; - for (int i_ant=0; i_ant<_Arity && compute; i_ant++) { - if (MemberOf(SourceFWAntsIdxs[i_ant],SourceFWIdxs[3*(idx-1)-2],SourceFWIdxs[3*(idx)-2])) { - //cerr << "Skipping, they have been calculated in the " << (i_ant+1) << "-th branch" << endl; - compute=false; - } - } - if (compute) { - int dom = DominanceSource(SourceFWIdxs[3*(idx-1)-2],SourceFWIdxs[3*idx-2]); - if (DEBUG) cerr << "dom = " << dom << endl; - if (maxdepth1+maxdepth2>0) { - string source1 = TD::Convert(SourceFWIdxs[3*(idx-1)-1]); - string source2 = TD::Convert(SourceFWIdxs[3*(idx)-1]); - if (maxdepth1>0) { - oss << source1 << "/"; - if (SourceFWAbsIdxs[3*(idx-1)-2]<=maxdepth1) - oss << SourceFWAbsIdxs[3*(idx-1)-2]; - else - oss << "X"; - WordID source1id = TD::Convert(oss.str()); - oss.str(""); - oss << source2 << "/"; - if (SourceFWAbsIdxs[3*idx-2]<=maxdepth1) - oss << SourceFWAbsIdxs[3*idx-2]; - else - oss << "X"; - WordID source2id = TD::Convert(oss.str()); - oss.str(""); - ScoreDominance(table,dom,source1id,source2id,SourceFWIdxs[3*(idx-1)],SourceFWIdxs[3*idx], - cost,bo1,bo2,false,alpha_doms,beta_doms); - } - if (maxdepth2>0) { - oss << source1 << "/"; - if (maxfwidx-SourceFWAbsIdxs[3*(idx-1)-2]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWAbsIdxs[3*(idx-1)-2]+1)*-1); - else - oss << "X"; - WordID source1id = TD::Convert(oss.str()); - oss.str(""); - oss << source2 << "/"; - if (maxfwidx-SourceFWAbsIdxs[3*idx-2]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWAbsIdxs[3*(idx-1)-2]+1)*-1); - else - oss << "X"; - WordID source2id = TD::Convert(oss.str()); - oss.str(""); - ScoreDominance(table,dom,source1id,source2id,SourceFWIdxs[3*(idx-1)],SourceFWIdxs[3*idx], - cost,bo1,bo2,false,alpha_doms,beta_doms); - } - } - } - } - if (SourceFWIdxs[0]>0) { - if (lfw>=0) { - int dom = DominanceSource(0,SourceFWIdxs[1]); - string source1 = TD::Convert(lfw); - string source2 = TD::Convert(SourceFWIdxs[2]); - if (maxdepth1>0) { - oss << source1 << "/"; - if (SourceFWAbsIdxs[1]-1<=maxdepth1) - oss << (SourceFWAbsIdxs[1]-1); - else - oss << "X"; - WordID source1id = TD::Convert(oss.str()); - oss.str(""); - oss << source2 << "/"; - if (SourceFWAbsIdxs[1]<=maxdepth1) - oss << SourceFWAbsIdxs[1]; - else - oss << "X"; - WordID source2id = TD::Convert(oss.str()); - oss.str(""); - ScoreDominance(table,dom,source1id,source2id,lfw,SourceFWIdxs[3],bonus,bo1_bonus,bo2_bonus,true,alpha_doms,beta_doms); - } - if (maxdepth2>0) { - oss << source1 << "/"; - if (maxfwidx-(SourceFWAbsIdxs[1]-1)+1<=maxdepth2) - oss << ((maxfwidx-(SourceFWAbsIdxs[1]-1)+1)*-1); - else - oss << "X"; - WordID source1id = TD::Convert(oss.str()); - oss.str(""); - oss << source2 << "/"; - if (maxfwidx-SourceFWAbsIdxs[1]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWAbsIdxs[1]+1)*-1); - else - oss << "X"; - WordID source2id = TD::Convert(oss.str()); - oss.str(""); - ScoreDominance(table,dom,source1id,source2id,lfw,SourceFWIdxs[3],bonus,bo1_bonus,bo2_bonus,true,alpha_doms,beta_doms); - } - } - if (rfw>=0) { - int dom = DominanceSource(SourceFWIdxs[3*SourceFWIdxs[0]-2],_J-1); - string source1 = TD::Convert(SourceFWIdxs[3*SourceFWIdxs[0]-1]); - string source2 = TD::Convert(rfw); - if (maxdepth1>0) { - oss << source1 << "/"; - if (SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]<=maxdepth1) - oss << SourceFWAbsIdxs[3*SourceFWIdxs[0]-2]; - else - oss << "X"; - WordID source1id = TD::Convert(oss.str()); - oss.str(""); - oss << source2 << "/"; - if (SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]+1<=maxdepth1) - oss << (SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]+1); - else - oss << "X"; - WordID source2id = TD::Convert(oss.str()); - ScoreDominance(table,dom,source1id,source2id,SourceFWIdxs[3*SourceFWIdxs[0]], - rfw,bonus,bo1_bonus,bo2_bonus,true,alpha_doms,beta_doms); - } - if (maxdepth2>0) { - oss << source1 << "/"; - if (maxfwidx-SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]+1<=maxdepth2) - oss << ((maxfwidx-SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]+1)*-1); - else - oss << "X"; - WordID source1id = TD::Convert(oss.str()); - oss.str(""); - oss << source2 << "/"; - if (maxfwidx-(SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]+1)+1<=maxdepth2) - oss << ((maxfwidx-(SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]+1)+1)*-1); - else - oss << "X"; - WordID source2id = TD::Convert(oss.str()); - oss.str(""); - ScoreDominance(table,dom,source1id,source2id,SourceFWIdxs[3*SourceFWIdxs[0]], - rfw,bonus,bo1_bonus,bo2_bonus,true,alpha_doms,beta_doms); - } - } - } -} - - -void Alignment::computeDominanceTarget(const CountTable& table, WordID lfw, WordID rfw, - double *cost, double *bonus, double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus) { - if (DEBUG) cerr << "computeDominanceTarget" << endl; - for (int idx=2; idx<=TargetFWIdxs[0]; idx++) { - if (DEBUG) cerr << "PrevTargetFWIdxs :" << TargetFWIdxs[3*(idx-1)-2] << "," << TargetFWIdxs[3*(idx-1)-1] << "," <=0) { - int dom = DominanceTarget(0,TargetFWIdxs[1]); - if (DEBUG) cerr << "dom target (with left) = " << dom << endl; - ScoreDominance(table,dom,lfw,lfw,TargetFWIdxs[2],TargetFWIdxs[3],bonus,bo1_bonus,bo2_bonus,true,alpha_domt,beta_domt); - } - if (rfw>=0) { - int dom = DominanceTarget(TargetFWIdxs[3*TargetFWIdxs[0]-2],_I-1); - if (DEBUG) cerr << "dom target (with right) = " << dom << endl; - ScoreDominance(table,dom,TargetFWIdxs[3*TargetFWIdxs[0]-1],TargetFWIdxs[3*TargetFWIdxs[0]], - rfw,rfw,bonus,bo1_bonus,bo2_bonus,true,alpha_domt,beta_domt); - } - } - - //cerr << "END of computeDominanceTarget" << endl; -} - -double Alignment::ScoreDominance(const CountTable& table, int dom, WordID source1, WordID source2, WordID target1, WordID target2) { - if (DEBUG) { - cerr << "ScoreDominance(source1=" << TD::Convert(source1) << ",source2=" << TD::Convert(source2) - << ",target1=" << TD::Convert(target1) << ",target2=" << TD::Convert(target2) << ", dom=" << dom << endl; - } - string _source1 = TD::Convert(source1); - string _source2 = TD::Convert(source2); - string _source1idx; string _source2idx; - if (table.mode==1) { - _source1idx = _source1; _source2idx = _source2; - _source1 = _source1idx.substr(0,_source1idx.find_last_of("/")); - _source2 = _source2idx.substr(0,_source2idx.find_last_of("/")); - } - string _target1 = TD::Convert(target1); - string _target2 = TD::Convert(target2); - - double count = table.ultimate[dom]; - double total = table.ultimate[4]; - double prob = count/total; - if (DEBUG) cerr << "level0 " << count << "/" << total << "=" << prob << endl; - double alpha = 0.1; - - string key = _source1 + " " + _source2; - WordID key_id = TD::Convert(key); - map::const_iterator it = table.model.find(key_id); - bool stop = (it==table.model.end()); - if (!stop) { - stop = true; - if (it->second[4]>=0) { - count = it->second[dom] + alpha*prob; - total = it->second[4] + alpha; - prob = count/total; - if (DEBUG) cerr << "level1 " << count << "/" << total << "=" << prob << endl; - stop = false; - } - } - if (stop) return prob; - - key = _source1 + " " + _source2 + " " + _target1 + " " + _target2; - key_id = TD::Convert(key); - it = table.model.find(key_id); - stop = (it==table.model.end()); - if (!stop) { - stop = true; - if (it->second[4]>=0) { - count = it->second[dom] + alpha*prob; - total = it->second[4] + alpha; - prob = count/total; - if (DEBUG) cerr << "level2 " << count << "/" << total << "=" << prob << endl; - stop = false; - } - } - - if (table.mode!=1 || stop) return prob; - key = _source1 + " " + _source2 + " " + _target1 + " " + _target2; - key_id = TD::Convert(key); - it = table.model.find(key_id); - if (it!=table.model.end()) { - if (it->second[4]>=0) { - count = it->second[dom] + alpha*prob; - total = it->second[4] + alpha; - if (DEBUG) cerr << "level3 " << count << "/" << total << "=" << prob << endl; - prob = count/total; - } - } - - return prob; -} - -void Alignment::ScoreDominance(const CountTable& table, int dom, WordID source1, WordID source2, WordID target1, WordID target2, double *cost, double *bo1, double *bo2, bool isBonus, double alpha2, double beta2) { - if (DEBUG) - cerr << "ScoreDominance(source1=" << TD::Convert(source1) << ",source2=" << TD::Convert(source2) - << ",target1=" << TD::Convert(target1) << ",target2=" << TD::Convert(target2) << ",isBonus=" << isBonus << ", alpha2 = " << alpha2 << ", beta2 = " << beta2 << endl; - if (DEBUG) cerr << " BEFORE=" << *cost << endl; - *cost += ScoreDominance(table,dom,source1,source2,target1,target2); - if (DEBUG) cerr << " AFTER=" << *cost << endl; -} - -WordID Alignment::F2EProjectionFromExternal(int idx, const vector& als, const string& delimiter) { - if (DEBUG) { - cerr << "F2EProjectionFromExternal=" << idx << endl; - for (int i=0; i< als.size(); i++) cerr << "als[" << i << "]=" << als[i] << " "; - cerr << endl; - } - vector alignedTo; - for (int i=0; i0) projection << delimiter; - projection << TD::Convert(_e[alignedTo[i]]); - } - if (DEBUG) { - cerr << "projection = " << projection.str() << endl; - cerr << "returns = " << TD::Convert(projection.str()) << endl; - } - return TD::Convert(projection.str()); - } -} - -WordID Alignment::E2FProjectionFromExternal(int idx, const vector& als, const string& delimiter) { - vector alignedTo; - for (int i=0; i0) projection << delimiter; - projection << TD::Convert(_f[alignedTo[i]]); - } - return TD::Convert(projection.str()); - } -} - - -WordID Alignment::F2EProjection(int idx, const string& delimiter) { - if (DEBUG) cerr << "F2EProjection(" << idx << ")" << endl; - int e = targetOf(idx); - if (e<0) { - if (DEBUG) cerr << "projection = NULL" << endl; - return TD::Convert("NULL"); - } else { - if (targetOf(idx,e+1)<0) { - if (DEBUG) cerr << "e-1=" << (e-1) << ", size=" << _e.size() << endl; - return getE(e-1); // if not aligned to many, why bother continuing - } - ostringstream projection; - bool firstTime = true; - do { - if (!firstTime) projection << delimiter; - projection << TD::Convert(_e[e-1]); // transform space - firstTime = false; - e = targetOf(idx,e+1); - //if (DEBUG) cerr << "projection = " << projection.str() << endl; - } while(e>=0); - return TD::Convert(projection.str()); - } -} - -WordID Alignment::E2FProjection(int idx, const string& delimiter) { - //cerr << "E2FProjection(" << idx << ")" << endl; - //cerr << "i" << endl; - int f = sourceOf(idx); - //cerr << "j, f=" << f << endl; - if (f<0) { - //cerr << "projection = NULL" << endl; - return TD::Convert("NULL"); - } else { - if (sourceOf(idx,f+1)<0) return getF(f-1); - bool firstTime = true; - ostringstream projection(ostringstream::out); - do { - if (!firstTime) projection << delimiter; - projection << TD::Convert(_f[f-1]); //transform space - firstTime = false; - f = sourceOf(idx,f+1); - //cerr << "projection = " << projection.str() << endl; - } while(f>=0); - return TD::Convert(projection.str()); - } -} -void Alignment::computeBorderDominanceSource(const CountTable& table, double *cost, double *bonus, double *state_mono, - double *state_nonmono, TRule &rule, const std::vector& ant_contexts, const map& sfw) { - // HACK: GOAL is assumed to always be "S" - if (DEBUG) cerr << "computeBorderDominanceSource" << endl; - std::vector f = rule.f(); - std::vector e = rule.e(); - int nt_index[f.size()]; - int nt_count=0; - for (int i=0; i als; - for (std::vector::const_iterator i = rule.als().begin(); i != rule.als().end(); ++i) { - int s = i->s_; int t = i->t_; - als.push_back(link(t,s)); - } - if (DEBUG) cerr << "rule.Arity=" << rule.Arity() << endl; - if (rule.Arity()>0) { - int ntc=0; - for (int s=0; s neither, 1 -> leftFirst, 2 -> rightFirst, 3 -> dontCare - // ScoreDominance(const CountTable& table, int dom, WordID source1, WordID source2, WordID target1, WordID target2) - int prevs = 0; - for (int i=0; i(ant_contexts[nt_index[s]-1]); - *cost += Dwarf::IntegerToDouble(ants[51]); // 50->mono, 51->non-mono - if (DEBUG) cerr << " adding "<< Dwarf::IntegerToDouble(ants[51]) << " into cost, resulting = " << *cost << endl; - } - flag[s] = true; - } - } - prevs = currs; - } - if (DEBUG) cerr << "bonus and state matter" << endl; - for (int s=0; s(ant_contexts[nt_index[s]-1]); - double indbonus = Dwarf::IntegerToDouble(ants[50]); - *bonus += indbonus; - *state_mono += indbonus; - *state_nonmono += Dwarf::IntegerToDouble(ants[51]); - if (DEBUG) cerr << " propagating state=" << *state_mono <<","<< *state_nonmono<< endl; - } - } - } - if (DEBUG) cerr << "LHS:" << rule.GetLHS() << ":" << TD::Convert(rule.GetLHS()*-1) <(ant_contexts[i]); - *cost += Dwarf::IntegerToDouble(ants[50]); - } - *bonus = 0; - } - if (DEBUG) cerr << "-->>>> cost="<<*cost<<", bonus="<<*bonus<<", state_mono="<<*state_mono<<", state_nonmono="<<*state_nonmono<& ant_contexts, const map& sfw, const map& tfw,const Lattice& sourcelattice, int spanstart, int spanend) { - if (DEBUG) cerr << "===Rule===" << rule.AsString() << endl; - _f = rule.f(); - _e = rule.e(); - _Arity = rule.Arity(); - if (DEBUG) { - cerr << "F: "; - for (int idx=0; idx<_f.size(); idx++) cerr << _f[idx] << " "; - cerr << endl; - cerr << "F': "; - for (int idx=0; idx<_f.size(); idx++) - if (_f[idx]>=0) { - cerr << TD::Convert(_f[idx]) << " "; - } else { - cerr << TD::Convert(_f[idx]*-1); - } - cerr << endl; - cerr << "E: "; - for (int idx=0; idx<_e.size(); idx++) - cerr << _e[idx] << " "; - cerr << endl; - cerr << "E': "; - for (int idx=0; idx<_e.size(); idx++) - if (_e[idx]>0) { - cerr << TD::Convert(_e[idx]) << " "; - } else { - cerr << "[NT]" << " "; - } - cerr << endl; - } - - SourceFWRuleIdxs[0]=0; - SourceFWRuleAbsIdxs[0]=0; - for (int idx=1; idx<=_f.size(); idx++) { // in transformed space - if (sfw.find(_f[idx-1])!=sfw.end()) { - SourceFWRuleIdxs[0]++; - SourceFWRuleAbsIdxs[++SourceFWRuleAbsIdxs[0]]=GetFWGlobalIdx(idx,sourcelattice,_f,spanstart,spanend,ant_contexts,sfw); - SourceFWRuleIdxs[3*SourceFWRuleIdxs[0]-2]=idx; - SourceFWRuleIdxs[3*SourceFWRuleIdxs[0]-1]=_f[idx-1]; - SourceFWRuleIdxs[3*SourceFWRuleIdxs[0]] =F2EProjectionFromExternal(idx-1,rule.a_,"_SEP_"); - } - } - TargetFWRuleIdxs[0]=0; - for (int idx=1; idx<=_e.size(); idx++) { // in transformed space - if (tfw.find(_e[idx-1])!=tfw.end()) { - TargetFWRuleIdxs[0]++; - TargetFWRuleIdxs[3*TargetFWRuleIdxs[0]-2]=idx; - TargetFWRuleIdxs[3*TargetFWRuleIdxs[0]-1]=E2FProjectionFromExternal(idx-1,rule.a_,"_SEP_"); - TargetFWRuleIdxs[3*TargetFWRuleIdxs[0]] =_e[idx-1]; - } - } - - if (DEBUG) { - cerr << "SourceFWRuleIdxs[" << SourceFWRuleIdxs[0] << "]:"; - for (int idx=1; idx<=SourceFWRuleIdxs[0]; idx++) { - cerr << " idx:" << SourceFWRuleIdxs[3*idx-2]; - cerr << " absidx:" << SourceFWRuleAbsIdxs[idx]; - cerr << " F:" << SourceFWRuleIdxs[3*idx-1]; - cerr << " E:" << SourceFWRuleIdxs[3*idx]; - cerr << "; "; - } - cerr << endl; - cerr << "TargetFWRuleIdxs[" << TargetFWRuleIdxs[0] << "]:"; - for (int idx=1; idx<=TargetFWRuleIdxs[0]; idx++) { - cerr << " idx:" << TargetFWRuleIdxs[3*idx-2]; - cerr << " F:" << TargetFWRuleIdxs[3*idx-1]; - cerr << " E:" << TargetFWRuleIdxs[3*idx]; - } - cerr << endl; - } - if (SourceFWRuleIdxs[0]+TargetFWRuleIdxs[0]==0) { - bool nofw = true; - for (int i_ant=0; i_ant<_Arity && nofw; i_ant++) { - const int* ants = reinterpret_cast(ant_contexts[i_ant]); - if (ants[0]>=0||ants[3]>=0||ants[6]>=0||ants[9]>=0) nofw=false; - } - if (nofw) return true; - } - //cerr << "clearing als first" << endl; - clearAls(_J,_I); - - if (DEBUG) cerr << "A["<< rule.a_.size() << "]: " ; - RuleAl[0]=0; - // add phrase start boundary - RuleAl[0]++; RuleAl[RuleAl[0]*2-1]=0; RuleAl[RuleAl[0]*2]=0; - if (DEBUG) cerr << RuleAl[RuleAl[0]*2-1] << "-" << RuleAl[RuleAl[0]*2] << " "; - for (int idx=0; idx(ant_contexts[i_ant]); - int span = ants[Dwarf::STATE_SIZE-1]; - if (DEBUG) { - cerr << "antcontexts[" << i_ant << "] "; - for (int idx=0; idx=0) { - // Given a span, give the index of the first function word - int firstfwidx = GetFirstFWIdx(source(span),target(span),sourcelattice,sfw); - if (DEBUG) cerr << " firstfwidx = " << firstfwidx << endl; - int fwcount = 0; - if (ants[1]>=0) { // one function word - SourceFWAntsIdxs[i_ant][0]++; SourceFWAntsIdxs[i_ant][1]=ants[0]; - SourceFWAntsIdxs[i_ant][2]=ants[1]; SourceFWAntsIdxs[i_ant][3]=ants[2]; - fwcount++; - } else { // if ants[1] < 0 then compound fws - //cerr << "ants[1]<0" << endl; - istringstream ossf(TD::Convert(ants[1]*-1)); string ffw; - istringstream osse(TD::Convert(ants[2])); string efw; //projection would be mostly NULL - int delta=ants[0]; - while (osse >> efw && ossf >> ffw) { - SourceFWAntsIdxs[i_ant][0]++; - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3-2]=ants[0]-(delta--); - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3-1]=TD::Convert(ffw); - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3] =TD::Convert(efw); - fwcount++; - } - } - if (DEBUG) cerr << " fwcount=" << fwcount << endl; - SourceFWAntsAbsIdxs[i_ant][0]=fwcount; - for (int i=1; i<=fwcount; i++) SourceFWAntsAbsIdxs[i_ant][i]=firstfwidx++; - } - if (ants[3]>=0) { - int lastfwidx = GetLastFWIdx(source(span),target(span),sourcelattice,sfw); - if (DEBUG) cerr << " lastfwidx = " << lastfwidx << endl; - int fwcount=0; - if (ants[4]>=0) { - fwcount++; - SourceFWAntsIdxs[i_ant][0]++; - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3-2]=ants[3]; - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3-1]=ants[4]; - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3] =ants[5]; - } else { // if ants[4] < 0 then compound fws - //cerr << "ants[4]<0" << endl; - istringstream ossf(TD::Convert(ants[4]*-1)); string ffw; - istringstream osse(TD::Convert(ants[5])); string efw; - int delta=0; - while (osse >> efw && ossf >> ffw) { - fwcount++; - SourceFWAntsIdxs[i_ant][0]++; - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3-2]=ants[3]+(delta++); - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3-1]=TD::Convert(ffw); - SourceFWAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][0]*3] =TD::Convert(efw); - } - } - if (DEBUG) cerr << " fwcount=" << fwcount << endl; - for (int i=1; i<=fwcount; i++) SourceFWAntsAbsIdxs[i_ant][SourceFWAntsAbsIdxs[i_ant][0]+i]=lastfwidx-fwcount+i; - SourceFWAntsAbsIdxs[i_ant][0]+=fwcount; - } - TargetFWAntsIdxs[i_ant][0]=0; - if (ants[6]>=0) { - if (ants[8]>=0) { // check the e part - TargetFWAntsIdxs[i_ant][0]++; - TargetFWAntsIdxs[i_ant][1]=ants[6]; - TargetFWAntsIdxs[i_ant][2]=ants[7]; - TargetFWAntsIdxs[i_ant][3]=ants[8]; - } else { // if ants[8] < 0 then compound fws - //cerr << "ants[8]<0" << endl; - //cerr << "ants[7]=" << TD::Convert(ants[7]) << endl; - //cerr << "ants[8]=" << TD::Convert(ants[8]*-1) << endl; - istringstream ossf(TD::Convert(ants[7])); string ffw; - istringstream osse(TD::Convert(ants[8]*-1)); string efw; - int delta=ants[6]; - while (osse >> efw && ossf >> ffw) { - //cerr << "efw="<< efw << ",ffw=" << ffw << endl; - TargetFWAntsIdxs[i_ant][0]++; - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3-2]=ants[6]-(delta--); - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3-1]=TD::Convert(ffw); - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3] =TD::Convert(efw); - } - } - } - if (ants[9]>=0) { - if (ants[11]>=0) { - TargetFWAntsIdxs[i_ant][0]++; - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3-2]=ants[9]; - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3-1]=ants[10]; - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3] =ants[11]; - } else { - //cerr << "ants[11]<0" << endl; - //cerr << "ants[10]=" << TD::Convert(ants[10]) << endl; - //cerr << "ants[11]=" << TD::Convert(ants[11]*-1) << endl; - istringstream ossf(TD::Convert(ants[10])); string ffw; - istringstream osse(TD::Convert(ants[11]*-1)); string efw; - int delta = 0; - while (osse >> efw && ossf >> ffw) { - //cerr << "efw="<< efw << ",ffw=" << ffw << endl; - TargetFWAntsIdxs[i_ant][0]++; - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3-2]=ants[9]+(delta++); - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3-1]=TD::Convert(ffw); - TargetFWAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][0]*3] =TD::Convert(efw); - } - } - } - AntsAl[i_ant][0]=ants[12];//number of alignments - for (int idx=1; idx<=AntsAl[i_ant][0]; idx++) { - AntsAl[i_ant][idx*2-1] = source(ants[12+idx]); - AntsAl[i_ant][idx*2] = target(ants[12+idx]); - } - } - - for (int i_ant=0; i_ant<_Arity; i_ant++) { - int length = AntsAl[i_ant][0]; - int maxs = -1000; - int maxt = -1000; - for (int idx=0; idx=0) { - TargetRuleIdxs[idx]=curr++; - } else { - int i_ant = TargetRuleIdxs[idx]*-1-1; - if (DEBUG) cerr << "TargetRuleIdxs[" << i_ant << "]" << endl; - for (int idx2=1; idx2<=TargetAntsIdxs[i_ant][0]; idx2++) { - TargetAntsIdxs[i_ant][idx2]=curr++; - if (DEBUG) cerr << TargetAntsIdxs[i_ant][idx2] << " "; - } - if (DEBUG) cerr << endl; - } - } - if (DEBUG) { - cerr << "TargetRuleIdxs" << endl; - for (int idx=1; idx<=TargetRuleIdxs[0]; idx++) cerr << TargetRuleIdxs[idx] << " "; - cerr << endl; - } - for (int idx=1; idx<=RuleAl[0]; idx++) { - if (DEBUG) { - cerr << RuleAl[idx*2-1] << " - " << RuleAl[idx*2] << " to "; - cerr << SourceRuleIdxs[RuleAl[idx*2-1]+1] << " - " << TargetRuleIdxs[RuleAl[idx*2]+1] << endl; - } - set(SourceRuleIdxs[RuleAl[idx*2-1]+1],TargetRuleIdxs[RuleAl[idx*2]+1]); - } - for (int i_ant=0; i_ant<_Arity; i_ant++) { - for (int idx=1; idx<=AntsAl[i_ant][0]; idx++) { - if (DEBUG) { - cerr << AntsAl[i_ant][2*idx-1] << " - " << AntsAl[i_ant][2*idx] << " to "; - cerr << SourceAntsIdxs[i_ant][AntsAl[i_ant][2*idx-1]+1] << " - "; - cerr << TargetAntsIdxs[i_ant][AntsAl[i_ant][2*idx]+1] << endl; - } - set(SourceAntsIdxs[i_ant][AntsAl[i_ant][2*idx-1]+1],TargetAntsIdxs[i_ant][AntsAl[i_ant][2*idx]+1]); - } - } - SourceFWIdxs[0]=0; - SourceFWAbsIdxs[0]=0; - if (DEBUG) cerr << "SourceFWRuleIdxs:" << endl; - for (int idx=1; idx<=SourceFWRuleIdxs[0]; idx++) { - if (DEBUG) cerr << SourceFWRuleIdxs[3*idx-2] << " to " << SourceRuleIdxs[SourceFWRuleIdxs[3*idx-2]+1] << endl; - SourceFWRuleIdxs[3*idx-2] = SourceRuleIdxs[SourceFWRuleIdxs[3*idx-2]+1]; - SourceFWAbsIdxs[0]++; - SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]=SourceFWRuleAbsIdxs[idx]; - SourceFWIdxs[0]++; - SourceFWIdxs[3*SourceFWIdxs[0]-2]=SourceFWRuleIdxs[3*idx-2]; - SourceFWIdxs[3*SourceFWIdxs[0]-1]=SourceFWRuleIdxs[3*idx-1]; - SourceFWIdxs[3*SourceFWIdxs[0]] =SourceFWRuleIdxs[3*idx]; - } - for (int i_ant=0; i_ant<_Arity; i_ant++) { - if (DEBUG) cerr << "SourceFWAntsIdxs[" << i_ant << "]" << endl; - for (int idx=1; idx<=SourceFWAntsIdxs[i_ant][0]; idx++) { - if (DEBUG) - cerr << SourceFWAntsIdxs[i_ant][3*idx-2] << " to " << SourceAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][3*idx-2]+1] << endl; - SourceFWAntsIdxs[i_ant][3*idx-2] = SourceAntsIdxs[i_ant][SourceFWAntsIdxs[i_ant][3*idx-2]+1]; - SourceFWAbsIdxs[0]++; - SourceFWAbsIdxs[3*SourceFWAbsIdxs[0]-2]=SourceFWAntsAbsIdxs[i_ant][idx]; - SourceFWIdxs[0]++; - SourceFWIdxs[3*SourceFWIdxs[0]-2]=SourceFWAntsIdxs[i_ant][3*idx-2]; - SourceFWIdxs[3*SourceFWIdxs[0]-1]=SourceFWAntsIdxs[i_ant][3*idx-1]; - SourceFWIdxs[3*SourceFWIdxs[0]] =SourceFWAntsIdxs[i_ant][3*idx]; - } - } - sort(SourceFWIdxs); - sort(SourceFWAbsIdxs); - if (DEBUG) { - cerr << "SourceFWIdxs : "; - for (int idx=1; idx<=SourceFWIdxs[0]; idx++) { - cerr << "idx:" << SourceFWIdxs[3*idx-2] << ","; - cerr << "F:" << SourceFWIdxs[3*idx-1] << ","; - cerr << "E:" << SourceFWIdxs[3*idx] << " "; - } - cerr << endl; - } - TargetFWIdxs[0]=0; - if (DEBUG) cerr << "TargetFWRuleIdxs:" << endl; - for (int idx=1; idx<=TargetFWRuleIdxs[0]; idx++) { - if (DEBUG) cerr << TargetFWRuleIdxs[3*idx-2] << " to " << TargetRuleIdxs[TargetFWRuleIdxs[3*idx-2]+1] << endl; - TargetFWRuleIdxs[3*idx-2] = TargetRuleIdxs[TargetFWRuleIdxs[3*idx-2]+1]; - TargetFWIdxs[0]++; - TargetFWIdxs[3*TargetFWIdxs[0]-2]=TargetFWRuleIdxs[3*idx-2]; - TargetFWIdxs[3*TargetFWIdxs[0]-1]=TargetFWRuleIdxs[3*idx-1]; - TargetFWIdxs[3*TargetFWIdxs[0]] =TargetFWRuleIdxs[3*idx]; - } - for (int i_ant=0; i_ant<_Arity; i_ant++) { - if (DEBUG) cerr << "TargetFWAntsIdxs[" << i_ant << "]" << endl; - for (int idx=1; idx<=TargetFWAntsIdxs[i_ant][0]; idx++) { - if (DEBUG) cerr << TargetFWAntsIdxs[i_ant][3*idx-2] << " to " << TargetAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][3*idx-2]+1] << endl; - TargetFWAntsIdxs[i_ant][3*idx-2] = TargetAntsIdxs[i_ant][TargetFWAntsIdxs[i_ant][3*idx-2]+1]; - TargetFWIdxs[0]++; - TargetFWIdxs[3*TargetFWIdxs[0]-2]=TargetFWAntsIdxs[i_ant][3*idx-2]; - TargetFWIdxs[3*TargetFWIdxs[0]-1]=TargetFWAntsIdxs[i_ant][3*idx-1]; - TargetFWIdxs[3*TargetFWIdxs[0]] =TargetFWAntsIdxs[i_ant][3*idx]; - } - } - sort(TargetFWIdxs); - if (DEBUG) { - cerr << "TargetFWIdxs : "; - for (int idx=1; idx<=TargetFWIdxs[0]; idx++) { - cerr << "idx:" << TargetFWIdxs[3*idx-2]<< ","; - cerr << "E:" << TargetFWIdxs[3*idx-1]<< ","; - cerr << "F:" << TargetFWIdxs[3*idx]<< " "; - } - cerr << endl; - cerr << AsString() << endl; - } - fas = firstSourceAligned(1); las = lastSourceAligned(_J-2); - fat = firstTargetAligned(1); lat = lastTargetAligned(_I-2); - if (DEBUG) cerr << "fas=" << fas << ", las=" << las << ", fat=" << fat << ", lat=" << lat << endl; - assert(fas<=las); - assert(fat<=lat); - SetCurrAlVector(); - if (DEBUG) cerr << "end prepare" << endl; - return false; -} - -string Alignment::AsStringSimple() { - ostringstream stream; - for (int j=0; j=0) { - stream << " " << j << "-" << t; - t = targetOf(j,t+1); - } - } - return stream.str(); -}; - - -string Alignment::AsString() { - ostringstream stream; - stream << "J:" << getJ() << " I:" << getI(); - for (int j=0; j=0) { - stream << " " << j << "-" << t; - t = targetOf(j,t+1); - } - } - stream << " TargetSpan:"; - for (int j=0; j=0) { - curr_al.push_back(link(j,i)); - i = targetOf(j,i+1); - } - } -} - -void CountTable::print() const { - cerr << "+++ Model +++" << endl; - for (map::const_iterator iter=model.begin(); iter!=model.end(); iter++) { - cerr << TD::Convert(iter->first) << " "; - for (int i=0; isecond[i] << " "; - cerr << endl; - } - cerr << "+++ Ultimate +++" << endl; - for (int i=0; i* ret) { - ret->clear(); - for (int i=0; i<_J; i++) { - int t = targetOf(i); - while (t>=0) { - ret->push_back(link(i,t)); - t = targetOf(i,t+1); - } - } -} - -int Alignment::GetFWGlobalIdx(int idx, const Lattice& sourcelattice, vector& sources, int spanstart, int spanend, const std::vector& ant_contexts, const map& sfw) { - // get the index of the function word in the lattice - if (DEBUG) cerr << " GetFWGlobalIdx(" << idx << "," << spanstart << "," << spanend << ")" << endl; - int curr = spanstart; int i_ant = 0; - for (int i=1; i and - if (sources[i]<0) { - const int* ants = reinterpret_cast(ant_contexts[i_ant++]); - int antstate = ants[Dwarf::STATE_SIZE-1]; - if (DEBUG) cerr << " found NT[" << target(antstate) << "," << source(antstate) << "]" << endl; - curr += target(antstate)-source(antstate); - } else { - curr++; - } - } - if (DEBUG) cerr << " curr = " << curr << endl; - //compute the fw index - int ret = 1; - for (int i=0; i=spanstart) return curr; - } - } -// assert(0); - return curr; -} - -int Alignment::GetLastFWIdx(int spanstart,int spanend, const Lattice& sourcelattice, const map& sfw) { - if (DEBUG) cerr << " GetLastFWIdx(" << spanstart << "," << spanend << ")" << endl; - int curr=0; - for (int i=0; i& tags, bool pos) { - if (!pos) { - map::const_iterator it = tags.find(original); - if (it!=tags.end()) { - return it->second; - } - } else { - string key,idx; - Dwarf::stripIndex(TD::Convert(original),&key,&idx); - map::const_iterator it = tags.find(TD::Convert(key)); - if (it!=tags.end()) { - ostringstream oss; - oss << TD::Convert(it->second) << "/" << idx; - return TD::Convert(oss.str()); - } - } - return original; -} - -int* Alignment::SOS() { - int* neighbor = new int[4]; - neighbor[0]=0; neighbor[1]=0; - neighbor[2]=0; neighbor[3]=0; - return neighbor; -} - -int* Alignment::EOS() { - int* neighbor = new int[4]; - neighbor[0]=getJ()-1; neighbor[1]=neighbor[0]; - neighbor[2]=getI()-1; neighbor[3]=neighbor[2]; - return neighbor; -} - -int* Alignment::neighborLeft(int startidx, int endidx, bool* getit) { - if (DEBUG) cerr << " neighborLeft("< -#include -#include -#include -#include -#include "wordid.h" -#include "lattice.h" -#include "trule.h" -#include "tdict.h" -#include -#include -#include - -using namespace std; -using namespace std::tr1; -using namespace boost::tuples; -using namespace boost; - -const static bool DEBUG = false; - -class CountTable { -public: - int* ultimate; - map model; - int mode; - int numColumn; - void print() const; - void setup(int _numcolumn, int _mode) { - mode = _mode; numColumn = _numcolumn; - } -}; - -class Alignment { -/* Alignment represents an alignment object in a 2D format to support function word-based models calculation - - A note about model's parameter estimation: - ========================================== - The model is estimated as a two-level Dirichlet process. - For orientation model, the first tier estimation is: - P(o|f,e) where *o* is the orientation value to estimate, *f* is the source function word aligned to *e* - its second tier is: P(o|f), while its third tier is P(o) - For dominance model, the first tier estimation is: - P(d|f1,f2,e1,e2) where *d* is a dominance value to estimate, *f1,f2* are the neighboring function words on the source - aligned to *e1,e2* on the target side - its second tier is: P(d|f1,f2) while its third tier is P(d) - - Taking orientation model as a case in point, a two level estimation proceeds as follow: - P(o|f,e) = c(o,f,e) + alpha { c(o,f) + beta [ c (o) / c(.) ] } - ------------------------------ - c(f) + beta - ------------------------------------------------- - c(f,e) + alpha - where c() is a count function, alpha and beta are the concentration parameter - of the first and second Dirichlet process respectively - To encourage or penalize the use of second and third tier statistics, bo1 and bo2 binary features are introduced -*/ -public: - const static int MAX_WORDS = 200; - const static int MINIMUM_INIT = 1000; - const static int MAXIMUM_INIT = -1000; - const static int MAX_ARITY = 2; - WordID kSOS; - WordID kEOS; - WordID kUNK; - double alpha_oris; // 1st concentration parameter for orientation model - double beta_oris; // 2nd concentration parameter for orientation model - double alpha_orit; // 1st concentration parameter for orientation model - double beta_orit; // 2nd concentration parameter for orientation model - double alpha_doms; // idem as above but for dominance model - double beta_doms; - double alpha_domt; // idem as above but for dominance model - double beta_domt; - - // ACCESS to alignment - void set(int j,int i); // j is the source index, while i is the target index - void reset(int j,int i); // idem as above - inline bool at(int j, int i) { return _matrix[j][i]; }; - inline int getJ() {return _J;}; // max source of the current alignment - inline int getI() {return _I;}; // max target of the current alignment - inline void setI(int I) { _I = I; }; - inline void setJ(int J) { _J = J; }; - inline void setF(vector f) { _f=f;}; - inline void setE(vector e) { _e=e;}; - inline WordID getF(int id) { if (id<0) return TD::Convert(""); if (id>=_f.size()) return TD::Convert(""); return _f[id];}; - inline WordID getE(int id) { if (id<0) return TD::Convert(""); if (id>=_e.size()) return TD::Convert(""); return _e[id];}; - void clearAls(int prevJ=200, int prevI=200); - int sourceOf(int i, int start = -1); - int targetOf(int j, int start = -1); - inline int minSSpan(int i) { return _sSpan[i][0];} - inline int maxSSpan(int i) { return _sSpan[i][1];} - inline int minTSpan(int j) { return _tSpan[j][0];} - inline int maxTSpan(int j) { return _tSpan[j][1];} - static inline int link(int s, int t) { return (s << 16) | t; } - static inline int source(int st) {return st >> 16; } - static inline int target(int st) {return st & 0xffff; } - inline void setAlphaOris(double val) { alpha_oris=val; } - inline void setAlphaOrit(double val) { alpha_orit=val; } - inline void setAlphaDoms(double val) { alpha_doms=val; } - inline void setAlphaDomt(double val) { alpha_domt=val; } - inline void setBetaOris(double val) { beta_oris=val; } - inline void setBetaOrit(double val) { beta_orit=val; } - inline void setBetaDoms(double val) { beta_doms=val; } - inline void setBetaDomt(double val) { beta_domt=val; } - inline void setFreqCutoff(int val) { cout << _freq_cutoff << " to " << val << endl; _freq_cutoff=val; } - string AsString(); - string AsStringSimple(); - int* SOS(); - int* EOS(); - - // Model related function - Alignment(); - // Given the current *rule* and its antecedents, construct an alignment space and mark the function word alignments - // according *sfw* and *tfw* - bool prepare(TRule& rule, const std::vector& ant_contexts, - const map& sfw, const map& tfw, const Lattice& sourcelattice, int spanstart, int spanend); - - // Compute orientation model score which parameters are stored in *table* and pass the values accordingly - // will call Orientation(Source|Target) and ScoreOrientation(Source|Target) - void computeOrientationSource(const CountTable& table, double *cost, double *bonus, double *bo1, - double *bo1_bonus, double *bo2, double *bo2_bonus); - void computeOrientationSourcePos(const CountTable& table, double *cost, double *bonus, - double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, int maxfwidx, int maxdepth1, int maxdepth2); - void computeOrientationSourceGen(const CountTable& table, double *cost, double *bonus, double *bo1, - double *bo1_bonus, double *bo2, double *bo2_bonus, const map& tags); - void computeOrientationSourceBackward(const CountTable& table, double *cost, double *bonus, double *bo1, - double *bo1_bonus, double *bo2, double *bo2_bonus); - void computeOrientationSourceBackwardPos(const CountTable& table, double *cost, double *bonus, double *bo1, - double *bo1_bonus, double *bo2, double *bo2_bonus, int maxfwidx, int maxdepth1, int maxdepth2); - void computeOrientationTarget(const CountTable& table, double *cost, double *bonus, double *bo1, - double *bo1_bonus, double *bo2, double *bo2_bonus); - void computeOrientationTargetBackward(const CountTable& table, double *cost, double *bonus, double *bo1, - double *bo1_bonus, double *bo2, double *bo2_bonus); - // Get the orientation value of a function word at a particular index *fw* - // assign the value to either *oril* or *orir* accoring to *Lcompute* and *Rcompute* - void OrientationSource(int fw, int*oril, int* orir, bool Lcompute=true, bool Rcompute=true); - void OrientationSource(int fw0, int fw1, int*oril, int* orir, bool Lcompute=true, bool Rcompute=true); - int OrientationSource(int* left, int* right); - void OrientationTarget(int fw, int*oril, int* orir, bool Lcompute=true, bool Rcompute=true); - void OrientationTarget(int fw0, int fw1, int*oril, int* orir, bool Lcompute=true, bool Rcompute=true); - - vector OrientationSourceLeft4Sampler(int fw0, int fw1); - vector OrientationSourceLeft4Sampler(int fw); - vector OrientationSourceRight4Sampler(int fw0, int fw1); - vector OrientationSourceRight4Sampler(int fw); - vector OrientationTargetLeft4Sampler(int fw0, int fw1); - vector OrientationTargetLeft4Sampler(int fw); - vector OrientationTargetRight4Sampler(int fw0, int fw1); - vector OrientationTargetRight4Sampler(int fw); - - // Given an orientation value *ori*, estimate the score accoding to *cond1*, *cond2* - // and assign the value accordingly according to *isBonus* and whether the first or the second tier estimation - // is used or not - void ScoreOrientationRight(const CountTable& table, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, - double *bo2, double *bo2_bonus, double alpha1, double beta1); - void ScoreOrientationLeft(const CountTable& table, int ori, WordID cond1, WordID cond, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, - double *bo2, double *bo2_bonus, double alpha1, double beta1); - double ScoreOrientationRight(const CountTable& table, int ori, WordID cond1, WordID cond2); - double ScoreOrientationLeft(const CountTable& table, int ori, WordID cond1, WordID cond); - void ScoreOrientationRightBackward(const CountTable& table, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, - double *bo2, double *bo2_bonus, double alpha1, double beta1); - void ScoreOrientationLeftBackward(const CountTable& table, int ori, WordID cond1, WordID cond, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, - double *bo2, double *bo2_bonus, double alpha1, double beta1); - double ScoreOrientationRightBackward(const CountTable& table, int ori, WordID cond1, WordID cond2); - double ScoreOrientationLeftBackward(const CountTable& table, int ori, WordID cond1, WordID cond); - void ScoreOrientation(const CountTable& table, int offset, int ori, WordID cond1, WordID cond2, - bool isBonus, double *cost, double *bonus, double *bo1, double *bo1_bonus, - double *bo2, double *bo2_bonus, double alpha1, double beta1); - double ScoreOrientation(const CountTable& table, int offset, int ori, WordID cond1, WordID cond2); - - // idem as above except these are for dominance model - void computeDominanceSource(const CountTable& table, WordID lfw, WordID rfw, double *cost, double *bonus, - double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus); - void computeDominanceSourcePos(const CountTable& table, WordID lfw, WordID rfw, double *cost, double *bonus, - double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus, int maxfwidx, int maxdepth1, int maxdepth2); - void computeDominanceTarget(const CountTable& table, WordID lfw, WordID rfw, double *cost, double *bonus, - double *bo1, double *bo1_bonus, double *bo2, double *bo2_bonus); - void computeBorderDominanceSource(const CountTable& table, double *cost, double *bonus, - double *state_mono, double *state_nonmono, - TRule &rule, const std::vector& ant_contexts, const map& sfw); - int DominanceSource(int fw1, int fw2); - int DominanceTarget(int fw1, int fw2); - vector DominanceSource4Sampler(int fw1, int fw2); - vector DominanceTarget4Sampler(int fw1, int fw2); - void ScoreDominance(const CountTable& table, int dom, WordID s1, WordID s2, WordID t1, WordID t2, - double *cost, double *bo1, double *bo2, bool isBonus, double alpha2, double beta2); - double ScoreDominance(const CountTable& table, int dom, WordID s1, WordID s2, WordID t1, WordID t2); - - // Remove all function word alignments except those at the borders - // May result in more than two function word alignments at each side, because this function - // will continue keeping function word alignments until the first aligned word at each side - void BorderingSFWsOnly(); - void BorderingTFWsOnly(); - void simplify(int *ret); // preparing the next state - void simplify_nofw(int *ret); // preparing the next state when no function word appears - // set the first part of the next state, which concerns with function word - // fas, las, fat, lat is the (f)irst or (l)ast function word alignments either on the (s)ource or (t)arget - // these parameters to anticipate cases where there are more than two function word alignments - void FillFWIdxsState(int *state, int fas, int las, int fat, int lat); - - // Helper function to obtain the aligned words on the other side - // WARNING!!! Only to be used if the als are in sync with either source or target sentences - WordID F2EProjectionFromExternal(int idx, const vector& als, const string& delimiter=" "); - WordID E2FProjectionFromExternal(int idx, const vector& als, const string& delimiter=" "); - // WARNING!!! Only to be used in dwarf_main.cc - // These two function words assume that the alignment contains phrase boundary - // but the source and target sentences do not - WordID F2EProjection(int idx, const string& delimiter=" "); - WordID E2FProjection(int idx, const string& delimiter=" "); - void SetCurrAlVector(); - int* blockSource(int fw1, int fw2); - int* blockTarget(int fw1, int fw2); - void ToArrayInt(vector* arr); - int* neighborLeft(int startidx, int endidx, bool* found); - int* neighborRight(int startidx, int endidx, bool* found); -private: - // Hash to avoid redundancy - unordered_map, int, boost::hash > > oris_hash; - unordered_map, int, boost::hash > > orit_hash; - unordered_map, int, boost::hash > > doms_hash; - unordered_map, int, boost::hash > > domt_hash; - unordered_map, vector, boost::hash > > simplify_hash; - unordered_map, vector, boost::hash > > prepare_hash; - - int _J; // effective source length; - int _I; // effective target length; - bool _matrix[MAX_WORDS][MAX_WORDS]; // true if aligned - short _sSpan[MAX_WORDS][2]; //the source span of a target index; 0->min, 1->max - short _tSpan[MAX_WORDS][2]; //the target span of a source index; 0->min, 2->max - int _freq_cutoff; - int SourceFWRuleIdxs[40]; //the indexes of function words in the rule; - // The following applies to all *FW*Idxs - // *FW*Idxs[0] = size - // *FW*Idxs[idx*3-2] = index in the alignment, where idx starts from 1 to size - // *FW*Idxs[idx*3-1] = source WordID - // *FW*Idxs[idx*3] = target WordID - int SourceFWRuleAbsIdxs[40]; - int TargetFWRuleIdxs[40]; //the indexes of function words in the rule; zeroth element is the count - int ** SourceFWAntsIdxs; //the indexes of function words in antecedents - int ** SourceFWAntsAbsIdxs; - int ** TargetFWAntsIdxs; //the indexes of function words in antecedents - int SourceRuleIdxs[40]; //the indexes of SOURCE tokens (zeroth element is the number of source tokens) - //>0 means terminal, -i means the i-th Xs - int TargetRuleIdxs[40]; //the indexes of TARGET tokens (zeroth element is the number of target tokens) - int ** SourceAntsIdxs; //the array of indexes of a particular antecedent's SOURCE tokens - int ** TargetAntsIdxs; //the array of indexes of a particular antecedent's TARGET tokens - int SourceFWIdxs[40]; - int SourceFWAbsIdxs[40]; - int TargetFWIdxs[40]; - // *sort* and *quickSort* are used to sort *FW*Idxs - void sort(int* num); - void quickSort(int arr[], int top, int bottom); - - // *block(Source|Target)* finds the minimum block that containts two indexes (fw1 and fw2) - inline int least(int i1, int i2) { return (i1i2)?i1:i2; } - void simplifyBackward(vector*blocks, int* block, const vector& danglings); - // used in simplify to check whether an atomic block according to source function words is also atomic according - // to target function words as well, otherwise break it - // the resulting blocks are added into *blocks* - int _Arity; - std::vector _f; // the source sentence of the **current** rule (may not consistent with the current alignment) - std::vector _e; // the target sentence of the **current** rule - int RuleAl[40]; - int **AntsAl; - int firstSourceAligned(int start); - int firstTargetAligned(int start); - int lastSourceAligned(int end); - int lastTargetAligned(int end); - int fas, las, fat, lat; // first aligned source, last aligned source, first aligned target, last aligned target - bool MemberOf(int* FWIdxs, int pos1, int pos2); // whether FWIdxs contains pos1 and pos2 consecutively - // Convert the alignment to vector form, will be used for hashing purposes - vector curr_al; - int GetFWGlobalIdx(int idx, const Lattice& sourcelattice, vector& sources, int spanstart, int spanend, const std::vector& ant_contexts, const map& sfw); - int GetFirstFWIdx(int spanstart,int spanend, const Lattice& sourcelattice, const map& sfw); - int GetLastFWIdx(int spanstart,int spanend, const Lattice& sourcelattice, const map& sfw); - WordID generalize(WordID original, const map& tags, bool pos=false); -}; - -#endif diff --git a/decoder/earley_composer.cc b/decoder/earley_composer.cc index efce70a6..32c387d3 100644 --- a/decoder/earley_composer.cc +++ b/decoder/earley_composer.cc @@ -4,8 +4,14 @@ #include #include #include -#include -#include +#ifdef HAVE_CXX11 +# include +# include +#else +# include +# include +namespace std { using std::tr1::unordered_map; using std::tr1::unordered_multiset; using std::tr1::unordered_set; } +#endif #include #include @@ -19,7 +25,6 @@ #include "hg_remove_eps.h" using namespace std; -using namespace std::tr1; // Define the following macro if you want to see lots of debugging output // when you run the chart parser diff --git a/decoder/factored_lexicon_helper.cc b/decoder/factored_lexicon_helper.cc index 7203b325..e7899215 100644 --- a/decoder/factored_lexicon_helper.cc +++ b/decoder/factored_lexicon_helper.cc @@ -2,6 +2,7 @@ #include "filelib.h" #include "stringlib.h" +#include "sentence_metadata.h" using namespace std; diff --git a/decoder/factored_lexicon_helper.h b/decoder/factored_lexicon_helper.h index 81c75275..7fedc517 100644 --- a/decoder/factored_lexicon_helper.h +++ b/decoder/factored_lexicon_helper.h @@ -6,7 +6,8 @@ #include #include #include "tdict.h" -#include "sentence_metadata.h" + +struct SentenceMetadata; // when computing features, it can be advantageous to: // 1) back off to less specific forms (e.g., less highly inflected forms, POS tags, etc) diff --git a/decoder/ff_dwarf.cc b/decoder/ff_dwarf.cc deleted file mode 100644 index fe7a472e..00000000 --- a/decoder/ff_dwarf.cc +++ /dev/null @@ -1,894 +0,0 @@ -#include -#include -#include -#include -#include -#include -#include "hg.h" -#include "ff_dwarf.h" -#include "dwarf.h" -#include "wordid.h" -#include "tdict.h" -#include "filelib.h" -#include "sentence_metadata.h" -#include "stringlib.h" - -using namespace std; - -Dwarf::Dwarf(const std::string& param) { -/* Param is a space separated string which contains any or all of the following: - oris|orit|doms|domt=filename - e.g. oris=/fs/clip-galep3eval/hendra/z2e/oris128.gz -*/ - sSOS=""; - sEOS=""; - kSOS=TD::Convert(sSOS); - kEOS=TD::Convert(sEOS); - kGOAL=TD::Convert("S")*-1; - _sent_id = (int *)malloc(sizeof(int)); - *_sent_id = -1; - if (DEBUG) cerr << "here = " << *_sent_id << endl; - _fwcount = (int *)malloc(sizeof(int)); - *_fwcount = -1; - cerr << "initializing dwarf" << endl; - flag_oris=false; flag_orit=false; flag_doms=false; flag_domt=false; flag_tfw_count=false; - flag_bdoms=false; flag_porislr=false, flag_porisrl=false, flag_goris=false; flag_pgorislr=false, flag_pgorisrl=false; - flag_pdomslr=false; flag_pdomsrl=false; flag_pgdomslr=false; flag_pgdomsrl=false; flag_gdoms=false; - flag_oris_backward=false; flag_orit_backward=false; - explicit_soseos=false; - SetStateSize(STATE_SIZE*sizeof(int)); - als = new Alignment(); - als->clearAls(Alignment::MAX_WORDS,Alignment::MAX_WORDS); - istringstream iss(param); string w; - while(iss >> w) { - int equal = w.find_first_of("="); - if (equal!=string::npos) { - string model = w.substr(0,equal); - vector params; - Tokenize(w.substr(equal+1),',',¶ms); - string fn = params[0]; - if (model == "minfreq") { - cerr << "model minfreq " << fn << endl; - als->setFreqCutoff(atoi(fn.c_str())); - } else if (model == "oris") { - flag_oris = readOrientation(&toris,fn,&sfw); - if (flag_oris) { - oris_ = FD::Convert("OrientationSource"); - //oris_bo1_ = FD::Convert("OrientationSource_BO1"); - //oris_bo2_ = FD::Convert("OrientationSource_BO2"); - } - if (params.size()>1) als->setAlphaOris(atof(params[1].c_str())); - if (params.size()>2) als->setBetaOris(atof(params[2].c_str())); - } else if (model == "porislr") { - flag_porislr = readOrientation(&tporislr,fn,&sfw,true); - poris_nlr = 0; - if (flag_porislr) { - porislr_ = FD::Convert("OrientationSourcePositionfulLeftRight"); - } - if (params.size()>1) poris_nlr = atoi(params[1].c_str()); - if (DEBUG) cerr << " maximum poris depth=" << poris_nlr << endl; - } else if (model == "porisrl") { - flag_porisrl = readOrientation(&tporisrl,fn,&sfw,true); - poris_nrl = 0; - if (flag_porisrl) { - porisrl_ = FD::Convert("OrientationSourcePositionfulRightLeft"); - } - if (params.size()>1) poris_nrl = atoi(params[1].c_str()); - if (DEBUG) cerr << " maximum poris depth=" << poris_nrl << endl; - } else if (model=="goris") { - flag_goris = readOrientation(&tgoris,fn,&sfw); - if (flag_goris) { - goris_ = FD::Convert("OrientationSourceGeneralized"); - } - if (params.size()>1) { - readTags(params[1],&tags); - generalizeOrientation(&tgoris,tags); - } - } else if (model=="pgorislr") { - flag_pgorislr = readOrientation(&tpgorislr,fn,&sfw,true); - pgoris_nlr = 0; - if (flag_pgorislr) { - pgorislr_ = FD::Convert("OrientationSourceGeneralizedPositionfulLeftRight"); - } - if (DEBUG) { - cerr << "BEFORE GENERALIZATION" << endl; - tpgorislr.print(); - } - if (params.size()>1) pgoris_nlr = atoi(params[1].c_str()); - if (params.size()>2) { - readTags(params[2],&tags); - generalizeOrientation(&tpgorislr,tags,true); - } - if (DEBUG) { - cerr << "AFTER GENERALIZATION" << endl; - tpgorislr.print(); - } - } else if (model=="pgorisrl") { - flag_pgorisrl = readOrientation(&tpgorisrl,fn,&sfw,true); - pgoris_nrl = 0; - if (flag_pgorisrl) { - pgorisrl_ = FD::Convert("OrientationSourceGeneralizedPositionfulLeftRight"); - } - if (params.size()>1) pgoris_nrl = atoi(params[1].c_str()); - if (params.size()>2) { - readTags(params[2],&tags); - generalizeOrientation(&tpgorisrl,tags,true); - } - } else if (model == "oris_backward") { - flag_oris_backward = true; - if (!flag_oris) readOrientation(&toris,fn,&sfw); - oris_backward_ = FD::Convert("OrientationSourceBackward"); - if (params.size()>1) als->setAlphaOris(atof(params[1].c_str())); - if (params.size()>2) als->setBetaOris(atof(params[2].c_str())); - } else if (model == "orit") { - flag_orit = readOrientation(&torit,fn,&tfw); - if (flag_orit) { - orit_ = FD::Convert("OrientationTarget"); - //orit_bo1_ = FD::Convert("OrientationTarget_BO1"); - //orit_bo2_ = FD::Convert("OrientationTarget_BO2"); - } - if (params.size()>1) als->setAlphaOrit(atof(params[1].c_str())); - if (params.size()>2) als->setBetaOrit(atof(params[2].c_str())); - } else if (model == "orit_backward") { - flag_orit_backward = true; - if (!flag_orit) readOrientation(&torit,fn,&tfw); - orit_backward_ = FD::Convert("OrientationTargetBackward"); - if (params.size()>1) als->setAlphaOrit(atof(params[1].c_str())); - if (params.size()>2) als->setBetaOrit(atof(params[2].c_str())); - } else if (model == "doms") { - flag_doms = readDominance(&tdoms,fn,&sfw); - if (flag_doms) { - doms_ = FD::Convert("DominanceSource"); - //doms_bo1_ = FD::Convert("DominanceSource_BO1"); - //doms_bo2_ = FD::Convert("DominanceSource_BO2"); - } - if (params.size()>1) als->setAlphaDoms(atof(params[1].c_str())); - if (params.size()>2) als->setBetaDoms(atof(params[2].c_str())); - } else if (model == "pdomsrl") { - flag_pdomsrl = readDominance(&tpdomsrl,fn,&sfw,true); - if (flag_pdomsrl) { - pdomsrl_ = FD::Convert("DominanceSourcePositionfulRightLeft"); - } - if (params.size()>1) pdoms_nrl = atoi(params[1].c_str()); - } else if (model == "pdomslr") { - flag_pdomslr = readDominance(&tpdomslr,fn,&sfw,true); - tpdomslr.print(); - if (flag_pdomslr) { - pdomslr_ = FD::Convert("DominanceSourcePositionfulLeftRight"); - } - if (params.size()>1) pdoms_nlr = atoi(params[1].c_str()); - } else if (model == "pgdomsrl") { - flag_pgdomsrl = readDominance(&tpgdomsrl,fn,&sfw,true); - if (flag_pgdomsrl) { - pgdomsrl_ = FD::Convert("DominanceSourceGeneralizedPositionfulRightLeft"); - } - if (params.size()>1) pgdoms_nrl = atoi(params[1].c_str()); - if (params.size()>2) { - readTags(params[2],&tags); - generalizeDominance(&tpgdomsrl,tags,true); - } - } else if (model == "pgdomslr") { - flag_pgdomslr = readDominance(&tpgdomslr,fn,&sfw,true); - if (flag_pgdomslr) { - pgdomslr_ = FD::Convert("DominanceSourceGeneralizedPositionfulLeftRight"); - } - if (params.size()>1) pgdoms_nlr = atoi(params[1].c_str()); - if (params.size()>2) { - readTags(params[2],&tags); - if (DEBUG) { - for (map::const_iterator it=tags.begin(); it!=tags.end(); it++) { - cerr << "tags = " << TD::Convert(it->first) << ", " << TD::Convert(it->second) << endl; - } - } - generalizeDominance(&tpgdomslr,tags,true); - } - if (DEBUG) tpgdomslr.print(); - } else if (model == "bdoms") { - flag_bdoms = readDominance(&tbdoms,fn,&sfw); - if (flag_bdoms) { - bdoms_ = FD::Convert("BorderDominanceSource"); - } - } else if (model == "domt") { - flag_domt = readDominance(&tdomt,fn,&tfw); - if (flag_domt) { - domt_ = FD::Convert("DominanceTarget"); - //domt_bo1_ = FD::Convert("DominanceTarget_BO1"); - //domt_bo2_ = FD::Convert("DominanceTarget_BO2"); - } - if (params.size()>1) als->setAlphaDomt(atof(params[1].c_str())); - if (params.size()>2) als->setBetaDomt(atof(params[2].c_str())); - } else if (model== "tfw_count") { - flag_tfw_count = readList(fn,&tfw); - tfw_count_ = FD::Convert("TargetFunctionWordsCount"); - } else { - cerr << "DWARF doesn't understand this model: " << model << endl; - } - } else { - if (w=="tfw_count") { - flag_tfw_count = true; - tfw_count_ = FD::Convert("TargetFunctionWordsCount"); - } else if (w=="oris_backward") { - flag_oris_backward = true; - oris_backward_ = FD::Convert("OrientationSourceBackward"); - } else if (w=="orit_backward") { - flag_orit_backward = true; - orit_backward_ = FD::Convert("OrientationTargetBackward"); - } else if (w=="explicit_soseos") { - explicit_soseos=true; - } else { - cerr << "DWARF doesn't need this param: " << param << endl; - } - } - } - for (map::const_iterator it=sfw.begin(); it!=sfw.end() && DEBUG; it++) { - cerr << " FW:" << TD::Convert(it->first) << endl; - } -} - -void Dwarf::TraversalFeaturesImpl(const SentenceMetadata& smeta, - const Hypergraph::Edge& edge, - const std::vector& ant_contexts, - SparseVector* features, - SparseVector* estimated_features, - void* context) const { - if (DEBUG) cerr << "TraversalFeaturesImpl" << endl; - double cost, bonus, bo1, bo2, bo1_bonus, bo2_bonus; - double bdoms_state_mono= 0; double bdoms_state_nonmono = 0; - TRule r = *edge.rule_; - if (DEBUG) cerr << " sent_id=" << *_sent_id << ", " << smeta.GetSentenceID() << endl; - if (DEBUG) cerr << "rule = " << r.AsString() << endl; - if (DEBUG) cerr << "rule[i,j] = " << edge.i_ << "," << edge.j_ << endl; - if (*_sent_id != smeta.GetSentenceID()) { //new sentence - *_sent_id = smeta.GetSentenceID(); - const Lattice l = smeta.GetSourceLattice(); - *_fwcount=0; - for (int i=0; icomputeDominanceSource(tdoms,_lfw,_rfw,&cost,&bonus, - &bo1,&bo1_bonus,&bo2,&bo2_bonus); - if (DEBUG) cerr << " COST=" << cost << ", BONUS=" << bonus << endl; - if (isFinal&&!explicit_soseos) { - cost += bonus; - if (DEBUG) cerr << " final and !explicit_soseos, thus cost = " << cost << endl; - bonus = 0; - } - features->set_value(doms_,cost); - estimated_features->set_value(doms_,bonus); - } - if (flag_pdomslr) { - if (DEBUG) cerr << " flag_pdomslr true, nofw=" << nofw << endl; - if (DEBUG) cerr << " lfw=" << _lfw << ", rfw=" << _rfw << endl; - if (DEBUG) cerr << " kSOS=" << kSOS << ", kEOS=" << kEOS << endl; - cost=0; bonus=0; bo1=0; bo2=0; bo1_bonus=0; bo2_bonus=0; - if (!nofw) als->computeDominanceSourcePos(tpdomslr,_lfw,_rfw,&cost,&bonus, - &bo1,&bo1_bonus,&bo2,&bo2_bonus,*_fwcount,pdoms_nlr,0); - if (isFinal&&!explicit_soseos) { - cost += bonus; - bonus = 0; - } - features->set_value(pdomslr_,cost); - estimated_features->set_value(pdomslr_,bonus); - } - if (flag_pdomsrl) { - cost=0; bonus=0; bo1=0; bo2=0; bo1_bonus=0; bo2_bonus=0; - if (!nofw) als->computeDominanceSourcePos(tpdomsrl,_lfw,_rfw,&cost,&bonus, - &bo1,&bo1_bonus,&bo2,&bo2_bonus,*_fwcount,0,pdoms_nrl); - if (isFinal&&!explicit_soseos) { - cost += bonus; - bonus = 0; - } - features->set_value(pdomsrl_,cost); - estimated_features->set_value(pdomsrl_,bonus); - } - if (flag_pgdomslr) { - cost=0; bonus=0; bo1=0; bo2=0; bo1_bonus=0; bo2_bonus=0; - if (!nofw) als->computeDominanceSourcePos(tpgdomslr,_lfw,_rfw,&cost,&bonus, - &bo1,&bo1_bonus,&bo2,&bo2_bonus,*_fwcount,pgdoms_nlr,0); - if (isFinal&&!explicit_soseos) { - cost += bonus; - bonus = 0; - } - features->set_value(pgdomslr_,cost); - estimated_features->set_value(pgdomslr_,bonus); - } - if (flag_pgdomsrl) { cost=0; bonus=0; bo1=0; bo2=0; bo1_bonus=0; bo2_bonus=0; - if (!nofw) als->computeDominanceSourcePos(tpgdomsrl,_lfw,_rfw,&cost,&bonus, - &bo1,&bo1_bonus,&bo2,&bo2_bonus,*_fwcount,0,pgdoms_nrl); - if (isFinal&&!explicit_soseos) { - cost += bonus; - bonus = 0; - } - features->set_value(pgdomsrl_,cost); - estimated_features->set_value(pgdomsrl_,bonus); - } - - - if (flag_bdoms) { - cost=0; bonus=0; bdoms_state_mono=0; bdoms_state_nonmono=0; - if (!nofw) - als->computeBorderDominanceSource(tbdoms,&cost,&bonus, - &bdoms_state_mono, &bdoms_state_nonmono,*edge.rule_, ant_contexts, sfw); - features->set_value(bdoms_,cost); - estimated_features->set_value(bdoms_,bonus); - } - if (flag_orit) { - cost=0; bonus=0; bo1=0; bo2=0; bo1_bonus=0; bo2_bonus=0; - if (!nofw) als->computeOrientationTarget(torit,&cost,&bonus,&bo1,&bo1_bonus,&bo2,&bo2_bonus); - if (DEBUG) cerr << "cost=" << cost << ", bonus=" << bonus << ", bo1=" << bo1 << ", bo1_bonus=" << bo1_bonus << ", bo2=" << bo2 << ", bo2_bonus=" << bo2_bonus << endl; - features->set_value(orit_,cost); - //features->set_value(orit_bo1_,bo1); - //features->set_value(orit_bo2_,bo2); - estimated_features->set_value(orit_,bonus); - //estimated_features->set_value(orit_bo1_,bo1_bonus); - //estimated_features->set_value(orit_bo2_,bo2_bonus); - } - if (flag_orit_backward) { - cost=0; bonus=0; - if (!nofw) als->computeOrientationTargetBackward(torit,&cost,&bonus,&bo1,&bo1_bonus,&bo2,&bo2_bonus); - features->set_value(orit_backward_,cost); - estimated_features->set_value(orit_backward_,bonus); - } - if (flag_domt) { - cost=0; bonus=0; bo1=0; bo2=0; bo1_bonus=0; bo2_bonus=0; - WordID _lfw=-1; int start = edge.i_; - WordID _rfw=-1; int end = edge.j_; - if (smeta.HasReference()) { - const Lattice l = smeta.GetReference(); - for (int idx=start-1; idx>=0; idx--) { - if (l.size()>0) - if (tfw.find(l[idx][0].label) !=tfw.end()) { - _lfw=l[idx][0].label; break; - } - } - for (int idx=end; idx0) - if (tfw.find(l[idx][0].label)!=tfw.end()) { - _rfw=l[idx][0].label; break; - } - } - } - //neighboringFWs(smeta.GetReference(),edge.i_,edge.j_,tfw,&_lfw,&_rfw); - if (!nofw) als->computeDominanceTarget(tdomt,_lfw,_rfw,&cost,&bonus, - &bo1,&bo1_bonus,&bo2,&bo2_bonus); - features->set_value(domt_,cost); - //features->set_value(domt_bo1_,bo1); - //features->set_value(domt_bo2_,bo2); - estimated_features->set_value(domt_,bonus); - //estimated_features->set_value(domt_bo1_,bo1_bonus); - //estimated_features->set_value(domt_bo2_,bo2_bonus); - } - int* vcontext = reinterpret_cast(context); - if (!nofw) { - als->BorderingSFWsOnly(); - als->BorderingTFWsOnly(); - als->simplify(vcontext); - } else { - als->simplify_nofw(vcontext); - } - vcontext[50] = DoubleToInteger(bdoms_state_mono); - vcontext[51] = DoubleToInteger(bdoms_state_nonmono); - vcontext[STATE_SIZE-1] = Alignment::link(edge.i_,edge.j_); - if (DEBUG) { - cerr << "state@traverse = "; - for (int idx=0; idxsetup(24,pos); - table->ultimate = new int[24]; - for (int i=0; i<24; i++) table->ultimate[i]=0; - ostringstream oss; - while (in) { - string line; - getline(in,line); - if (line=="") break; - istringstream tokenizer(line); - string sourceidx, source, target, word; - tokenizer >> source >> target; - if (pos) { - sourceidx = source; - source = sourceidx.substr(0,sourceidx.find_last_of("/")); - } - if (fw->find(TD::Convert(source))==fw->end()) fw->insert(pair(TD::Convert(source),1)); - - - int* element = new int[24]; - element[5] = 0; - for (int i=0; i<5; i++) { - element[i] = 0; - if (tokenizer >> word) element[i] = atoi(word.c_str()); - element[5] += element[i]; - } - element[11] = 0; - for (int i=6; i<11; i++) { - element[i] = 0; - if (tokenizer >> word) element[i] = atoi(word.c_str()); - element[11] += element[i]; - } - element[17] = 0; - for (int i=12; i<17; i++) { - element[i] = 0; - if (tokenizer >> word) element[i] = atoi(word.c_str()); - element[17] += element[i]; - } - element[23] = 0; - for (int i=18; i<23; i++) { - element[i] = 0; - if (tokenizer >> word) element[i] = atoi(word.c_str()); - element[23] += element[i]; - } - for (int i=0; i<24; i++) table->ultimate[i] += element[i]; - oss << source << " " << target; - WordID key_id = TD::Convert(oss.str()); - oss.str(""); - if (table->model.find(key_id)!=table->model.end()) { - for (int i=0; i<24; i++) table->model[key_id][i]+=element[i]; - } else { - int* el2 = new int[24]; - for (int i=0; i<24; i++) el2[i] = element[i]; - table->model.insert(pair(key_id,el2)); - } - - oss << source; - key_id = TD::Convert(oss.str()); - oss.str(""); - if (table->model.find(key_id)!=table->model.end()) { - for (int i=0; i<24; i++) table->model[key_id][i]+=element[i]; - } else { - int* el2 = new int[24]; - for (int i=0; i<24; i++) el2[i] = element[i]; - table->model.insert(pair(key_id,el2)); - } - - if (pos) { - oss << sourceidx << " " << target; - key_id = TD::Convert(oss.str()); - oss.str(""); - if (table->model.find(key_id)!=table->model.end()) { - for (int i=0; i<24; i++) table->model[key_id][i]+=element[i]; - } else { - int* el2 = new int[24]; - for (int i=0; i<24; i++) el2[i] = element[i]; - table->model.insert(pair(key_id,el2)); - } - } - delete[] element; - } - return true; -} - -bool Dwarf::readList(const std::string& filename, std::map* fw) { - ReadFile rf(filename); - istream& in = *rf.stream(); - while (in) { - string word; - getline(in,word); - if (fw->find(TD::Convert(word))==fw->end()) fw->insert(pair(TD::Convert(word),1)); - } - return true; -} - -bool Dwarf::readDominance(CountTable* table, const std::string& filename, std::map* fw, bool pos) { - // the input format is - // source1 source2 target1 target2 0 1 2 3 - // 0 -> dontcase 1->leftfirst 2->rightfirst 3->neither - if (DEBUG) cerr << "readDominance(" << filename << ",pos="<< pos << ")" << endl; - ReadFile rf(filename); - istream& in = *rf.stream(); - table->ultimate = new int[5]; - table->setup(5,pos); - for (int i=0; i<5; i++) table->ultimate[i]=0; - while (in) { - string line, word; - getline(in,line); - if (line=="") break; - string source1idx, source2idx, target1, target2, source1, source2; - ostringstream oss; - WordID key_id; - istringstream tokenizer(line); - tokenizer >> source1 >> source2 >> target1 >> target2; - if (pos) { - source1idx = source1; - source2idx = source2; - source1 = source1idx.substr(0,source1idx.find_last_of("/")); - source2 = source2idx.substr(0,source2idx.find_last_of("/")); - } - if (fw->find(TD::Convert(source1))==fw->end()) fw->insert(pair(TD::Convert(source1),1)); - if (fw->find(TD::Convert(source2))==fw->end()) fw->insert(pair(TD::Convert(source2),1)); - - int* element = new int[5]; - element[4]=0; - for (int i=0; i<4; i++) { - element[i] = 0; - if (tokenizer >> word) element[i] = atoi(word.c_str()); - element[4]+=element[i]; - } - for (int i=0; i<5; i++) table->ultimate[i] += element[i]; - - oss << source1 << " " << source2 << " " << target1 << " " << target2; - key_id = TD::Convert(oss.str()); - oss.str(""); - if (table->model.find(key_id)!=table->model.end()) { - for (int i=0; i<5; i++) table->model[key_id][i]+=element[i]; - } else { - int* el2 = new int[5]; - for (int i=0; i<5; i++) el2[i]=element[i]; - table->model.insert(pair(key_id,el2)); - } - - oss << source1 << " " << source2; - key_id = TD::Convert(oss.str()); - oss.str(""); - if (table->model.find(key_id)!=table->model.end()) { - for (int i=0; i<5; i++) table->model[key_id][i]+=element[i]; - } else { - int* el2 = new int[5]; - for (int i=0; i<5; i++) el2[i]=element[i]; - table->model.insert(pair(key_id,el2)); - } - - if (pos) { - oss << source1idx << " " << source2idx << " " << target1 << " " << target2; - key_id = TD::Convert(oss.str()); - oss.str(""); - if (table->model.find(key_id)!=table->model.end()) { - for (int i=0; i<5; i++) table->model[key_id][i]+=element[i]; - } else { - int* el2 = new int[5]; - for (int i=0; i<5; i++) el2[i]=element[i]; - table->model.insert(pair(key_id,el2)); - } - } - delete element; - } - - return true; -} - -bool Dwarf::readTags(const std::string& filename, std::map* tags) { - ReadFile rf(filename); - istream& in = *rf.stream(); - while(in) { - string line, word, tag; - getline(in,line); - if (line=="") break; - istringstream tokenizer(line); - tokenizer >> tag >> word; - tags->insert(pair(TD::Convert(word),TD::Convert(tag))); - } - return true; -} - -bool Dwarf::generalizeOrientation(CountTable* table, const std::map& tags, bool pos) { - map generalized; - for (map::iterator it=table->model.begin(); it!=table->model.end(); it++) { - string source, target; - istringstream tokenizer(TD::Convert(it->first)); - tokenizer >> source >> target; - string idx = ""; - if (pos) { - int found = source.find_last_of("/"); - if (found!=string::npos && found>0) { - idx = source.substr(found+1); - source = source.substr(0,found); - } - } - map::const_iterator tags_iter = tags.find(TD::Convert(source)); - if (tags_iter!=tags.end()) { - ostringstream genkey; - genkey << TD::Convert(tags_iter->second); - if (idx!="") genkey << "/" << idx; - if (target!="") genkey << " " << target; - int* model; - if (generalized.find(genkey.str())!=generalized.end()) { - model = generalized[genkey.str()]; - for (int i=0; i<24; i++) model[i] += it->second[i]; - } else { - int* el = new int[24]; - for (int i=0; i<24; i++) el[i] = it->second[i]; - generalized.insert(pair(genkey.str(),el)); - } - } - } - for (map::iterator it=table->model.begin(); it!=table->model.end(); it++) { - string source, target; - istringstream tokenizer(TD::Convert(it->first)); - tokenizer >> source >> target; - string idx = ""; - if (pos) { - int found = source.find_last_of("/"); - if (found!=string::npos && found>0) { - idx = source.substr(found+1); - source = source.substr(0,found); - } - } - map::const_iterator tags_iter = tags.find(TD::Convert(source)); - if (tags_iter!=tags.end()) { - ostringstream genkey; - genkey << TD::Convert(tags_iter->second); - if (idx!="") genkey << "/" << idx; - if (target!="") genkey << " " << target; - if (generalized.find(genkey.str())!=generalized.end()) { - delete it->second; - it->second = generalized[genkey.str()]; - } - } - } - return false; // no idea if this is right -} - - - -bool Dwarf::generalizeDominance(CountTable* table, const std::map& tags, bool pos) { - map generalized; - ostringstream oss; - for (map::iterator it=table->model.begin(); it!=table->model.end(); it++) { - string source1, source2, target1, target2; - string idx1 = ""; string idx2 = ""; - istringstream tokenizer(TD::Convert(it->first)); - tokenizer >> source1 >> source2 >> target1 >> target2; - if (DEBUG) cerr << "source1=|" << source1 << "|, source2=|" << source2 << "|, target1=|" << target1 << "|, target2=|" << target2 << "|" << endl; - if (pos) { - int found1 = source1.find_last_of("/"); - int found2 = source2.find_last_of("/"); - if (found1!=string::npos && found2!=string::npos && found1>0 && found2>0) { - idx1 = source1.substr(found1+1); - source1 = source1.substr(0,found1); - idx2 = source2.substr(found2+1); - source2 = source2.substr(0,found2); - } - } - if (DEBUG) - cerr << "[U]source1='" << source1 << "', idx1='"<< idx1 << "', source2='" << source2 << "', idx2='"<< idx2 << "', target1='" << target1 << "', target2='" << target2 << "'" << endl; - map::const_iterator tags_iter1 = tags.find(TD::Convert(source1)); - map::const_iterator tags_iter2 = tags.find(TD::Convert(source2)); - if (tags_iter1!=tags.end()) - source1 = TD::Convert(tags_iter1->second); - oss << source1; - if (idx1!="") oss << "/" << idx1; - if (tags_iter2!=tags.end()) - source2 = TD::Convert(tags_iter2->second); - oss << " " << source2; - if (idx2!="") oss << "/" << idx2; - if (target1!="" && target2!="") oss << " " << target1 << " " << target2; - - if (DEBUG) cerr << "generalized key = '" << oss.str() << "'" << endl; - if (generalized.find(oss.str())!=generalized.end()) { - int* model = generalized[oss.str()]; - for (int i=0; i<5; i++) model[i] += it->second[i]; - } else { - int* model = new int[5]; - for (int i=0; i<5; i++) model[i] = it->second[i]; - generalized.insert(pair(oss.str(),model)); - } - oss.str(""); - } - - if (DEBUG) { - for (map::const_iterator it=generalized.begin(); it!=generalized.end(); it++) { - cerr << "GENERALIZED = " << it->first << ", "; - for (int i=0; i<5; i++) cerr << it->second[i] << " "; - cerr << endl; - } - } - - for (map::iterator it=table->model.begin(); it!=table->model.end(); it++) { - string source1, source2, target1, target2; - string idx1 = ""; string idx2 = ""; - istringstream tokenizer(TD::Convert(it->first)); - tokenizer >> source1 >> source2 >> target1 >> target2; - if (pos) { - int found1 = source1.find_last_of("/"); - int found2 = source2.find_last_of("/"); - if (found1!=string::npos && found2!=string::npos && found1>0 && found2>0) { - idx1 = source1.substr(found1+1); - source1 = source1.substr(0,found1); - idx2 = source2.substr(found2+1); - source2 = source2.substr(0,found2); - } - } - map::const_iterator tags_iter1 = tags.find(TD::Convert(source1)); - map::const_iterator tags_iter2 = tags.find(TD::Convert(source2)); - if (tags_iter1!=tags.end()) - source1 = TD::Convert(tags_iter1->second); - oss << source1; - if (idx1!="") oss << "/" << idx1; - if (tags_iter2!=tags.end()) - source2 = TD::Convert(tags_iter2->second); - oss << " " << source2; - if (idx2!="") oss << "/" << idx2; - if (target1!="" && target2!="") oss << " " << target1 << " " << target2; - - if (generalized.find(oss.str())!=generalized.end()) { - if (DEBUG) cerr << " generalizing "<< TD::Convert(it->first) << " into " << oss.str() << endl; - if (DEBUG) { - cerr << " model from "; - for (int i=0; i<5; i++) cerr << it->second[i] << " "; - cerr << endl; - } - delete it->second; - it->second = generalized[oss.str()]; - if (DEBUG) { - cerr << " into "; - for (int i=0; i<5; i++) cerr << it->second[i] << " "; - cerr << endl; - } - } - oss.str(""); - } - -} diff --git a/decoder/ff_dwarf.h b/decoder/ff_dwarf.h deleted file mode 100644 index 3d6a7da6..00000000 --- a/decoder/ff_dwarf.h +++ /dev/null @@ -1,100 +0,0 @@ -#include -#include -#include -#include "ff.h" -#include "dwarf.h" -#include "lattice.h" - -using namespace std; - -class Dwarf : public FeatureFunction { - public: - Dwarf(const std::string& param); - /* State-related param - STATE_SIZE: the number of ints - MAXIMUM_ALIGNMENTS: the maximum number of alignments in the states, - each alignment point is encoded in one int - (the first two bytes for source, and the remaining one for target) - */ - static const int STATE_SIZE=53; - static const int IMPOSSIBLY_LARGE_POS = 9999999; - static const int MAXIMUM_ALIGNMENTS=37; - /* Read from file the Orientation(Source|Target model parameter. */ - static bool readOrientation(CountTable* table, const std::string& filename, std::map *fw, bool pos=false); - /* Read from file the Dominance(Source|Target) model parameter. */ - static bool readDominance(CountTable* table, const std::string& filename, std::map *fw, bool pos=false); - static bool readList(const std::string& filename, std::map* fw); - static double IntegerToDouble(int val); - static int DoubleToInteger(double val); - bool readTags(const std::string& filename, std::map* tags); - bool generalizeOrientation(CountTable* table, const std::map& tags, bool pos=false); - bool generalizeDominance(CountTable* table, const std::map& tags, bool pos=false); - static void stripIndex(const string& source, string* pkey, string* pidx) { - if (DEBUG) cerr << " stripIndex(" << source << ")" << endl; - int found = source.find_last_of("/"); - string idx = source.substr(found+1); - string key = source.substr(0,found); - if (DEBUG) cerr << " found=" << found << "," << key << "," << idx << endl; - pkey = &key; - pidx = &idx; - } - - - protected: - /* The high-level workflow is as follow: - 1. call *als->prepare*, which constructs the full alignment of the edge while taking into account the antecedents - also in this call, function words are identified. Most of the work in this call is to make sure the indexes - of the alignments (including the function words) are consistent with the newly created alignment - 2. call *als->computeOrientationSource*, *als->computeOrientationTarget*, - *als->computeDominanceSource*, or *als->computeDominanceTarget* - and pass the resulting score to either *features* or to *estimated_features* - 3. call *als->BorderingSFWsOnly()* and *als->BorderingTFWsOnly()*, which removes records of all function word - alignments except those at the borders. Note that fw alignments kept may be more than two on each side - for examples if there are a number of unaligned fw alignments before the leftmost alignment or the rightmost one - 4. call *als->simplify()*, which assigns the state of this edge (*context*). It simplifies the alignment space to - its most compact representation, enough to compute the unscored models. This is done by observing the surviving - function word alignments set by 3. - */ - void TraversalFeaturesImpl(const SentenceMetadata& smeta, - const HG::Edge& edge, - const std::vector& ant_contexts, - SparseVector* features, - SparseVector* estimated_features, - void* context) const; - private: - Alignment* als; - /* Feature IDs set by calling FD::Convert(model's string) */ - int oris_, oris_bo1_, oris_bo2_, orit_, orit_bo1_, orit_bo2_; - int oris_backward_, orit_backward_, porislr_, porisrl_, goris_, pgorislr_, pgorisrl_; - int pdomslr_, pdomsrl_, pgdomslr_, pgdomsrl_; - int doms_, doms_bo1_, doms_bo2_, domt_, domt_bo1_, domt_bo2_; - int tfw_count_; - int bdoms_; - int poris_count; - int pgoris_count; - int poris_nlr, poris_nrl; // maximum depth (1->from the beginning of the sentence, 2-> from the end of the sentence) - int pgoris_nlr, pgoris_nrl; - int pdoms_nlr, pdoms_nrl; - int pgdoms_nlr, pgdoms_nrl; - int* _sent_id; - int* _fwcount; - WordID kSOS; - WordID kEOS; - string sSOS; - string sEOS; - WordID kGOAL; - /* model's flag, if set true will invoke the model scoring */ - bool flag_oris, flag_orit, flag_doms, flag_domt, flag_tfw_count, flag_oris_backward, flag_orit_backward, flag_bdoms; - bool flag_porislr, flag_porisrl, flag_goris, flag_pgorislr, flag_pgorisrl; - bool explicit_soseos; - bool flag_pdomslr, flag_pdomsrl, flag_pgdomslr, flag_pgdomsrl, flag_gdoms; - /* a collection of Source function words (sfw) and Target function words (tfw) */ - std::map sfw; - std::map tfw; - std::map tags; - /* a collection of model's parameter */ - CountTable toris, torit, tdoms, tbdoms, tdomt, tporislr, tporisrl, tgoris, tpgorislr, tpgorisrl; - CountTable tpdomslr, tpdomsrl, tpgdomslr, tpgdomsrl; - void neighboringFWs(const Lattice& l, const int& i, const int& j, const map& fw_hash, int* lfw, int* rfw); -}; - diff --git a/decoder/ff_lm.cc b/decoder/ff_lm.cc index 6ec7b4f3..bc51076f 100644 --- a/decoder/ff_lm.cc +++ b/decoder/ff_lm.cc @@ -61,11 +61,6 @@ char const* usage_verbose="-n determines the name of the feature (and its weight #include "hg.h" #include "stringlib.h" -#ifdef HAVE_RANDLM -// http://randlm.sourceforge.net/ -#include "RandLM.h" -#endif - using namespace std; string LanguageModel::usage(bool param,bool verbose) { @@ -542,99 +537,3 @@ void LanguageModel::FinalTraversalFeatures(const void* ant_state, features->set_value(fid_, imp().FinalTraversalCost(ant_state)); } -#ifdef HAVE_RANDLM -struct RandLMImpl : public LanguageModelImpl { - RandLMImpl(int order, randlm::RandLM* rlm) : - LanguageModelImpl(order), - rlm_(rlm), - oov_(rlm->getWordID(rlm->getOOV())), - rb_(1000, oov_) { - map map_cdec2randlm; - int max_wordid = 0; - for(map::const_iterator it = rlm->vocabStart(); - it != rlm->vocabEnd(); ++it) { - const int cur = TD::Convert(it->first); - map_cdec2randlm[TD::Convert(it->first)] = it->second; - if (cur > max_wordid) max_wordid = cur; - } - cdec2randlm_.resize(max_wordid + 1, oov_); - for (map::iterator it = map_cdec2randlm.begin(); - it != map_cdec2randlm.end(); ++it) - cdec2randlm_[it->first] = it->second; - map_cdec2randlm.clear(); - } - - inline randlm::WordID Convert2RandLM(int w) { - return (w < cdec2randlm_.size() ? cdec2randlm_[w] : oov_); - } - - virtual double WordProb(int word, int* context) { - int i = order_; - int c = 1; - rb_[i] = Convert2RandLM(word); - while (i > 1 && *context > 0) { - --i; - rb_[i] = Convert2RandLM(*context); - ++context; - ++c; - } - const void* finalState = 0; - int found; - //cerr << "I = " << i << endl; - return rlm_->getProb(&rb_[i], c, &found, &finalState); - } - private: - boost::shared_ptr rlm_; - randlm::WordID oov_; - vector cdec2randlm_; - vector rb_; -}; - -LanguageModelRandLM::LanguageModelRandLM(const string& param) : - fid_(FD::Convert("RandLM")) { - vector argv; - int argc = SplitOnWhitespace(param, &argv); - int order = 3; - // TODO add support for -n FeatureName - string filename; - if (argc < 1) { cerr << "RandLM requires a filename, minimally!\n"; abort(); } - else if (argc == 1) { filename = argv[0]; } - else if (argc == 2 || argc > 3) { cerr << "Don't understand 'RandLM " << param << "'\n"; } - else if (argc == 3) { - if (argv[0] == "-o") { - order = atoi(argv[1].c_str()); - filename = argv[2]; - } else if (argv[1] == "-o") { - order = atoi(argv[2].c_str()); - filename = argv[0]; - } - } -// set_order(order); - int cache_MB = 200; // increase cache size - randlm::RandLM* rlm = randlm::RandLM::initRandLM(filename, order, cache_MB); - assert(rlm != NULL); - pimpl_ = new RandLMImpl(order, rlm); -} - -LanguageModelRandLM::~LanguageModelRandLM() { - delete pimpl_; -} - -void LanguageModelRandLM::TraversalFeaturesImpl(const SentenceMetadata& smeta, - const Hypergraph::Edge& edge, - const vector& ant_states, - SparseVector* features, - SparseVector* estimated_features, - void* state) const { - (void) smeta; - features->set_value(fid_, imp().LookupWords(*edge.rule_, ant_states, state)); - estimated_features->set_value(fid_, imp().EstimateProb(state)); -} - -void LanguageModelRandLM::FinalTraversalFeatures(const void* ant_state, - SparseVector* features) const { - features->set_value(fid_, imp().FinalTraversalCost(ant_state)); -} - -#endif - diff --git a/decoder/ff_lm.h b/decoder/ff_lm.h index 94e18f00..85e79704 100644 --- a/decoder/ff_lm.h +++ b/decoder/ff_lm.h @@ -69,26 +69,4 @@ class LanguageModel : public FeatureFunction { /* mutable */ LanguageModelInterface* pimpl_; }; -#ifdef HAVE_RANDLM -class LanguageModelRandLM : public FeatureFunction { - public: - // param = "filename.lm [-o n]" - LanguageModelRandLM(const std::string& param); - ~LanguageModelRandLM(); - virtual void FinalTraversalFeatures(const void* context, - SparseVector* features) const; - std::string DebugStateToString(const void* state) const; - protected: - virtual void TraversalFeaturesImpl(const SentenceMetadata& smeta, - const HG::Edge& edge, - const std::vector& ant_contexts, - SparseVector* features, - SparseVector* estimated_features, - void* out_context) const; - private: - const int fid_; - mutable LanguageModelImpl* pimpl_; -}; -#endif - #endif diff --git a/decoder/ff_source_syntax.cc b/decoder/ff_source_syntax.cc index a1997695..95709076 100644 --- a/decoder/ff_source_syntax.cc +++ b/decoder/ff_source_syntax.cc @@ -2,6 +2,12 @@ #include #include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_set; } +#endif #include "hg.h" #include "sentence_metadata.h" diff --git a/decoder/ff_source_syntax2.cc b/decoder/ff_source_syntax2.cc index 08ece917..622c6908 100644 --- a/decoder/ff_source_syntax2.cc +++ b/decoder/ff_source_syntax2.cc @@ -3,7 +3,6 @@ #include #include #include -#include #include "sentence_metadata.h" #include "array2d.h" @@ -126,8 +125,7 @@ struct SourceSyntaxFeatures2Impl { Array2D src_tree; // src_tree(i,j) NT = type mutable Array2D > fids_ef; // fires for fully lexicalized - tr1::unordered_set feature_filter; - + unordered_set feature_filter; }; SourceSyntaxFeatures2::SourceSyntaxFeatures2(const string& param) : diff --git a/decoder/ff_source_syntax2_p.cc b/decoder/ff_source_syntax2_p.cc index dfa791ea..130144fa 100644 --- a/decoder/ff_source_syntax2_p.cc +++ b/decoder/ff_source_syntax2_p.cc @@ -3,7 +3,12 @@ #include #include #include -#include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_set; } +#endif #include "sentence_metadata.h" #include "array2d.h" @@ -133,8 +138,7 @@ struct PSourceSyntaxFeatures2Impl { Array2D src_tree; // src_tree(i,j) NT = type mutable Array2D > fids_ef; // fires for fully lexicalized - tr1::unordered_set feature_filter; - + unordered_set feature_filter; }; PSourceSyntaxFeatures2::PSourceSyntaxFeatures2(const string& param) : diff --git a/decoder/ff_source_syntax_p.cc b/decoder/ff_source_syntax_p.cc index cd081544..1d3dc497 100644 --- a/decoder/ff_source_syntax_p.cc +++ b/decoder/ff_source_syntax_p.cc @@ -2,7 +2,12 @@ #include #include -#include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_map; using std::tr1::unordered_set; } +#endif #include "sentence_metadata.h" #include "array2d.h" @@ -139,7 +144,7 @@ struct PSourceSyntaxFeaturesImpl { Array2D src_tree; // src_tree(i,j) NT = type // mutable Array2D fids_cat; // this tends to overfit baddly mutable Array2D > fids_ef; // fires for fully lexicalized - tr1::unordered_set feature_filter; + unordered_set feature_filter; }; PSourceSyntaxFeatures::PSourceSyntaxFeatures(const string& param) : diff --git a/decoder/ff_wordalign.cc b/decoder/ff_wordalign.cc index 1491819d..8ed053c2 100644 --- a/decoder/ff_wordalign.cc +++ b/decoder/ff_wordalign.cc @@ -7,7 +7,12 @@ #include #include #include -#include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_map; } +#endif #include #include "boost/tuple/tuple_comparison.hpp" @@ -249,7 +254,7 @@ void NewJump::FireFeature(const SentenceMetadata& smeta, if (fp1_) get<6>(key) = GetSourceWord(id, cur_src_index + 1); if (fprev_) get<7>(key) = GetSourceWord(id, prev_src_index); - static std::tr1::unordered_map fids; + static std::unordered_map fids; int& fid = fids[key]; if (!fid) { ostringstream os; diff --git a/decoder/ff_wordalign.h b/decoder/ff_wordalign.h index ba3d0b9b..50f0dafa 100644 --- a/decoder/ff_wordalign.h +++ b/decoder/ff_wordalign.h @@ -5,8 +5,16 @@ #include "array2d.h" #include "factored_lexicon_helper.h" +#include +#include #include #include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_map; } +#endif class RelativeSentencePosition : public FeatureFunction { public: @@ -124,9 +132,6 @@ class LexicalTranslationTrigger : public FeatureFunction { std::vector > triggers_; }; -#include -#include -#include class BlunsomSynchronousParseHack : public FeatureFunction { public: BlunsomSynchronousParseHack(const std::string& param); @@ -196,7 +201,7 @@ class BlunsomSynchronousParseHack : public FeatureFunction { const int fid_; mutable int cur_sent_; - typedef std::tr1::unordered_map, int, boost::hash > > Vec2Int; + typedef std::unordered_map, int, boost::hash > > Vec2Int; mutable Vec2Int cur_map_; const std::vector mutable * cur_ref_; mutable std::vector > refs_; diff --git a/decoder/ff_wordset.cc b/decoder/ff_wordset.cc index 70cea7de..9be6f2e0 100644 --- a/decoder/ff_wordset.cc +++ b/decoder/ff_wordset.cc @@ -2,21 +2,67 @@ #include "hg.h" #include "fdict.h" +#include "filelib.h" +#include #include #include using namespace std; +void WordSet::parseArgs(const string& args, string* featName, string* vocabFile, bool* oovMode) { + vector toks(10); + boost::split(toks, args, boost::is_any_of(" ")); + + *oovMode = false; + + // skip initial feature name + for(vector::const_iterator it = toks.begin(); it != toks.end(); ++it) { + if(*it == "-v") { + *vocabFile = *++it; // copy + + } else if(*it == "-N") { + *featName = *++it; + } else if(*it == "--oov") { + *oovMode = true; + } else { + cerr << "Unrecognized argument: " << *it << endl; + exit(1); + } + } + + if(*featName == "") { + cerr << "featName (-N) not specified for WordSet" << endl; + exit(1); + } + if(*vocabFile == "") { + cerr << "vocabFile (-v) not specified for WordSet" << endl; + exit(1); + } +} + +void WordSet::loadVocab(const string& vocabFile, unordered_set* vocab) { + ReadFile rf(vocabFile); + if (!rf) { + cerr << "Unable to open file: " << vocabFile; + abort(); + } + string line; + while (getline(*rf.stream(), line)) { + boost::trim(line); + if(line.empty()) continue; + WordID vocabId = TD::Convert(line); + vocab->insert(vocabId); + } +} + void WordSet::TraversalFeaturesImpl(const SentenceMetadata& /*smeta*/ , const Hypergraph::Edge& edge, const vector& /* ant_contexts */, SparseVector* features, SparseVector* /* estimated_features */, void* /* context */) const { - double addScore = 0.0; - for(std::vector::const_iterator it = edge.rule_->e_.begin(); it != edge.rule_->e_.end(); ++it) { - + for(vector::const_iterator it = edge.rule_->e_.begin(); it != edge.rule_->e_.end(); ++it) { bool inVocab = (vocab_.find(*it) != vocab_.end()); if(oovMode_ && !inVocab) { addScore += 1.0; diff --git a/decoder/ff_wordset.h b/decoder/ff_wordset.h index 639e1514..affee2f4 100644 --- a/decoder/ff_wordset.h +++ b/decoder/ff_wordset.h @@ -4,14 +4,18 @@ #include "ff.h" #include "tdict.h" -#include -#include - #include #include #include #include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_set; } +#endif + class WordSet : public FeatureFunction { public: // we depend on the order of the initializer list @@ -42,69 +46,12 @@ class WordSet : public FeatureFunction { void* context) const; private: - static void loadVocab(const std::string& vocabFile, std::tr1::unordered_set* vocab) { - - std::ifstream file; - std::string line; - - file.open(vocabFile.c_str(), std::fstream::in); - if (file.is_open()) { - unsigned lineNum = 0; - while (!file.eof()) { - ++lineNum; - getline(file, line); - boost::trim(line); - if(line.empty()) { - continue; - } - - WordID vocabId = TD::Convert(line); - vocab->insert(vocabId); - } - file.close(); - } else { - std::cerr << "Unable to open file: " << vocabFile; - exit(1); - } - } - - static void parseArgs(const std::string& args, std::string* featName, std::string* vocabFile, bool* oovMode) { - - std::vector toks(10); - boost::split(toks, args, boost::is_any_of(" ")); - - *oovMode = false; - - // skip initial feature name - for(std::vector::const_iterator it = toks.begin(); it != toks.end(); ++it) { - if(*it == "-v") { - *vocabFile = *++it; // copy - - } else if(*it == "-N") { - *featName = *++it; - - } else if(*it == "--oov") { - *oovMode = true; - - } else { - std::cerr << "Unrecognized argument: " << *it << std::endl; - exit(1); - } - } - - if(*featName == "") { - std::cerr << "featName (-N) not specified for WordSet" << std::endl; - exit(1); - } - if(*vocabFile == "") { - std::cerr << "vocabFile (-v) not specified for WordSet" << std::endl; - exit(1); - } - } + static void parseArgs(const std::string& args, std::string* featName, std::string* vocabFile, bool* oovMode); + static void loadVocab(const std::string& vocabFile, std::unordered_set* vocab); int fid_; bool oovMode_; - std::tr1::unordered_set vocab_; + std::unordered_set vocab_; }; #endif diff --git a/decoder/grammar.cc b/decoder/grammar.cc index ee43f537..f2530d35 100644 --- a/decoder/grammar.cc +++ b/decoder/grammar.cc @@ -3,15 +3,20 @@ #include #include #include -#include -#include +#ifdef HAVE_CXX11 +# include +# include +#else +# include +# include +namespace std { using std::tr1::unordered_map; using std::tr1::unordered_set; } +#endif #include "rule_lexer.h" #include "filelib.h" #include "tdict.h" using namespace std; -using namespace std::tr1; const vector Grammar::NO_RULES; diff --git a/decoder/hg_intersect.cc b/decoder/hg_intersect.cc index ad5b701a..c5f1cc91 100644 --- a/decoder/hg_intersect.cc +++ b/decoder/hg_intersect.cc @@ -1,7 +1,12 @@ #include "hg_intersect.h" #include -#include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_map; } +#endif #include "fast_lexical_cast.hpp" #include @@ -13,7 +18,6 @@ #include "bottom_up_parser.h" using boost::lexical_cast; -using namespace std::tr1; using namespace std; struct RuleFilter { diff --git a/decoder/kbest.h b/decoder/kbest.h index 44c23151..cd386aef 100644 --- a/decoder/kbest.h +++ b/decoder/kbest.h @@ -3,7 +3,12 @@ #include #include -#include +#ifdef HAVE_CXX11 +# include +#else +# include +namespace std { using std::tr1::unordered_set; } +#endif #include #include @@ -22,7 +27,7 @@ namespace KBest { // optional, filter unique yield strings struct FilterUnique { - std::tr1::unordered_set, boost::hash > > unique; + std::unordered_set, boost::hash > > unique; bool operator()(const std::vector& yield) { return !unique.insert(yield).second; @@ -111,7 +116,7 @@ namespace KBest { }; typedef std::vector CandidateHeap; typedef std::vector DerivationList; - typedef std::tr1::unordered_set< + typedef std::unordered_set< const Derivation*, DerivationUniquenessHash, DerivationUniquenessEquals> UniqueDerivationSet; struct NodeDerivationState { diff --git a/decoder/maxtrans_blunsom.cc b/decoder/maxtrans_blunsom.cc index 774e4170..8d1d471c 100644 --- a/decoder/maxtrans_blunsom.cc +++ b/decoder/maxtrans_blunsom.cc @@ -2,8 +2,14 @@ #include #include -#include -#include +#ifdef HAVE_CXX11 +# include +# include +#else +# include +# include +namespace std { using std::tr1::unordered_map; using std::tr1::unordered_set; } +#endif #include #include @@ -14,7 +20,6 @@ using boost::tuple; using namespace std; -using namespace std::tr1; namespace Hack { diff --git a/decoder/phrasebased_translator.cc b/decoder/phrasebased_translator.cc index d65e44d1..321fb286 100644 --- a/decoder/phrasebased_translator.cc +++ b/decoder/phrasebased_translator.cc @@ -2,8 +2,14 @@ #include #include -#include -#include +#ifdef HAVE_CXX11 +# include +# include +#else +# include +# include +namespace std { using std::tr1::unordered_map; using std::tr1::unordered_set; } +#endif #include #include @@ -17,7 +23,6 @@ #include "array2d.h" using namespace std; -using namespace std::tr1; using namespace boost::tuples; struct Coverage : public vector { diff --git a/decoder/scfg_translator.cc b/decoder/scfg_translator.cc index 6f0b003b..a506c591 100644 --- a/decoder/scfg_translator.cc +++ b/decoder/scfg_translator.cc @@ -1,13 +1,9 @@ -//TODO: bottom-up pruning, with actual final models' (appropriately weighted) heuristics and local scores. - -//TODO: grammar heuristic (min cost of reachable rule set) for binarizations (active edges) if we wish to prune those also - -#include "hash.h" -#include "translator.h" #include #include #include #include +#include "hash.h" +#include "translator.h" #include "hg.h" #include "grammar.h" #include "bottom_up_parser.h" @@ -16,13 +12,11 @@ #include "tdict.h" #include "viterbi.h" #include "verbose.h" -#include #define foreach BOOST_FOREACH #define reverse_foreach BOOST_REVERSE_FOREACH using namespace std; -using namespace std::tr1; static bool printGrammarsUsed = false; struct GlueGrammar : public TextGrammar { diff --git a/decoder/sentence_metadata.h b/decoder/sentence_metadata.h index eab9f15d..52586331 100644 --- a/decoder/sentence_metadata.h +++ b/decoder/sentence_metadata.h @@ -5,7 +5,9 @@ #include #include #include "lattice.h" -#include "scorer.h" + +struct DocScorer; // deprecated, will be removed +struct Score; // deprecated, will be removed struct SentenceMetadata { friend class DecoderImpl; -- cgit v1.2.3