mosesdecoder/moses/src/ChartManager.cpp

474 lines
17 KiB
C++
Raw Normal View History

// $Id$
// vim:tabstop=2
/***********************************************************************
Moses - factored phrase-based language decoder
Copyright (C) 2010 Hieu Hoang
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
***********************************************************************/
#include <stdio.h>
#include "ChartManager.h"
#include "ChartCell.h"
#include "ChartHypothesis.h"
#include "ChartTrellisDetourQueue.h"
#include "ChartTrellisNode.h"
#include "ChartTrellisPath.h"
#include "ChartTrellisPathList.h"
#include "StaticData.h"
#include "DecodeStep.h"
#include "TreeInput.h"
#include "DummyScoreProducers.h"
using namespace std;
using namespace Moses;
namespace Moses
{
extern bool g_debug;
2012-07-03 21:11:53 +04:00
/* constructor. Initialize everything prior to decoding a particular sentence.
* \param source the sentence to be decoded
* \param system which particular set of models to use.
*/
ChartManager::ChartManager(InputType const& source, const TranslationSystem* system)
:m_source(source)
,m_hypoStackColl(source, *this)
,m_system(system)
,m_start(clock())
,m_hypothesisId(0)
,m_translationOptionList(StaticData::Instance().GetRuleLimit())
,m_decodeGraphList(system->GetDecodeGraphs())
{
m_system->InitializeBeforeSentenceProcessing(source);
const std::vector<PhraseDictionaryFeature*> &dictionaries = m_system->GetPhraseDictionaries();
m_ruleLookupManagers.reserve(dictionaries.size());
for (std::vector<PhraseDictionaryFeature*>::const_iterator p = dictionaries.begin();
p != dictionaries.end(); ++p) {
PhraseDictionaryFeature *pdf = *p;
const PhraseDictionary *dict = pdf->GetDictionary();
PhraseDictionary *nonConstDict = const_cast<PhraseDictionary*>(dict);
m_ruleLookupManagers.push_back(nonConstDict->CreateRuleLookupManager(source, m_hypoStackColl));
}
}
ChartManager::~ChartManager()
{
m_system->CleanUpAfterSentenceProcessing(m_source);
RemoveAllInColl(m_ruleLookupManagers);
RemoveAllInColl(m_unksrcs);
RemoveAllInColl(m_cacheTargetPhraseCollection);
clock_t end = clock();
float et = (end - m_start);
et /= (float)CLOCKS_PER_SEC;
VERBOSE(1, "Translation took " << et << " seconds" << endl);
}
2012-07-03 21:11:53 +04:00
//! decode the sentence. This contains the main laps. Basically, the CKY++ algorithm
void ChartManager::ProcessSentence()
{
VERBOSE(1,"Translating: " << m_source << endl);
ResetSentenceStats(m_source);
VERBOSE(2,"Decoding: " << endl);
//ChartHypothesis::ResetHypoCount();
AddXmlChartOptions();
// MAIN LOOP
size_t size = m_source.GetSize();
for (size_t width = 1; width <= size; ++width) {
for (size_t startPos = 0; startPos <= size-width; ++startPos) {
size_t endPos = startPos + width - 1;
WordsRange range(startPos, endPos);
// create trans opt
CreateTranslationOptionsForRange(range);
// decode
ChartCell &cell = m_hypoStackColl.Get(range);
cell.ProcessSentence(m_translationOptionList, m_hypoStackColl);
m_translationOptionList.Clear();
cell.PruneToSize();
cell.CleanupArcList();
cell.SortHypotheses();
}
}
IFVERBOSE(1) {
for (size_t startPos = 0; startPos < size; ++startPos) {
cerr.width(3);
cerr << startPos << " ";
}
cerr << endl;
for (size_t width = 1; width <= size; width++) {
for( size_t space = 0; space < width-1; space++ ) {
cerr << " ";
}
for (size_t startPos = 0; startPos <= size-width; ++startPos) {
WordsRange range(startPos, startPos+width-1);
cerr.width(3);
cerr << m_hypoStackColl.Get(range).GetSize() << " ";
}
cerr << endl;
}
}
}
2012-07-03 21:11:53 +04:00
/** add specific translation options and hypotheses according to the XML override translation scheme.
* Doesn't seem to do anything about walls and zones.
* @todo check walls & zones. Check that the implementation doesn't leak, xml options sometimes does if you're not careful
*/
void ChartManager::AddXmlChartOptions() {
const std::vector <ChartTranslationOption*> xmlChartOptionsList = m_source.GetXmlChartTranslationOptions();
IFVERBOSE(2) { cerr << "AddXmlChartOptions " << xmlChartOptionsList.size() << endl; }
if (xmlChartOptionsList.size() == 0) return;
for(std::vector<ChartTranslationOption*>::const_iterator i = xmlChartOptionsList.begin();
i != xmlChartOptionsList.end(); ++i) {
ChartTranslationOption* opt = *i;
Moses::Scores wordPenaltyScore(1, -0.434294482); // TODO what is this number?
opt->GetTargetPhraseCollection().GetCollection()[0]->SetScore((ScoreProducer*)m_system->GetWordPenaltyProducer(), wordPenaltyScore);
const WordsRange &range = opt->GetSourceWordsRange();
RuleCubeItem* item = new RuleCubeItem( *opt, m_hypoStackColl );
ChartHypothesis* hypo = new ChartHypothesis(*opt, *item, *this);
hypo->CalcScore();
ChartCell &cell = m_hypoStackColl.Get(range);
cell.AddHypothesis(hypo);
}
}
2012-07-03 21:11:53 +04:00
//! get best complete translation from the top chart cell.
const ChartHypothesis *ChartManager::GetBestHypothesis() const
{
size_t size = m_source.GetSize();
if (size == 0) // empty source
return NULL;
else {
WordsRange range(0, size-1);
const ChartCell &lastCell = m_hypoStackColl.Get(range);
return lastCell.GetBestHypothesis();
}
}
2012-07-03 21:11:53 +04:00
/** Calculate the n-best paths through the output hypergraph.
* Return the list of paths with the variable ret
* \param count how may paths to return
* \param ret return argument
* \param onlyDistinct whether to check for distinct output sentence or not (default - don't check, just return top n-paths)
*/
void ChartManager::CalcNBest(size_t count, ChartTrellisPathList &ret,bool onlyDistinct) const
{
size_t size = m_source.GetSize();
if (count == 0 || size == 0)
return;
// Build a ChartTrellisPath for the 1-best path, if any.
WordsRange range(0, size-1);
const ChartCell &lastCell = m_hypoStackColl.Get(range);
const ChartHypothesis *hypo = lastCell.GetBestHypothesis();
if (hypo == NULL) {
// no hypothesis
return;
}
boost::shared_ptr<ChartTrellisPath> basePath(new ChartTrellisPath(*hypo));
// Add it to the n-best list.
ret.Add(basePath);
if (count == 1) {
return;
}
// Record the output phrase if distinct translations are required.
set<Phrase> distinctHyps;
if (onlyDistinct) {
distinctHyps.insert(basePath->GetOutputPhrase());
}
// Set a limit on the number of detours to pop. If the n-best list is
// restricted to distinct translations then this limit should be bigger
// than n. The n-best factor determines how much bigger the limit should be.
const size_t nBestFactor = StaticData::Instance().GetNBestFactor();
size_t popLimit;
if (!onlyDistinct) {
popLimit = count-1;
} else if (nBestFactor == 0) {
// 0 = 'unlimited.' This actually sets a large-ish limit in case too many
// translations are identical.
popLimit = count * 1000;
} else {
popLimit = count * nBestFactor;
}
// Create an empty priority queue of detour objects. It is bounded to
// contain no more than popLimit items.
ChartTrellisDetourQueue contenders(popLimit);
// Create a ChartTrellisDetour for each single-point deviation from basePath
// and add them to the queue.
CreateDeviantPaths(basePath, contenders);
// MAIN loop
for (size_t i = 0;
ret.GetSize() < count && !contenders.Empty() && i < popLimit;
++i) {
// Get the best detour from the queue.
std::auto_ptr<const ChartTrellisDetour> detour(contenders.Pop());
CHECK(detour.get());
// Create a full base path from the chosen detour.
basePath.reset(new ChartTrellisPath(*detour));
// Generate new detours from this base path and add them to the queue of
// contenders. The new detours deviate from the base path by a single
// replacement along the previous detour sub-path.
CHECK(basePath->GetDeviationPoint());
CreateDeviantPaths(basePath, *(basePath->GetDeviationPoint()), contenders);
// If the n-best list is allowed to contain duplicate translations (at the
// surface level) then add the new path unconditionally, otherwise check
// whether the translation has seen before.
if (!onlyDistinct) {
ret.Add(basePath);
} else {
Phrase tgtPhrase = basePath->GetOutputPhrase();
if (distinctHyps.insert(tgtPhrase).second) {
ret.Add(basePath);
}
}
}
}
void ChartManager::GetSearchGraph(long translationId, std::ostream &outputSearchGraphStream) const
{
size_t size = m_source.GetSize();
// which hypotheses are reachable?
std::map<unsigned,bool> reachable;
WordsRange fullRange(0, size-1);
const ChartCell &lastCell = m_hypoStackColl.Get(fullRange);
const ChartHypothesis *hypo = lastCell.GetBestHypothesis();
if (hypo == NULL) {
// no hypothesis
return;
}
FindReachableHypotheses( hypo, reachable);
for (size_t width = 1; width <= size; ++width) {
for (size_t startPos = 0; startPos <= size-width; ++startPos) {
size_t endPos = startPos + width - 1;
WordsRange range(startPos, endPos);
TRACE_ERR(" " << range << "=");
const ChartCell &cell = m_hypoStackColl.Get(range);
cell.GetSearchGraph(translationId, outputSearchGraphStream, reachable);
}
}
}
void ChartManager::FindReachableHypotheses( const ChartHypothesis *hypo, std::map<unsigned,bool> &reachable ) const
{
// do not recurse, if already visited
if (reachable.find(hypo->GetId()) != reachable.end())
{
return;
}
// recurse
reachable[ hypo->GetId() ] = true;
const std::vector<const ChartHypothesis*> &previous = hypo->GetPrevHypos();
for(std::vector<const ChartHypothesis*>::const_iterator i = previous.begin(); i != previous.end(); ++i)
{
FindReachableHypotheses( *i, reachable );
}
// also loop over recombined hypotheses (arcs)
const ChartArcList *arcList = hypo->GetArcList();
if (arcList) {
ChartArcList::const_iterator iterArc;
for (iterArc = arcList->begin(); iterArc != arcList->end(); ++iterArc) {
const ChartHypothesis &arc = **iterArc;
FindReachableHypotheses( &arc, reachable );
}
}
}
void ChartManager::CreateDeviantPaths(
boost::shared_ptr<const ChartTrellisPath> basePath,
ChartTrellisDetourQueue &q)
{
CreateDeviantPaths(basePath, basePath->GetFinalNode(), q);
}
void ChartManager::CreateDeviantPaths(
boost::shared_ptr<const ChartTrellisPath> basePath,
const ChartTrellisNode &substitutedNode,
ChartTrellisDetourQueue &queue)
{
const ChartArcList *arcList = substitutedNode.GetHypothesis().GetArcList();
if (arcList) {
for (ChartArcList::const_iterator iter = arcList->begin();
iter != arcList->end(); ++iter) {
const ChartHypothesis &replacement = **iter;
queue.Push(new ChartTrellisDetour(basePath, substitutedNode,
replacement));
}
}
// recusively create deviant paths for child nodes
const ChartTrellisNode::NodeChildren &children = substitutedNode.GetChildren();
ChartTrellisNode::NodeChildren::const_iterator iter;
for (iter = children.begin(); iter != children.end(); ++iter) {
const ChartTrellisNode &child = **iter;
CreateDeviantPaths(basePath, child, queue);
}
}
void ChartManager::CreateTranslationOptionsForRange(const WordsRange &wordsRange)
{
assert(m_decodeGraphList.size() == m_ruleLookupManagers.size());
m_translationOptionList.Clear();
std::vector <DecodeGraph*>::const_iterator iterDecodeGraph;
std::vector <ChartRuleLookupManager*>::const_iterator iterRuleLookupManagers = m_ruleLookupManagers.begin();
for (iterDecodeGraph = m_decodeGraphList.begin(); iterDecodeGraph != m_decodeGraphList.end(); ++iterDecodeGraph, ++iterRuleLookupManagers) {
const DecodeGraph &decodeGraph = **iterDecodeGraph;
assert(decodeGraph.GetSize() == 1);
ChartRuleLookupManager &ruleLookupManager = **iterRuleLookupManagers;
size_t maxSpan = decodeGraph.GetMaxChartSpan();
if (maxSpan == 0 || wordsRange.GetNumWordsCovered() <= maxSpan) {
ruleLookupManager.GetChartRuleCollection(wordsRange, m_translationOptionList);
}
}
if (wordsRange.GetNumWordsCovered() == 1 && wordsRange.GetStartPos() != 0 && wordsRange.GetStartPos() != m_source.GetSize()-1) {
bool alwaysCreateDirectTranslationOption = StaticData::Instance().IsAlwaysCreateDirectTranslationOption();
if (m_translationOptionList.GetSize() == 0 || alwaysCreateDirectTranslationOption) {
// create unknown words for 1 word coverage where we don't have any trans options
const Word &sourceWord = m_source.GetWord(wordsRange.GetStartPos());
ProcessOneUnknownWord(sourceWord, wordsRange);
}
}
m_translationOptionList.ApplyThreshold();
}
//! special handling of ONE unknown words.
void ChartManager::ProcessOneUnknownWord(const Word &sourceWord, const WordsRange &range)
{
// unknown word, add as trans opt
const StaticData &staticData = StaticData::Instance();
const UnknownWordPenaltyProducer *unknownWordPenaltyProducer = m_system->GetUnknownWordPenaltyProducer();
vector<float> wordPenaltyScore(1, -0.434294482); // TODO what is this number?
const ChartCell &chartCell = m_hypoStackColl.Get(range);
const ChartCellLabel &sourceWordLabel = chartCell.GetSourceWordLabel();
size_t isDigit = 0;
if (staticData.GetDropUnknown()) {
const Factor *f = sourceWord[0]; // TODO hack. shouldn't know which factor is surface
const string &s = f->GetString();
isDigit = s.find_first_of("0123456789");
if (isDigit == string::npos)
isDigit = 0;
else
isDigit = 1;
// modify the starting bitmap
}
Phrase* m_unksrc = new Phrase(1);
m_unksrc->AddWord() = sourceWord;
m_unksrcs.push_back(m_unksrc);
//TranslationOption *transOpt;
if (! staticData.GetDropUnknown() || isDigit) {
// loop
const UnknownLHSList &lhsList = staticData.GetUnknownLHS();
UnknownLHSList::const_iterator iterLHS;
for (iterLHS = lhsList.begin(); iterLHS != lhsList.end(); ++iterLHS) {
const string &targetLHSStr = iterLHS->first;
float prob = iterLHS->second;
// lhs
//const Word &sourceLHS = staticData.GetInputDefaultNonTerminal();
Word targetLHS(true);
targetLHS.CreateFromString(Output, staticData.GetOutputFactorOrder(), targetLHSStr, true);
CHECK(targetLHS.GetFactor(0) != NULL);
// add to dictionary
TargetPhrase *targetPhrase = new TargetPhrase(Output);
TargetPhraseCollection *tpc = new TargetPhraseCollection;
tpc->Add(targetPhrase);
m_cacheTargetPhraseCollection.push_back(tpc);
Word &targetWord = targetPhrase->AddWord();
targetWord.CreateUnknownWord(sourceWord);
// scores
vector<float> unknownScore(1, FloorScore(TransformScore(prob)));
//targetPhrase->SetScore();
targetPhrase->SetScore(unknownWordPenaltyProducer, unknownScore);
targetPhrase->SetScore(m_system->GetWordPenaltyProducer(), wordPenaltyScore);
targetPhrase->SetSourcePhrase(m_unksrc);
targetPhrase->SetTargetLHS(targetLHS);
// chart rule
m_translationOptionList.Add(*tpc, m_emptyStackVec, range);
} // for (iterLHS
} else {
// drop source word. create blank trans opt
vector<float> unknownScore(1, FloorScore(-numeric_limits<float>::infinity()));
TargetPhrase *targetPhrase = new TargetPhrase(Output);
TargetPhraseCollection *tpc = new TargetPhraseCollection;
tpc->Add(targetPhrase);
// loop
const UnknownLHSList &lhsList = staticData.GetUnknownLHS();
UnknownLHSList::const_iterator iterLHS;
for (iterLHS = lhsList.begin(); iterLHS != lhsList.end(); ++iterLHS) {
const string &targetLHSStr = iterLHS->first;
//float prob = iterLHS->second;
Word targetLHS(true);
targetLHS.CreateFromString(Output, staticData.GetOutputFactorOrder(), targetLHSStr, true);
CHECK(targetLHS.GetFactor(0) != NULL);
m_cacheTargetPhraseCollection.push_back(tpc);
targetPhrase->SetSourcePhrase(m_unksrc);
targetPhrase->SetScore(unknownWordPenaltyProducer, unknownScore);
targetPhrase->SetTargetLHS(targetLHS);
// chart rule
m_translationOptionList.Add(*tpc, m_emptyStackVec, range);
}
}
}
} // namespace Moses