void writeCountOfCounts( const string &fileNameCountOfCounts )
{
// open file
Moses::OutputFileStream countOfCountsFile;
bool success = countOfCountsFile.Open(fileNameCountOfCounts.c_str());
if (!success) {
cerr << "ERROR: could not open count-of-counts file "
<< fileNameCountOfCounts << endl;
return;
}
// Kneser-Ney needs the total number of phrase pairs
countOfCountsFile << totalDistinct << endl;
// write out counts
for(int i=1; i<=COC_MAX; i++) {
countOfCountsFile << countOfCounts[ i ] << endl;
}
countOfCountsFile.Close();
}
void processFiles( const std::string& fileNameDirect,
const std::string& fileNameIndirect,
const std::string& fileNameConsolidated,
const std::string& fileNameCountOfCounts,
const std::string& fileNameSourceLabelSet,
const std::string& fileNamePartsOfSpeechVocabulary )
{
if (goodTuringFlag || kneserNeyFlag)
loadCountOfCounts( fileNameCountOfCounts );
// open input files
Moses::InputFileStream fileDirect(fileNameDirect);
UTIL_THROW_IF2(fileDirect.fail(), "could not open phrase table file " << fileNameDirect);
Moses::InputFileStream fileIndirect(fileNameIndirect);
UTIL_THROW_IF2(fileIndirect.fail(), "could not open phrase table file " << fileNameIndirect);
// open output file: consolidated phrase table
Moses::OutputFileStream fileConsolidated;
bool success = fileConsolidated.Open(fileNameConsolidated);
UTIL_THROW_IF2(!success, "could not open output file " << fileNameConsolidated);
// create properties consolidator
// (in case any additional phrase property requires further processing)
MosesTraining::PropertiesConsolidator propertiesConsolidator = MosesTraining::PropertiesConsolidator();
if (sourceLabelsFlag) {
propertiesConsolidator.ActivateSourceLabelsProcessing(fileNameSourceLabelSet);
}
if (partsOfSpeechFlag) {
propertiesConsolidator.ActivatePartsOfSpeechProcessing(fileNamePartsOfSpeechVocabulary);
}
// loop through all extracted phrase translations
int i=0;
while(true) {
i++;
if (i%100000 == 0) std::cerr << "." << std::flush;
std::vector< std::string > itemDirect, itemIndirect;
if (! getLine(fileIndirect, itemIndirect) ||
! getLine(fileDirect, itemDirect))
break;
// direct: target source alignment probabilities
// indirect: source target probabilities
// consistency checks
UTIL_THROW_IF2(itemDirect[0].compare( itemIndirect[0] ) != 0,
"target phrase does not match in line " << i << ": '" << itemDirect[0] << "' != '" << itemIndirect[0] << "'");
UTIL_THROW_IF2(itemDirect[1].compare( itemIndirect[1] ) != 0,
"source phrase does not match in line " << i << ": '" << itemDirect[1] << "' != '" << itemIndirect[1] << "'");
// SCORES ...
std::string directScores, directSparseScores, indirectScores, indirectSparseScores;
breakdownCoreAndSparse( itemDirect[3], directScores, directSparseScores );
breakdownCoreAndSparse( itemIndirect[3], indirectScores, indirectSparseScores );
std::vector<std::string> directCounts;
Moses::Tokenize( directCounts, itemDirect[4] );
std::vector<std::string> indirectCounts;
Moses::Tokenize( indirectCounts, itemIndirect[4] );
float countF = Moses::Scan<float>(directCounts[0]);
float countE = Moses::Scan<float>(indirectCounts[0]);
float countEF = Moses::Scan<float>(indirectCounts[1]);
float n1_F, n1_E;
if (kneserNeyFlag) {
n1_F = Moses::Scan<float>(directCounts[2]);
n1_E = Moses::Scan<float>(indirectCounts[2]);
}
// Good Turing discounting
float adjustedCountEF = countEF;
if (goodTuringFlag && countEF+0.99999 < goodTuringDiscount.size()-1)
adjustedCountEF *= goodTuringDiscount[(int)(countEF+0.99998)];
float adjustedCountEF_indirect = adjustedCountEF;
// Kneser Ney discounting [Foster et al, 2006]
if (kneserNeyFlag) {
float D = kneserNey_D3;
if (countEF < 2) D = kneserNey_D1;
else if (countEF < 3) D = kneserNey_D2;
if (D > countEF) D = countEF - 0.01; // sanity constraint
float p_b_E = n1_E / totalCount; // target phrase prob based on distinct
float alpha_F = D * n1_F / countF; // available mass
adjustedCountEF = countEF - D + countF * alpha_F * p_b_E;
// for indirect
float p_b_F = n1_F / totalCount; // target phrase prob based on distinct
float alpha_E = D * n1_E / countE; // available mass
adjustedCountEF_indirect = countEF - D + countE * alpha_E * p_b_F;
}
// drop due to MinScore thresholding
if ((minScore0 > 0 && adjustedCountEF_indirect/countE < minScore0) ||
(minScore2 > 0 && adjustedCountEF /countF < minScore2)) {
continue;
}
// output phrase pair
fileConsolidated << itemDirect[0] << " ||| ";
if (partsOfSpeechFlag) {
// write POS factor from property
std::vector<std::string> targetTokens;
Moses::Tokenize( targetTokens, itemDirect[1] );
std::vector<std::string> propertyValuePOS;
propertiesConsolidator.GetPOSPropertyValueFromPropertiesString(itemDirect[5], propertyValuePOS);
size_t targetTerminalIndex = 0;
for (std::vector<std::string>::const_iterator targetTokensIt=targetTokens.begin();
targetTokensIt!=targetTokens.end(); ++targetTokensIt) {
fileConsolidated << *targetTokensIt;
if (!isNonTerminal(*targetTokensIt)) {
assert(propertyValuePOS.size() > targetTerminalIndex);
fileConsolidated << "|" << propertyValuePOS[targetTerminalIndex];
++targetTerminalIndex;
}
fileConsolidated << " ";
}
fileConsolidated << "|||";
} else {
fileConsolidated << itemDirect[1] << " |||";
}
// prob indirect
if (!onlyDirectFlag) {
fileConsolidated << " " << maybeLogProb(adjustedCountEF_indirect/countE);
fileConsolidated << " " << indirectScores;
}
// prob direct
fileConsolidated << " " << maybeLogProb(adjustedCountEF/countF);
fileConsolidated << " " << directScores;
// phrase count feature
if (phraseCountFlag) {
fileConsolidated << " " << maybeLogProb(2.718);
}
// low count feature
if (lowCountFlag) {
fileConsolidated << " " << maybeLogProb(std::exp(-1.0/countEF));
}
// count bin feature (as a core feature)
if (countBin.size()>0 && !sparseCountBinFeatureFlag) {
bool foundBin = false;
for(size_t i=0; i < countBin.size(); i++) {
if (!foundBin && countEF <= countBin[i]) {
fileConsolidated << " " << maybeLogProb(2.718);
foundBin = true;
} else {
fileConsolidated << " " << maybeLogProb(1);
}
}
fileConsolidated << " " << maybeLogProb( foundBin ? 1 : 2.718 );
}
// alignment
fileConsolidated << " |||";
if (!itemDirect[2].empty()) {
fileConsolidated << " " << itemDirect[2];;
}
// counts, for debugging
fileConsolidated << " ||| " << countE << " " << countF << " " << countEF;
// sparse features
fileConsolidated << " |||";
if (directSparseScores.compare("") != 0)
fileConsolidated << " " << directSparseScores;
if (indirectSparseScores.compare("") != 0)
fileConsolidated << " " << indirectSparseScores;
// count bin feature (as a sparse feature)
if (sparseCountBinFeatureFlag) {
bool foundBin = false;
for(size_t i=0; i < countBin.size(); i++) {
if (!foundBin && countEF <= countBin[i]) {
fileConsolidated << " cb_";
if (i == 0 && countBin[i] > 1)
fileConsolidated << "1_";
else if (i > 0 && countBin[i-1]+1 < countBin[i])
fileConsolidated << (countBin[i-1]+1) << "_";
fileConsolidated << countBin[i] << " 1";
foundBin = true;
}
}
if (!foundBin) {
fileConsolidated << " cb_max 1";
}
}
// arbitrary key-value pairs
fileConsolidated << " |||";
if (itemDirect.size() >= 6) {
propertiesConsolidator.ProcessPropertiesString(itemDirect[5], fileConsolidated);
}
if (countsProperty) {
fileConsolidated << " {{Counts " << countE << " " << countF << " " << countEF << "}}";
}
fileConsolidated << std::endl;
}
fileDirect.Close();
fileIndirect.Close();
fileConsolidated.Close();
}
void processFiles( char* fileNameDirect, char* fileNameIndirect, char* fileNameConsolidated, char* fileNameCountOfCounts )
{
if (goodTuringFlag || kneserNeyFlag)
loadCountOfCounts( fileNameCountOfCounts );
// open input files
Moses::InputFileStream fileDirect(fileNameDirect);
Moses::InputFileStream fileIndirect(fileNameIndirect);
if (fileDirect.fail()) {
cerr << "ERROR: could not open phrase table file " << fileNameDirect << endl;
exit(1);
}
istream &fileDirectP = fileDirect;
if (fileIndirect.fail()) {
cerr << "ERROR: could not open phrase table file " << fileNameIndirect << endl;
exit(1);
}
istream &fileIndirectP = fileIndirect;
// open output file: consolidated phrase table
Moses::OutputFileStream fileConsolidated;
bool success = fileConsolidated.Open(fileNameConsolidated);
if (!success) {
cerr << "ERROR: could not open output file " << fileNameConsolidated << endl;
exit(1);
}
// loop through all extracted phrase translations
int i=0;
while(true) {
i++;
if (i%100000 == 0) cerr << "." << flush;
vector< string > itemDirect, itemIndirect;
if (! getLine(fileIndirectP,itemIndirect) ||
! getLine(fileDirectP, itemDirect ))
break;
// direct: target source alignment probabilities
// indirect: source target probabilities
// consistency checks
if (itemDirect[0].compare( itemIndirect[0] ) != 0) {
cerr << "ERROR: target phrase does not match in line " << i << ": '"
<< itemDirect[0] << "' != '" << itemIndirect[0] << "'" << endl;
exit(1);
}
if (itemDirect[1].compare( itemIndirect[1] ) != 0) {
cerr << "ERROR: source phrase does not match in line " << i << ": '"
<< itemDirect[1] << "' != '" << itemIndirect[1] << "'" << endl;
exit(1);
}
// output hierarchical phrase pair (with separated labels)
fileConsolidated << itemDirect[0] << " ||| " << itemDirect[1] << " |||";
// SCORES ...
string directScores, directSparseScores, indirectScores, indirectSparseScores;
breakdownCoreAndSparse( itemDirect[3], directScores, directSparseScores );
breakdownCoreAndSparse( itemIndirect[3], indirectScores, indirectSparseScores );
vector<string> directCounts = tokenize(itemDirect[4].c_str());
vector<string> indirectCounts = tokenize(itemIndirect[4].c_str());
float countF = atof(directCounts[0].c_str());
float countE = atof(indirectCounts[0].c_str());
float countEF = atof(indirectCounts[1].c_str());
float n1_F, n1_E;
if (kneserNeyFlag) {
n1_F = atof(directCounts[2].c_str());
n1_E = atof(indirectCounts[2].c_str());
}
// Good Turing discounting
float adjustedCountEF = countEF;
if (goodTuringFlag && countEF+0.99999 < goodTuringDiscount.size()-1)
adjustedCountEF *= goodTuringDiscount[(int)(countEF+0.99998)];
float adjustedCountEF_indirect = adjustedCountEF;
// Kneser Ney discounting [Foster et al, 2006]
if (kneserNeyFlag) {
float D = kneserNey_D3;
if (countEF < 2) D = kneserNey_D1;
else if (countEF < 3) D = kneserNey_D2;
if (D > countEF) D = countEF - 0.01; // sanity constraint
float p_b_E = n1_E / totalCount; // target phrase prob based on distinct
float alpha_F = D * n1_F / countF; // available mass
adjustedCountEF = countEF - D + countF * alpha_F * p_b_E;
// for indirect
float p_b_F = n1_F / totalCount; // target phrase prob based on distinct
float alpha_E = D * n1_E / countE; // available mass
adjustedCountEF_indirect = countEF - D + countE * alpha_E * p_b_F;
}
// prob indirect
if (!onlyDirectFlag) {
fileConsolidated << " " << maybeLogProb(adjustedCountEF_indirect/countE);
fileConsolidated << " " << indirectScores;
}
// prob direct
fileConsolidated << " " << maybeLogProb(adjustedCountEF/countF);
fileConsolidated << " " << directScores;
// phrase count feature
if (phraseCountFlag) {
fileConsolidated << " " << maybeLogProb(2.718);
}
// low count feature
if (lowCountFlag) {
fileConsolidated << " " << maybeLogProb(exp(-1.0/countEF));
}
// count bin feature (as a core feature)
if (countBin.size()>0 && !sparseCountBinFeatureFlag) {
bool foundBin = false;
for(size_t i=0; i < countBin.size(); i++) {
if (!foundBin && countEF <= countBin[i]) {
fileConsolidated << " " << maybeLogProb(2.718);
foundBin = true;
} else {
fileConsolidated << " " << maybeLogProb(1);
}
}
fileConsolidated << " " << maybeLogProb( foundBin ? 1 : 2.718 );
}
// alignment
fileConsolidated << " ||| " << itemDirect[2];
// counts, for debugging
fileConsolidated << "||| " << countE << " " << countF << " " << countEF;
// count bin feature (as a sparse feature)
fileConsolidated << " |||";
if (directSparseScores.compare("") != 0)
fileConsolidated << " " << directSparseScores;
if (indirectSparseScores.compare("") != 0)
fileConsolidated << " " << indirectSparseScores;
if (sparseCountBinFeatureFlag) {
bool foundBin = false;
for(size_t i=0; i < countBin.size(); i++) {
if (!foundBin && countEF <= countBin[i]) {
fileConsolidated << " cb_";
if (i == 0 && countBin[i] > 1)
fileConsolidated << "1_";
else if (i > 0 && countBin[i-1]+1 < countBin[i])
fileConsolidated << (countBin[i-1]+1) << "_";
fileConsolidated << countBin[i] << " 1";
foundBin = true;
}
}
if (!foundBin) {
fileConsolidated << " cb_max 1";
}
}
// arbitrary key-value pairs
if (itemDirect.size() >= 6) {
fileConsolidated << " ||| " << itemDirect[5];
}
fileConsolidated << endl;
}
fileDirect.Close();
fileIndirect.Close();
fileConsolidated.Close();
}
int main(int argc, char* argv[])
{
cerr << "extract-rules, written by Philipp Koehn\n"
<< "rule extraction from an aligned parallel corpus\n";
RuleExtractionOptions options;
int sentenceOffset = 0;
#ifdef WITH_THREADS
int thread_count = 1;
#endif
if (argc < 5) {
cerr << "syntax: extract-rules corpus.target corpus.source corpus.align extract ["
<< " --GlueGrammar FILE"
<< " | --UnknownWordLabel FILE"
<< " | --OnlyDirect"
<< " | --MaxSpan[" << options.maxSpan << "]"
<< " | --MinHoleTarget[" << options.minHoleTarget << "]"
<< " | --MinHoleSource[" << options.minHoleSource << "]"
<< " | --MinWords[" << options.minWords << "]"
<< " | --MaxSymbolsTarget[" << options.maxSymbolsTarget << "]"
<< " | --MaxSymbolsSource[" << options.maxSymbolsSource << "]"
<< " | --MaxNonTerm[" << options.maxNonTerm << "]"
<< " | --MaxScope[" << options.maxScope << "]"
<< " | --SourceSyntax | --TargetSyntax"
<< " | --AllowOnlyUnalignedWords | --DisallowNonTermConsecTarget |--NonTermConsecSource | --NoNonTermFirstWord | --NoFractionalCounting"
<< " | --UnpairedExtractFormat"
<< " | --ConditionOnTargetLHS ]"
<< " | --BoundaryRules[" << options.boundaryRules << "]"
<< " | --FlexibilityScore\n";
exit(1);
}
char* &fileNameT = argv[1];
char* &fileNameS = argv[2];
char* &fileNameA = argv[3];
string fileNameGlueGrammar;
string fileNameUnknownWordLabel;
string fileNameExtract = string(argv[4]);
int optionInd = 5;
for(int i=optionInd; i<argc; i++) {
// maximum span length
if (strcmp(argv[i],"--MaxSpan") == 0) {
options.maxSpan = atoi(argv[++i]);
if (options.maxSpan < 1) {
cerr << "extract error: --maxSpan should be at least 1" << endl;
exit(1);
}
} else if (strcmp(argv[i],"--MinHoleTarget") == 0) {
options.minHoleTarget = atoi(argv[++i]);
if (options.minHoleTarget < 1) {
cerr << "extract error: --minHoleTarget should be at least 1" << endl;
exit(1);
}
} else if (strcmp(argv[i],"--MinHoleSource") == 0) {
options.minHoleSource = atoi(argv[++i]);
if (options.minHoleSource < 1) {
cerr << "extract error: --minHoleSource should be at least 1" << endl;
exit(1);
}
}
// maximum number of words in hierarchical phrase
else if (strcmp(argv[i],"--MaxSymbolsTarget") == 0) {
options.maxSymbolsTarget = atoi(argv[++i]);
if (options.maxSymbolsTarget < 1) {
cerr << "extract error: --MaxSymbolsTarget should be at least 1" << endl;
exit(1);
}
}
// maximum number of words in hierarchical phrase
else if (strcmp(argv[i],"--MaxSymbolsSource") == 0) {
options.maxSymbolsSource = atoi(argv[++i]);
if (options.maxSymbolsSource < 1) {
cerr << "extract error: --MaxSymbolsSource should be at least 1" << endl;
exit(1);
}
}
// minimum number of words in hierarchical phrase
else if (strcmp(argv[i],"--MinWords") == 0) {
options.minWords = atoi(argv[++i]);
if (options.minWords < 0) {
cerr << "extract error: --MinWords should be at least 0" << endl;
exit(1);
}
}
// maximum number of non-terminals
else if (strcmp(argv[i],"--MaxNonTerm") == 0) {
options.maxNonTerm = atoi(argv[++i]);
if (options.maxNonTerm < 1) {
cerr << "extract error: --MaxNonTerm should be at least 1" << endl;
exit(1);
}
}
// maximum scope (see Hopkins and Langmead (2010))
else if (strcmp(argv[i],"--MaxScope") == 0) {
options.maxScope = atoi(argv[++i]);
if (options.maxScope < 0) {
cerr << "extract error: --MaxScope should be at least 0" << endl;
exit(1);
}
} else if (strcmp(argv[i], "--GZOutput") == 0) {
options.gzOutput = true;
}
// allow consecutive non-terminals (X Y | X Y)
else if (strcmp(argv[i],"--TargetSyntax") == 0) {
options.targetSyntax = true;
} else if (strcmp(argv[i],"--SourceSyntax") == 0) {
options.sourceSyntax = true;
} else if (strcmp(argv[i],"--AllowOnlyUnalignedWords") == 0) {
options.requireAlignedWord = false;
} else if (strcmp(argv[i],"--DisallowNonTermConsecTarget") == 0) {
options.nonTermConsecTarget = false;
} else if (strcmp(argv[i],"--NonTermConsecSource") == 0) {
options.nonTermConsecSource = true;
} else if (strcmp(argv[i],"--NoNonTermFirstWord") == 0) {
options.nonTermFirstWord = false;
} else if (strcmp(argv[i],"--OnlyOutputSpanInfo") == 0) {
options.onlyOutputSpanInfo = true;
} else if (strcmp(argv[i],"--OnlyDirect") == 0) {
options.onlyDirectFlag = true;
} else if (strcmp(argv[i],"--GlueGrammar") == 0) {
options.glueGrammarFlag = true;
if (++i >= argc) {
cerr << "ERROR: Option --GlueGrammar requires a file name" << endl;
exit(0);
}
fileNameGlueGrammar = string(argv[i]);
cerr << "creating glue grammar in '" << fileNameGlueGrammar << "'" << endl;
} else if (strcmp(argv[i],"--UnknownWordLabel") == 0) {
options.unknownWordLabelFlag = true;
if (++i >= argc) {
cerr << "ERROR: Option --UnknownWordLabel requires a file name" << endl;
exit(0);
}
fileNameUnknownWordLabel = string(argv[i]);
cerr << "creating unknown word labels in '" << fileNameUnknownWordLabel << "'" << endl;
}
// TODO: this should be a useful option
//else if (strcmp(argv[i],"--ZipFiles") == 0) {
// zipFiles = true;
//}
// if an source phrase is paired with two target phrases, then count(t|s) = 0.5
else if (strcmp(argv[i],"--NoFractionalCounting") == 0) {
options.fractionalCounting = false;
} else if (strcmp(argv[i],"--PCFG") == 0) {
options.pcfgScore = true;
} else if (strcmp(argv[i],"--UnpairedExtractFormat") == 0) {
options.unpairedExtractFormat = true;
} else if (strcmp(argv[i],"--ConditionOnTargetLHS") == 0) {
options.conditionOnTargetLhs = true;
} else if (strcmp(argv[i],"--FlexibilityScore") == 0) {
options.flexScoreFlag = true;
} else if (strcmp(argv[i],"-threads") == 0 ||
strcmp(argv[i],"--threads") == 0 ||
strcmp(argv[i],"--Threads") == 0) {
#ifdef WITH_THREADS
thread_count = atoi(argv[++i]);
#else
cerr << "thread support not compiled in." << '\n';
exit(1);
#endif
} else if (strcmp(argv[i], "--SentenceOffset") == 0) {
if (i+1 >= argc || argv[i+1][0] < '0' || argv[i+1][0] > '9') {
cerr << "extract: syntax error, used switch --SentenceOffset without a number" << endl;
exit(1);
}
sentenceOffset = atoi(argv[++i]);
} else if (strcmp(argv[i],"--BoundaryRules") == 0) {
options.boundaryRules = true;
} else {
cerr << "extract: syntax error, unknown option '" << string(argv[i]) << "'\n";
exit(1);
}
}
cerr << "extracting hierarchical rules" << endl;
// open input files
Moses::InputFileStream tFile(fileNameT);
Moses::InputFileStream sFile(fileNameS);
Moses::InputFileStream aFile(fileNameA);
istream *tFileP = &tFile;
istream *sFileP = &sFile;
istream *aFileP = &aFile;
// open output files
string fileNameExtractInv = fileNameExtract + ".inv" + (options.gzOutput?".gz":"");
Moses::OutputFileStream extractFile;
Moses::OutputFileStream extractFileInv;
Moses::OutputFileStream extractFileContext;
Moses::OutputFileStream extractFileContextInv;
extractFile.Open((fileNameExtract + (options.gzOutput?".gz":"")).c_str());
if (!options.onlyDirectFlag)
extractFileInv.Open(fileNameExtractInv.c_str());
if (options.flexScoreFlag) {
string fileNameExtractContext = fileNameExtract + ".context" + (options.gzOutput?".gz":"");
extractFileContext.Open(fileNameExtractContext.c_str());
if (!options.onlyDirectFlag) {
string fileNameExtractContextInv = fileNameExtract + ".context.inv" + (options.gzOutput?".gz":"");
extractFileContextInv.Open(fileNameExtractContextInv.c_str());
}
}
// stats on labels for glue grammar and unknown word label probabilities
set< string > targetLabelCollection, sourceLabelCollection;
map< string, int > targetTopLabelCollection, sourceTopLabelCollection;
// loop through all sentence pairs
size_t i=sentenceOffset;
string targetString, sourceString, alignmentString;
while(getline(*tFileP, targetString)) {
i++;
getline(*sFileP, sourceString);
getline(*aFileP, alignmentString);
if (i%1000 == 0) cerr << i << " " << flush;
SentenceAlignmentWithSyntax sentence
(targetLabelCollection, sourceLabelCollection,
targetTopLabelCollection, sourceTopLabelCollection, options);
//az: output src, tgt, and alingment line
if (options.onlyOutputSpanInfo) {
cout << "LOG: SRC: " << sourceString << endl;
cout << "LOG: TGT: " << targetString << endl;
cout << "LOG: ALT: " << alignmentString << endl;
cout << "LOG: PHRASES_BEGIN:" << endl;
}
if (sentence.create(targetString.c_str(), sourceString.c_str(), alignmentString.c_str(),"", i, options.boundaryRules)) {
if (options.unknownWordLabelFlag) {
collectWordLabelCounts(sentence);
}
ExtractTask *task = new ExtractTask(sentence, options, extractFile, extractFileInv, extractFileContext, extractFileContextInv);
task->Run();
delete task;
}
if (options.onlyOutputSpanInfo) cout << "LOG: PHRASES_END:" << endl; //az: mark end of phrases
}
tFile.Close();
sFile.Close();
aFile.Close();
// only close if we actually opened it
if (!options.onlyOutputSpanInfo) {
extractFile.Close();
if (!options.onlyDirectFlag) extractFileInv.Close();
}
if (options.flexScoreFlag) {
extractFileContext.Close();
if (!options.onlyDirectFlag) extractFileContextInv.Close();
}
if (options.glueGrammarFlag)
writeGlueGrammar(fileNameGlueGrammar, options, targetLabelCollection, targetTopLabelCollection);
if (options.unknownWordLabelFlag)
writeUnknownWordLabel(fileNameUnknownWordLabel);
}
int main(int argc, char* argv[])
{
cerr << "PhraseExtract v1.4, written by Philipp Koehn\n"
<< "phrase extraction from an aligned parallel corpus\n";
if (argc < 6) {
cerr << "syntax: extract en de align extract max-length [orientation [ --model [wbe|phrase|hier]-[msd|mslr|mono] ] | --OnlyOutputSpanInfo | --NoTTable | --SentenceId]\n";
exit(1);
}
char* &fileNameE = argv[1];
char* &fileNameF = argv[2];
char* &fileNameA = argv[3];
string fileNameExtract = string(argv[4]);
maxPhraseLength = atoi(argv[5]);
for(int i=6; i<argc; i++) {
if (strcmp(argv[i],"--OnlyOutputSpanInfo") == 0) {
onlyOutputSpanInfo = true;
} else if (strcmp(argv[i],"orientation") == 0 || strcmp(argv[i],"--Orientation") == 0) {
orientationFlag = true;
} else if (strcmp(argv[i],"--NoTTable") == 0) {
translationFlag = false;
} else if (strcmp(argv[i], "--SentenceId") == 0) {
sentenceIdFlag = true;
} else if (strcmp(argv[i], "--GZOutput") == 0) {
gzOutput = true;
} else if(strcmp(argv[i],"--model") == 0) {
if (i+1 >= argc) {
cerr << "extract: syntax error, no model's information provided to the option --model " << endl;
exit(1);
}
char* modelParams = argv[++i];
char* modelName = strtok(modelParams, "-");
char* modelType = strtok(NULL, "-");
REO_MODEL_TYPE intModelType;
if(strcmp(modelName, "wbe") == 0) {
wordModel = true;
if(strcmp(modelType, "msd") == 0)
wordType = REO_MSD;
else if(strcmp(modelType, "mslr") == 0)
wordType = REO_MSLR;
else if(strcmp(modelType, "mono") == 0 || strcmp(modelType, "monotonicity") == 0)
wordType = REO_MONO;
else {
cerr << "extract: syntax error, unknown reordering model type: " << modelType << endl;
exit(1);
}
} else if(strcmp(modelName, "phrase") == 0) {
phraseModel = true;
if(strcmp(modelType, "msd") == 0)
phraseType = REO_MSD;
else if(strcmp(modelType, "mslr") == 0)
phraseType = REO_MSLR;
else if(strcmp(modelType, "mono") == 0 || strcmp(modelType, "monotonicity") == 0)
phraseType = REO_MONO;
else {
cerr << "extract: syntax error, unknown reordering model type: " << modelType << endl;
exit(1);
}
} else if(strcmp(modelName, "hier") == 0) {
hierModel = true;
if(strcmp(modelType, "msd") == 0)
hierType = REO_MSD;
else if(strcmp(modelType, "mslr") == 0)
hierType = REO_MSLR;
else if(strcmp(modelType, "mono") == 0 || strcmp(modelType, "monotonicity") == 0)
hierType = REO_MONO;
else {
cerr << "extract: syntax error, unknown reordering model type: " << modelType << endl;
exit(1);
}
} else {
cerr << "extract: syntax error, unknown reordering model: " << modelName << endl;
exit(1);
}
allModelsOutputFlag = true;
} else {
cerr << "extract: syntax error, unknown option '" << string(argv[i]) << "'\n";
exit(1);
}
}
// default reordering model if no model selected
// allows for the old syntax to be used
if(orientationFlag && !allModelsOutputFlag) {
wordModel = true;
wordType = REO_MSD;
}
// open input files
Moses::InputFileStream eFile(fileNameE);
Moses::InputFileStream fFile(fileNameF);
Moses::InputFileStream aFile(fileNameA);
istream *eFileP = &eFile;
istream *fFileP = &fFile;
istream *aFileP = &aFile;
// open output files
if (translationFlag) {
string fileNameExtractInv = fileNameExtract + ".inv" + (gzOutput?".gz":"");
extractFile.Open( (fileNameExtract + (gzOutput?".gz":"")).c_str());
extractFileInv.Open(fileNameExtractInv.c_str());
}
if (orientationFlag) {
string fileNameExtractOrientation = fileNameExtract + ".o" + (gzOutput?".gz":"");
extractFileOrientation.Open(fileNameExtractOrientation.c_str());
}
if (sentenceIdFlag) {
string fileNameExtractSentenceId = fileNameExtract + ".sid" + (gzOutput?".gz":"");
extractFileSentenceId.Open(fileNameExtractSentenceId.c_str());
}
int i=0;
while(true) {
i++;
if (i%10000 == 0) cerr << "." << flush;
char englishString[LINE_MAX_LENGTH];
char foreignString[LINE_MAX_LENGTH];
char alignmentString[LINE_MAX_LENGTH];
SAFE_GETLINE((*eFileP), englishString, LINE_MAX_LENGTH, '\n', __FILE__);
if (eFileP->eof()) break;
SAFE_GETLINE((*fFileP), foreignString, LINE_MAX_LENGTH, '\n', __FILE__);
SAFE_GETLINE((*aFileP), alignmentString, LINE_MAX_LENGTH, '\n', __FILE__);
SentenceAlignment sentence;
// cout << "read in: " << englishString << " & " << foreignString << " & " << alignmentString << endl;
//az: output src, tgt, and alingment line
if (onlyOutputSpanInfo) {
cout << "LOG: SRC: " << foreignString << endl;
cout << "LOG: TGT: " << englishString << endl;
cout << "LOG: ALT: " << alignmentString << endl;
cout << "LOG: PHRASES_BEGIN:" << endl;
}
if (sentence.create( englishString, foreignString, alignmentString, i)) {
extract(sentence);
}
if (onlyOutputSpanInfo) cout << "LOG: PHRASES_END:" << endl; //az: mark end of phrases
}
eFile.Close();
fFile.Close();
aFile.Close();
//az: only close if we actually opened it
if (!onlyOutputSpanInfo) {
if (translationFlag) {
extractFile.Close();
extractFileInv.Close();
}
if (orientationFlag) extractFileOrientation.Close();
if (sentenceIdFlag) {
extractFileSentenceId.Close();
}
}
}
int main(int argc, char* argv[])
{
cerr << "extract-rules, written by Philipp Koehn\n"
<< "rule extraction from an aligned parallel corpus\n";
RuleExtractionOptions options;
#ifdef WITH_THREADS
int thread_count = 1;
#endif
if (argc < 5) {
cerr << "syntax: extract-rules corpus.target corpus.source corpus.align extract ["
#ifdef WITH_THREADS
<< " --threads NUM |"
#endif
<< " --GlueGrammar FILE"
<< " | --UnknownWordLabel FILE"
<< " | --OnlyDirect"
<< " | --OutputNTLengths"
<< " | --MaxSpan[" << options.maxSpan << "]"
<< " | --MinHoleTarget[" << options.minHoleTarget << "]"
<< " | --MinHoleSource[" << options.minHoleSource << "]"
<< " | --MinWords[" << options.minWords << "]"
<< " | --MaxSymbolsTarget[" << options.maxSymbolsTarget << "]"
<< " | --MaxSymbolsSource[" << options.maxSymbolsSource << "]"
<< " | --MaxNonTerm[" << options.maxNonTerm << "]"
<< " | --MaxScope[" << options.maxScope << "]"
<< " | --SourceSyntax | --TargetSyntax"
<< " | --AllowOnlyUnalignedWords | --DisallowNonTermConsecTarget |--NonTermConsecSource | --NoNonTermFirstWord | --NoFractionalCounting ]\n";
exit(1);
}
char* &fileNameT = argv[1];
char* &fileNameS = argv[2];
char* &fileNameA = argv[3];
string fileNameGlueGrammar;
string fileNameUnknownWordLabel;
string fileNameExtract = string(argv[4]);
int optionInd = 5;
for(int i=optionInd; i<argc; i++) {
// maximum span length
if (strcmp(argv[i],"--MaxSpan") == 0) {
options.maxSpan = atoi(argv[++i]);
if (options.maxSpan < 1) {
cerr << "extract error: --maxSpan should be at least 1" << endl;
exit(1);
}
} else if (strcmp(argv[i],"--MinHoleTarget") == 0) {
options.minHoleTarget = atoi(argv[++i]);
if (options.minHoleTarget < 1) {
cerr << "extract error: --minHoleTarget should be at least 1" << endl;
exit(1);
}
} else if (strcmp(argv[i],"--MinHoleSource") == 0) {
options.minHoleSource = atoi(argv[++i]);
if (options.minHoleSource < 1) {
cerr << "extract error: --minHoleSource should be at least 1" << endl;
exit(1);
}
}
// maximum number of words in hierarchical phrase
else if (strcmp(argv[i],"--MaxSymbolsTarget") == 0) {
options.maxSymbolsTarget = atoi(argv[++i]);
if (options.maxSymbolsTarget < 1) {
cerr << "extract error: --MaxSymbolsTarget should be at least 1" << endl;
exit(1);
}
}
// maximum number of words in hierarchical phrase
else if (strcmp(argv[i],"--MaxSymbolsSource") == 0) {
options.maxSymbolsSource = atoi(argv[++i]);
if (options.maxSymbolsSource < 1) {
cerr << "extract error: --MaxSymbolsSource should be at least 1" << endl;
exit(1);
}
}
// minimum number of words in hierarchical phrase
else if (strcmp(argv[i],"--MinWords") == 0) {
options.minWords = atoi(argv[++i]);
if (options.minWords < 0) {
cerr << "extract error: --MinWords should be at least 0" << endl;
exit(1);
}
}
// maximum number of non-terminals
else if (strcmp(argv[i],"--MaxNonTerm") == 0) {
options.maxNonTerm = atoi(argv[++i]);
if (options.maxNonTerm < 1) {
cerr << "extract error: --MaxNonTerm should be at least 1" << endl;
exit(1);
}
}
// maximum scope (see Hopkins and Langmead (2010))
else if (strcmp(argv[i],"--MaxScope") == 0) {
options.maxScope = atoi(argv[++i]);
if (options.maxScope < 0) {
cerr << "extract error: --MaxScope should be at least 0" << endl;
exit(1);
}
}
else if (strcmp(argv[i], "--GZOutput") == 0) {
options.gzOutput = true;
}
// allow consecutive non-terminals (X Y | X Y)
else if (strcmp(argv[i],"--TargetSyntax") == 0) {
options.targetSyntax = true;
} else if (strcmp(argv[i],"--SourceSyntax") == 0) {
options.sourceSyntax = true;
} else if (strcmp(argv[i],"--AllowOnlyUnalignedWords") == 0) {
options.requireAlignedWord = false;
} else if (strcmp(argv[i],"--DisallowNonTermConsecTarget") == 0) {
options.nonTermConsecTarget = false;
} else if (strcmp(argv[i],"--NonTermConsecSource") == 0) {
options.nonTermConsecSource = true;
} else if (strcmp(argv[i],"--NoNonTermFirstWord") == 0) {
options.nonTermFirstWord = false;
} else if (strcmp(argv[i],"--OnlyOutputSpanInfo") == 0) {
options.onlyOutputSpanInfo = true;
} else if (strcmp(argv[i],"--OnlyDirect") == 0) {
options.onlyDirectFlag = true;
} else if (strcmp(argv[i],"--GlueGrammar") == 0) {
options.glueGrammarFlag = true;
if (++i >= argc) {
cerr << "ERROR: Option --GlueGrammar requires a file name" << endl;
exit(0);
}
fileNameGlueGrammar = string(argv[i]);
cerr << "creating glue grammar in '" << fileNameGlueGrammar << "'" << endl;
} else if (strcmp(argv[i],"--UnknownWordLabel") == 0) {
options.unknownWordLabelFlag = true;
if (++i >= argc) {
cerr << "ERROR: Option --UnknownWordLabel requires a file name" << endl;
exit(0);
}
fileNameUnknownWordLabel = string(argv[i]);
cerr << "creating unknown word labels in '" << fileNameUnknownWordLabel << "'" << endl;
}
// TODO: this should be a useful option
//else if (strcmp(argv[i],"--ZipFiles") == 0) {
// zipFiles = true;
//}
// if an source phrase is paired with two target phrases, then count(t|s) = 0.5
else if (strcmp(argv[i],"--NoFractionalCounting") == 0) {
options.fractionalCounting = false;
} else if (strcmp(argv[i],"--OutputNTLengths") == 0) {
options.outputNTLengths = true;
#ifdef WITH_THREADS
} else if (strcmp(argv[i],"-threads") == 0 ||
strcmp(argv[i],"--threads") == 0 ||
strcmp(argv[i],"--Threads") == 0) {
thread_count = atoi(argv[++i]);
#endif
} else {
cerr << "extract: syntax error, unknown option '" << string(argv[i]) << "'\n";
exit(1);
}
}
cerr << "extracting hierarchical rules" << endl;
// open input files
Moses::InputFileStream tFile(fileNameT);
Moses::InputFileStream sFile(fileNameS);
Moses::InputFileStream aFile(fileNameA);
istream *tFileP = &tFile;
istream *sFileP = &sFile;
istream *aFileP = &aFile;
// open output files
string fileNameExtractInv = fileNameExtract + ".inv" + (options.gzOutput?".gz":"");
Moses::OutputFileStream extractFile;
Moses::OutputFileStream extractFileInv;
extractFile.Open((fileNameExtract + (options.gzOutput?".gz":"")).c_str());
if (!options.onlyDirectFlag)
extractFileInv.Open(fileNameExtractInv.c_str());
// output into file
Moses::OutputCollector* extractCollector = new Moses::OutputCollector(&extractFile);
Moses::OutputCollector* extractCollectorInv = new Moses::OutputCollector(&extractFileInv);
// stats on labels for glue grammar and unknown word label probabilities
set< string > targetLabelCollection, sourceLabelCollection;
map< string, int > targetTopLabelCollection, sourceTopLabelCollection;
#ifdef WITH_THREADS
// set up thread pool
Moses::ThreadPool pool(thread_count);
pool.SetQueueLimit(1000);
#endif
// loop through all sentence pairs
size_t i=0;
while(true) {
i++;
if (i%1000 == 0) cerr << "." << flush;
if (i%10000 == 0) cerr << ":" << flush;
if (i%100000 == 0) cerr << "!" << flush;
char targetString[LINE_MAX_LENGTH];
char sourceString[LINE_MAX_LENGTH];
char alignmentString[LINE_MAX_LENGTH];
SAFE_GETLINE((*tFileP), targetString, LINE_MAX_LENGTH, '\n', __FILE__);
if (tFileP->eof()) break;
SAFE_GETLINE((*sFileP), sourceString, LINE_MAX_LENGTH, '\n', __FILE__);
SAFE_GETLINE((*aFileP), alignmentString, LINE_MAX_LENGTH, '\n', __FILE__);
SentenceAlignmentWithSyntax *sentence = new SentenceAlignmentWithSyntax
(targetLabelCollection, sourceLabelCollection,
targetTopLabelCollection, sourceTopLabelCollection, options);
//az: output src, tgt, and alingment line
if (options.onlyOutputSpanInfo) {
cout << "LOG: SRC: " << sourceString << endl;
cout << "LOG: TGT: " << targetString << endl;
cout << "LOG: ALT: " << alignmentString << endl;
cout << "LOG: PHRASES_BEGIN:" << endl;
}
if (sentence->create(targetString, sourceString, alignmentString, i)) {
if (options.unknownWordLabelFlag) {
collectWordLabelCounts(*sentence);
}
ExtractTask *task = new ExtractTask(i-1, sentence, options, extractCollector, extractCollectorInv);
#ifdef WITH_THREADS
if (thread_count == 1) {
task->Run();
delete task;
}
else {
pool.Submit(task);
}
#else
task->Run();
delete task;
#endif
}
if (options.onlyOutputSpanInfo) cout << "LOG: PHRASES_END:" << endl; //az: mark end of phrases
}
#ifdef WITH_THREADS
// wait for all threads to finish
pool.Stop(true);
#endif
tFile.Close();
sFile.Close();
aFile.Close();
// only close if we actually opened it
if (!options.onlyOutputSpanInfo) {
extractFile.Close();
if (!options.onlyDirectFlag) extractFileInv.Close();
}
if (options.glueGrammarFlag)
writeGlueGrammar(fileNameGlueGrammar, options, targetLabelCollection, targetTopLabelCollection);
if (options.unknownWordLabelFlag)
writeUnknownWordLabel(fileNameUnknownWordLabel);
}
int main(int argc, char* argv[])
{
cerr << "Extract v2.0, written by Philipp Koehn\n"
<< "rule extraction from an aligned parallel corpus\n";
//time_t starttime = time(NULL);
Global *global = new Global();
g_global = global;
int sentenceOffset = 0;
if (argc < 5) {
cerr << "syntax: extract-mixed-syntax corpus.target corpus.source corpus.align extract "
<< " [ --Hierarchical | --Orientation"
<< " | --GlueGrammar FILE | --UnknownWordLabel FILE"
<< " | --OnlyDirect"
<< " | --MinHoleSpanSourceDefault[" << global->minHoleSpanSourceDefault << "]"
<< " | --MaxHoleSpanSourceDefault[" << global->maxHoleSpanSourceDefault << "]"
<< " | --MinHoleSpanSourceSyntax[" << global->minHoleSpanSourceSyntax << "]"
<< " | --MaxHoleSpanSourceSyntax[" << global->maxHoleSpanSourceSyntax << "]"
<< " | --MaxSymbols[" << global->maxSymbols<< "]"
<< " | --MaxNonTerm[" << global->maxNonTerm << "]"
<< " | --SourceSyntax | --TargetSyntax"
<< " | --UppermostOnly[" << g_global->uppermostOnly << "]"
<< endl;
exit(1);
}
char* &fileNameT = argv[1];
char* &fileNameS = argv[2];
char* &fileNameA = argv[3];
string fileNameGlueGrammar;
string fileNameUnknownWordLabel;
string fileNameExtract = string(argv[4]);
int optionInd = 5;
for(int i=optionInd;i<argc;i++)
{
if (strcmp(argv[i],"--MinHoleSpanSourceDefault") == 0) {
global->minHoleSpanSourceDefault = atoi(argv[++i]);
if (global->minHoleSpanSourceDefault < 1) {
cerr << "extract error: --minHoleSourceDefault should be at least 1" << endl;
exit(1);
}
}
else if (strcmp(argv[i],"--MaxHoleSpanSourceDefault") == 0) {
global->maxHoleSpanSourceDefault = atoi(argv[++i]);
if (global->maxHoleSpanSourceDefault < 1) {
cerr << "extract error: --maxHoleSourceDefault should be at least 1" << endl;
exit(1);
}
}
else if (strcmp(argv[i],"--MinHoleSpanSourceSyntax") == 0) {
global->minHoleSpanSourceSyntax = atoi(argv[++i]);
if (global->minHoleSpanSourceSyntax < 1) {
cerr << "extract error: --minHoleSourceSyntax should be at least 1" << endl;
exit(1);
}
}
else if (strcmp(argv[i],"--UppermostOnly") == 0) {
global->uppermostOnly = atoi(argv[++i]);
}
else if (strcmp(argv[i],"--MaxHoleSpanSourceSyntax") == 0) {
global->maxHoleSpanSourceSyntax = atoi(argv[++i]);
if (global->maxHoleSpanSourceSyntax < 1) {
cerr << "extract error: --maxHoleSourceSyntax should be at least 1" << endl;
exit(1);
}
}
// maximum number of words in hierarchical phrase
else if (strcmp(argv[i],"--maxSymbols") == 0) {
global->maxSymbols = atoi(argv[++i]);
if (global->maxSymbols < 1) {
cerr << "extract error: --maxSymbols should be at least 1" << endl;
exit(1);
}
}
// maximum number of non-terminals
else if (strcmp(argv[i],"--MaxNonTerm") == 0) {
global->maxNonTerm = atoi(argv[++i]);
if (global->maxNonTerm < 1) {
cerr << "extract error: --MaxNonTerm should be at least 1" << endl;
exit(1);
}
}
// allow consecutive non-terminals (X Y | X Y)
else if (strcmp(argv[i],"--TargetSyntax") == 0) {
global->targetSyntax = true;
}
else if (strcmp(argv[i],"--SourceSyntax") == 0) {
global->sourceSyntax = true;
}
// do not create many part00xx files!
else if (strcmp(argv[i],"--NoFileLimit") == 0) {
// now default
}
else if (strcmp(argv[i],"--GlueGrammar") == 0) {
global->glueGrammarFlag = true;
if (++i >= argc)
{
cerr << "ERROR: Option --GlueGrammar requires a file name" << endl;
exit(0);
}
fileNameGlueGrammar = string(argv[i]);
cerr << "creating glue grammar in '" << fileNameGlueGrammar << "'" << endl;
}
else if (strcmp(argv[i],"--UnknownWordLabel") == 0) {
global->unknownWordLabelFlag = true;
if (++i >= argc)
{
cerr << "ERROR: Option --UnknownWordLabel requires a file name" << endl;
exit(0);
}
fileNameUnknownWordLabel = string(argv[i]);
cerr << "creating unknown word labels in '" << fileNameUnknownWordLabel << "'" << endl;
}
// TODO: this should be a useful option
//else if (strcmp(argv[i],"--ZipFiles") == 0) {
// zipFiles = true;
//}
// if an source phrase is paired with two target phrases, then count(t|s) = 0.5
else if (strcmp(argv[i],"--Mixed") == 0) {
global->mixed = true;
}
else if (strcmp(argv[i],"--AllowDefaultNonTermEdge") == 0) {
global->allowDefaultNonTermEdge = atoi(argv[++i]);
}
else if (strcmp(argv[i], "--GZOutput") == 0) {
global->gzOutput = true;
}
else if (strcmp(argv[i],"--MaxSpan") == 0) {
// ignore
++i;
}
else if (strcmp(argv[i],"--SentenceOffset") == 0) {
if (i+1 >= argc || argv[i+1][0] < '0' || argv[i+1][0] > '9') {
cerr << "extract: syntax error, used switch --SentenceOffset without a number" << endl;
exit(1);
}
sentenceOffset = atoi(argv[++i]);
}
else {
cerr << "extract: syntax error, unknown option '" << string(argv[i]) << "'\n";
exit(1);
}
}
// open input files
Moses::InputFileStream tFile(fileNameT);
Moses::InputFileStream sFile(fileNameS);
Moses::InputFileStream aFile(fileNameA);
// open output files
string fileNameExtractInv = fileNameExtract + ".inv";
if (global->gzOutput) {
fileNameExtract += ".gz";
fileNameExtractInv += ".gz";
}
Moses::OutputFileStream extractFile;
Moses::OutputFileStream extractFileInv;
extractFile.Open(fileNameExtract.c_str());
extractFileInv.Open(fileNameExtractInv.c_str());
// loop through all sentence pairs
int i = sentenceOffset;
while(true) {
i++;
if (i % 1000 == 0) {
cerr << i << " " << flush;
}
string targetString;
string sourceString;
string alignmentString;
bool ok = getline(tFile, targetString);
if (!ok)
break;
getline(sFile, sourceString);
getline(aFile, alignmentString);
//cerr << endl << targetString << endl << sourceString << endl << alignmentString << endl;
//time_t currTime = time(NULL);
//cerr << "A " << (currTime - starttime) << endl;
SentenceAlignment sentencePair;
if (sentencePair.Create( targetString, sourceString, alignmentString, i, *global ))
{
//cerr << sentence.sourceTree << endl;
//cerr << sentence.targetTree << endl;
sentencePair.FindTunnels(*g_global);
//cerr << "C " << (time(NULL) - starttime) << endl;
//cerr << sentencePair << endl;
sentencePair.CreateLattice(*g_global);
//cerr << "D " << (time(NULL) - starttime) << endl;
//cerr << sentencePair << endl;
sentencePair.CreateRules(*g_global);
//cerr << "E " << (time(NULL) - starttime) << endl;
//cerr << sentence.lattice->GetRules().GetSize() << endl;
sentencePair.GetLattice().GetRules().Output(extractFile);
sentencePair.GetLattice().GetRules().OutputInv(extractFileInv);
}
}
tFile.Close();
sFile.Close();
aFile.Close();
extractFile.Close();
extractFileInv.Close();
if (global->glueGrammarFlag) {
writeGlueGrammar(fileNameGlueGrammar, *global, targetLabelCollection, targetTopLabelCollection);
}
delete global;
}
int main(int argc, char* argv[])
{
cerr << "PhraseExtract v1.4, written by Philipp Koehn\n"
<< "phrase extraction from an aligned parallel corpus\n";
if (argc < 6) {
cerr << "syntax: extract en de align extract max-length [orientation [ --model [wbe|phrase|hier]-[msd|mslr|mono] ] ";
cerr<<"| --OnlyOutputSpanInfo | --NoTTable | --GZOutput | --IncludeSentenceId | --SentenceOffset n | --InstanceWeights filename ]\n";
exit(1);
}
Moses::OutputFileStream extractFileOrientation;
const char* const &fileNameE = argv[1];
const char* const &fileNameF = argv[2];
const char* const &fileNameA = argv[3];
const string fileNameExtract = string(argv[4]);
PhraseExtractionOptions options(atoi(argv[5]));
for(int i=6; i<argc; i++) {
if (strcmp(argv[i],"--OnlyOutputSpanInfo") == 0) {
options.initOnlyOutputSpanInfo(true);
} else if (strcmp(argv[i],"orientation") == 0 || strcmp(argv[i],"--Orientation") == 0) {
options.initOrientationFlag(true);
} else if (strcmp(argv[i],"--FlexibilityScore") == 0) {
options.initFlexScoreFlag(true);
} else if (strcmp(argv[i],"--NoTTable") == 0) {
options.initTranslationFlag(false);
} else if (strcmp(argv[i], "--IncludeSentenceId") == 0) {
options.initIncludeSentenceIdFlag(true);
} else if (strcmp(argv[i], "--SentenceOffset") == 0) {
if (i+1 >= argc || argv[i+1][0] < '0' || argv[i+1][0] > '9') {
cerr << "extract: syntax error, used switch --SentenceOffset without a number" << endl;
exit(1);
}
sentenceOffset = atoi(argv[++i]);
} else if (strcmp(argv[i], "--GZOutput") == 0) {
options.initGzOutput(true);
} else if (strcmp(argv[i], "--InstanceWeights") == 0) {
if (i+1 >= argc) {
cerr << "extract: syntax error, used switch --InstanceWeights without file name" << endl;
exit(1);
}
options.initInstanceWeightsFile(argv[++i]);
} else if (strcmp(argv[i], "--Debug") == 0) {
options.debug = true;
} else if (strcmp(argv[i], "--MinPhraseLength") == 0) {
options.minPhraseLength = atoi(argv[++i]);
} else if (strcmp(argv[i], "--Separator") == 0) {
options.separator = argv[++i];
} else if(strcmp(argv[i],"--model") == 0) {
if (i+1 >= argc) {
cerr << "extract: syntax error, no model's information provided to the option --model " << endl;
exit(1);
}
char* modelParams = argv[++i];
char* modelName = strtok(modelParams, "-");
char* modelType = strtok(NULL, "-");
// REO_MODEL_TYPE intModelType;
if(strcmp(modelName, "wbe") == 0) {
options.initWordModel(true);
if(strcmp(modelType, "msd") == 0)
options.initWordType(REO_MSD);
else if(strcmp(modelType, "mslr") == 0)
options.initWordType(REO_MSLR);
else if(strcmp(modelType, "mono") == 0 || strcmp(modelType, "monotonicity") == 0)
options.initWordType(REO_MONO);
else {
cerr << "extract: syntax error, unknown reordering model type: " << modelType << endl;
exit(1);
}
} else if(strcmp(modelName, "phrase") == 0) {
options.initPhraseModel(true);
if(strcmp(modelType, "msd") == 0)
options.initPhraseType(REO_MSD);
else if(strcmp(modelType, "mslr") == 0)
options.initPhraseType(REO_MSLR);
else if(strcmp(modelType, "mono") == 0 || strcmp(modelType, "monotonicity") == 0)
options.initPhraseType(REO_MONO);
else {
cerr << "extract: syntax error, unknown reordering model type: " << modelType << endl;
exit(1);
}
} else if(strcmp(modelName, "hier") == 0) {
options.initHierModel(true);
if(strcmp(modelType, "msd") == 0)
options.initHierType(REO_MSD);
else if(strcmp(modelType, "mslr") == 0)
options.initHierType(REO_MSLR);
else if(strcmp(modelType, "mono") == 0 || strcmp(modelType, "monotonicity") == 0)
options.initHierType(REO_MONO);
else {
cerr << "extract: syntax error, unknown reordering model type: " << modelType << endl;
exit(1);
}
} else {
cerr << "extract: syntax error, unknown reordering model: " << modelName << endl;
exit(1);
}
options.initAllModelsOutputFlag(true);
} else {
cerr << "extract: syntax error, unknown option '" << string(argv[i]) << "'\n";
exit(1);
}
}
// default reordering model if no model selected
// allows for the old syntax to be used
if(options.isOrientationFlag() && !options.isAllModelsOutputFlag()) {
options.initWordModel(true);
options.initWordType(REO_MSD);
}
// open input files
Moses::InputFileStream eFile(fileNameE);
Moses::InputFileStream fFile(fileNameF);
Moses::InputFileStream aFile(fileNameA);
istream *eFileP = &eFile;
istream *fFileP = &fFile;
istream *aFileP = &aFile;
istream *iwFileP = NULL;
auto_ptr<Moses::InputFileStream> instanceWeightsFile;
if (options.getInstanceWeightsFile().length()) {
instanceWeightsFile.reset(new Moses::InputFileStream(options.getInstanceWeightsFile()));
iwFileP = instanceWeightsFile.get();
}
// open output files
if (options.isOrientationFlag()) {
string fileNameExtractOrientation = fileNameExtract + ".o" + (options.isGzOutput()?".gz":"");
extractFileOrientation.Open(fileNameExtractOrientation.c_str());
}
int i = sentenceOffset;
while(true) {
i++;
if (i%10000 == 0) cerr << "." << flush;
char englishString[LINE_MAX_LENGTH];
char foreignString[LINE_MAX_LENGTH];
char alignmentString[LINE_MAX_LENGTH];
char weightString[LINE_MAX_LENGTH];
SAFE_GETLINE((*eFileP), englishString, LINE_MAX_LENGTH, '\n', __FILE__);
if (eFileP->eof()) break;
SAFE_GETLINE((*fFileP), foreignString, LINE_MAX_LENGTH, '\n', __FILE__);
SAFE_GETLINE((*aFileP), alignmentString, LINE_MAX_LENGTH, '\n', __FILE__);
if (iwFileP) {
SAFE_GETLINE((*iwFileP), weightString, LINE_MAX_LENGTH, '\n', __FILE__);
}
SentenceAlignment sentence;
// cout << "read in: " << englishString << " & " << foreignString << " & " << alignmentString << endl;
//az: output src, tgt, and alingment line
if (options.isOnlyOutputSpanInfo()) {
cout << "LOG: SRC: " << foreignString << endl;
cout << "LOG: TGT: " << englishString << endl;
cout << "LOG: ALT: " << alignmentString << endl;
cout << "LOG: PHRASES_BEGIN:" << endl;
}
if (sentence.create( englishString, foreignString, alignmentString, weightString, i, false)) {
ExtractTask *task = new ExtractTask(i-1, sentence, options, extractFileOrientation);
task->Run();
delete task;
}
if (options.isOnlyOutputSpanInfo()) cout << "LOG: PHRASES_END:" << endl; //az: mark end of phrases
}
eFile.Close();
fFile.Close();
aFile.Close();
//az: only close if we actually opened it
if (!options.isOnlyOutputSpanInfo()) {
if (options.isOrientationFlag()) {
extractFileOrientation.Close();
}
}
}