1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
|
// $Id$
/***********************************************************************
Moses - factored phrase-based language decoder
Copyright (C) 2006 University of Edinburgh
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 <cassert>
#include <iostream>
#include <limits>
#include "TranslationOption.h"
#include "TranslationOptionCollection.h"
#include "DummyScoreProducers.h"
#include "Hypothesis.h"
#include "Util.h"
#include "SquareMatrix.h"
#include "LexicalReordering.h"
#include "StaticData.h"
#include "Input.h"
#include "LMList.h"
#include "hash.h"
using namespace std;
unsigned int Hypothesis::s_numNodes = 0;
unsigned int Hypothesis::s_HypothesesCreated = 0;
ObjectPool<Hypothesis> Hypothesis::s_objectPool("Hypothesis", 300000);
Hypothesis::Hypothesis(InputType const& source, const TargetPhrase &emptyTarget)
: m_prevHypo(NULL)
, m_targetPhrase(emptyTarget)
, m_sourcePhrase(0)
, m_sourceCompleted(source.GetSize())
, m_sourceInput(source)
, m_currSourceWordsRange(NOT_FOUND, NOT_FOUND)
, m_currTargetWordsRange(NOT_FOUND, NOT_FOUND)
, m_wordDeleted(false)
, m_languageModelStates(StaticData::Instance()->GetLMSize(), LanguageModel::UnknownState)
#ifdef N_BEST
, m_arcList(NULL)
#endif
, m_id(s_HypothesesCreated++)
{ // used for initial seeding of trans process
// initialize scores
_hash_computed = false;
ResetScore();
}
/***
* continue prevHypo by appending the phrases in transOpt
*/
Hypothesis::Hypothesis(const Hypothesis &prevHypo, const TranslationOption &transOpt)
: m_prevHypo(&prevHypo)
, m_targetPhrase(transOpt.GetTargetPhrase())
, m_sourcePhrase(0)
, m_sourceCompleted (prevHypo.m_sourceCompleted )
, m_sourceInput (prevHypo.m_sourceInput)
, m_currSourceWordsRange (transOpt.GetSourceWordsRange())
, m_currTargetWordsRange ( prevHypo.m_currTargetWordsRange.GetEndPos() + 1
,prevHypo.m_currTargetWordsRange.GetEndPos() + transOpt.GetTargetPhrase().GetSize())
, m_wordDeleted(false)
, m_totalScore(0.0f)
, m_futureScore(0.0f)
, m_scoreBreakdown (prevHypo.m_scoreBreakdown)
, m_languageModelStates(prevHypo.m_languageModelStates)
#ifdef N_BEST
, m_arcList(NULL)
#endif
, m_id(s_HypothesesCreated++)
{
// assert that we are not extending our hypothesis by retranslating something
// that this hypothesis has already translated!
assert(!m_sourceCompleted.Overlap(m_currSourceWordsRange));
_hash_computed = false;
m_sourceCompleted.SetValue(m_currSourceWordsRange.GetStartPos(), m_currSourceWordsRange.GetEndPos(), true);
m_wordDeleted = transOpt.IsDeletionOption();
m_scoreBreakdown.PlusEquals(transOpt.GetScoreBreakdown());
}
Hypothesis::~Hypothesis()
{
#ifdef N_BEST
if (m_arcList)
{
ArcList::iterator iter;
for (iter = m_arcList->begin() ; iter != m_arcList->end() ; ++iter)
{
s_objectPool.freeObject (*iter);
}
m_arcList->clear();
delete m_arcList;
}
#endif
}
#ifdef N_BEST
void Hypothesis::AddArc(Hypothesis *loserHypo)
{
if (!m_arcList) {
if (loserHypo->m_arcList) // we don't have an arcList, but loser does
{
this->m_arcList = loserHypo->m_arcList; // take ownership, we'll delete
loserHypo->m_arcList = 0; // prevent a double deletion
}
else
{ this->m_arcList = new ArcList(); }
} else {
if (loserHypo->m_arcList) { // both have an arc list: merge. delete loser
size_t my_size = m_arcList->size();
size_t add_size = loserHypo->m_arcList->size();
this->m_arcList->resize(my_size + add_size, 0);
std::memcpy(&(*m_arcList)[0] + my_size, &(*m_arcList)[0], add_size * sizeof(Hypothesis *));
delete loserHypo->m_arcList;
loserHypo->m_arcList = 0;
} else { // loserHypo doesn't have any arcs
// DO NOTHING
}
}
m_arcList->push_back(loserHypo);
}
#endif
/***
* return the subclass of Hypothesis most appropriate to the given translation option
*/
Hypothesis* Hypothesis::CreateNext(const TranslationOption &transOpt) const
{
return Create(*this, transOpt);
}
/***
* return the subclass of Hypothesis most appropriate to the given translation option
*/
Hypothesis* Hypothesis::Create(const Hypothesis &prevHypo, const TranslationOption &transOpt)
{
Hypothesis *ptr = s_objectPool.getPtr();
return new(ptr) Hypothesis(prevHypo, transOpt);
}
/***
* return the subclass of Hypothesis most appropriate to the given target phrase
*/
Hypothesis* Hypothesis::Create(InputType const& m_source, const TargetPhrase &emptyTarget)
{
Hypothesis *ptr = s_objectPool.getPtr();
return new(ptr) Hypothesis(m_source, emptyTarget);
}
bool Hypothesis::IsCompatible(const Phrase &phrase) const
{
// make sure factors don't contradict each other
// similar to phrase comparison
if (m_currTargetWordsRange.GetWordsCount() != phrase.GetSize())
{
return false;
}
size_t hypoSize = GetSize();
size_t transOptPos = 0;
for (size_t hypoPos = hypoSize - m_currTargetWordsRange.GetWordsCount() ; hypoPos < hypoSize ; hypoPos++)
{
for (unsigned int currFactor = 0 ; currFactor < NUM_FACTORS ; currFactor++)
{
FactorType factorType = static_cast<FactorType>(currFactor);
const Factor *thisFactor = GetFactor(hypoPos, factorType)
,*compareFactor = phrase.GetFactor(transOptPos, factorType);
if (thisFactor != NULL && compareFactor != NULL && thisFactor != compareFactor)
return false;
}
transOptPos++;
}
return true;
}
#if 0
void Hypothesis::GenerateNGramCompareKey(size_t contextSize)
{
struct MD5Context md5c;
MD5Init(&md5c);
size_t thisSize = this->GetSize();
size_t effectiveContextSize = std::min(thisSize, contextSize);
int start = thisSize - effectiveContextSize;
if (m_currTargetWordsRange.GetWordsCount() > 0) // initial hypothesis check
{
const Hypothesis *curHyp = this;
int curStart = 0;
while (start < (curStart = curHyp->m_currTargetWordsRange.GetStartPos())) {
for (int col = curHyp->m_currTargetWordsRange.GetEndPos(); col >= curStart; col--) {
MD5Update(&md5c,
(unsigned char*)curHyp->GetCurrFactorArray(col - curStart),
sizeof(FactorArray));
curHyp = curHyp->m_prevHypo;
}
}
for (int col = curHyp->m_currTargetWordsRange.GetEndPos(); col >= (int)start; col--) {
MD5Update(&md5c,
(unsigned char*)curHyp->GetCurrFactorArray(col - curStart),
sizeof(FactorArray));
}
}
MD5Final(m_compSignature, &md5c);
}
#endif
void Hypothesis::GenerateNGramCompareHash() const
{
_hash = quick_hash((const char*)&m_languageModelStates[0], sizeof(LanguageModel::State) * m_languageModelStates.size(), 0xcafe5137);
_hash_computed = true;
vector<size_t> wordCoverage = m_sourceCompleted.GetCompressedReprentation();
_hash = quick_hash((const char*)&wordCoverage[0], sizeof(size_t)*wordCoverage.size(), _hash);
}
int Hypothesis::NGramCompare(const Hypothesis &compare) const
{ // -1 = this < compare
// +1 = this > compare
// 0 = this ==compare
if (m_languageModelStates < compare.m_languageModelStates) return -1;
if (m_languageModelStates > compare.m_languageModelStates) return 1;
return 0;
}
/**
* Calculates the overall language model score by combining the scores
* of language models generated for each of the factors. Because the factors
* represent a variety of tag sets, and because factors with smaller tag sets
* (such as POS instead of words) allow us to calculate richer statistics, we
* allow a different length of n-gram to be specified for each factor.
* /param lmListInitial todo - describe this parameter
* /param lmListEnd todo - describe this parameter
*/
void Hypothesis::CalcLMScore(const LMList &languageModels)
{
const size_t startPos = m_currTargetWordsRange.GetStartPos();
LMList::const_iterator iterLM;
size_t lmIdx = 0;
// already have LM scores from previous and trigram score of poss trans.
// just need trigram score of the words of the start of current phrase
for (iterLM = languageModels.begin() ; iterLM != languageModels.end() ; ++iterLM,++lmIdx)
{
const LanguageModel &languageModel = **iterLM;
FactorType factorType = languageModel.GetFactorType();
size_t nGramOrder = languageModel.GetNGramOrder();
size_t currEndPos = m_currTargetWordsRange.GetEndPos();
float lmScore;
if(m_currTargetWordsRange.GetWordsCount() == 0) {
lmScore = 0; //the score associated with dropping source words is not part of the language model
} else { //non-empty target phrase
// 1st n-gram
vector<const Factor*> contextFactor(nGramOrder);
size_t index = 0;
for (int currPos = (int) startPos - (int) nGramOrder + 1 ; currPos <= (int) startPos ; currPos++)
{
if (currPos >= 0)
contextFactor[index++] = GetFactor(currPos, factorType);
else
contextFactor[index++] = languageModel.GetSentenceStart();
}
lmScore = languageModel.GetValue(contextFactor);
//cout<<"context factor: "<<languageModel.GetValue(contextFactor)<<endl;
// main loop
size_t endPos = std::min(startPos + nGramOrder - 2
, currEndPos);
for (size_t currPos = startPos + 1 ; currPos <= endPos ; currPos++)
{
// shift all args down 1 place
for (size_t i = 0 ; i < nGramOrder - 1 ; i++)
contextFactor[i] = contextFactor[i + 1];
// add last factor
contextFactor.back() = GetFactor(currPos, factorType);
lmScore += languageModel.GetValue(contextFactor);
//cout<<"context factor: "<<languageModel.GetValue(contextFactor)<<endl;
}
// end of sentence
if (m_sourceCompleted.IsComplete())
{
const size_t size = GetSize();
contextFactor.back() = languageModel.GetSentenceEnd();
for (size_t i = 0 ; i < nGramOrder - 1 ; i ++)
{
int currPos = size - nGramOrder + i + 1;
if (currPos < 0)
contextFactor[i] = languageModel.GetSentenceStart();
else
contextFactor[i] = GetFactor((size_t)currPos, factorType);
}
lmScore += languageModel.GetValue(contextFactor, &m_languageModelStates[lmIdx]);
} else {
for (size_t currPos = endPos+1; currPos <= currEndPos; currPos++) {
for (size_t i = 0 ; i < nGramOrder - 1 ; i++)
contextFactor[i] = contextFactor[i + 1];
contextFactor.back() = GetFactor(currPos, factorType);
}
m_languageModelStates[lmIdx]=languageModel.GetState(contextFactor);
}
}
m_scoreBreakdown.PlusEquals(&languageModel, lmScore);
}
}
void Hypothesis::CalcDistortionScore()
{
const DistortionScoreProducer *dsp = StaticData::Instance()->GetDistortionScoreProducer();
float distortionScore = dsp->CalculateDistortionScore(
m_prevHypo->GetCurrSourceWordsRange(),
this->GetCurrSourceWordsRange()
);
m_scoreBreakdown.PlusEquals(dsp, distortionScore);
}
void Hypothesis::ResetScore()
{
m_scoreBreakdown.ZeroAll();
m_futureScore = m_totalScore = 0.0f;
}
/***
* calculate the logarithm of our total translation score (sum up components)
*/
void Hypothesis::CalcScore(const StaticData& staticData, const SquareMatrix &futureScore)
{
// DISTORTION COST
CalcDistortionScore();
// LANGUAGE MODEL COST
CalcLMScore(staticData.GetAllLM());
// WORD PENALTY
m_scoreBreakdown.PlusEquals(staticData.GetWordPenaltyProducer(), - (float) m_currTargetWordsRange.GetWordsCount());
// FUTURE COST
CalcFutureScore(futureScore);
//LEXICAL REORDERING COST
LexicalReordering *m_lexReorder = staticData.GetLexReorder();
if (m_lexReorder) {
m_scoreBreakdown.PlusEquals(m_lexReorder, m_lexReorder->CalcScore(this));
}
// TOTAL
m_totalScore = m_scoreBreakdown.InnerProduct(staticData.GetAllWeights()) + m_futureScore;
}
void Hypothesis::CalcFutureScore(const SquareMatrix &futureScore)
{
const size_t maxSize= numeric_limits<size_t>::max();
size_t start = maxSize;
m_futureScore = 0.0f;
for(size_t currPos = 0 ; currPos < m_sourceCompleted.GetSize() ; currPos++)
{
if(m_sourceCompleted.GetValue(currPos) == 0 && start == maxSize)
{
start = currPos;
}
if(m_sourceCompleted.GetValue(currPos) == 1 && start != maxSize)
{
// m_score[ScoreType::FutureScoreEnum] += futureScore[start][currPos - 1];
m_futureScore += futureScore.GetScore(start, currPos - 1);
start = maxSize;
}
}
if (start != maxSize)
{
// m_score[ScoreType::FutureScoreEnum] += futureScore[start][m_sourceCompleted.GetSize() - 1];
m_futureScore += futureScore.GetScore(start, m_sourceCompleted.GetSize() - 1);
}
// add future costs for distortion model
if(StaticData::Instance()->UseDistortionFutureCosts())
m_futureScore += m_sourceCompleted.GetFutureCosts(m_currSourceWordsRange.GetEndPos()) * StaticData::Instance()->GetWeightDistortion();
}
int Hypothesis::GetId()const{
return m_id;
}
const Hypothesis* Hypothesis::GetPrevHypo()const{
return m_prevHypo;
}
/**
* print hypothesis information for pharaoh-style logging
*/
void Hypothesis::PrintHypothesis(const InputType &source, float weightDistortion, float weightWordPenalty) const{
cout<<"creating hypothesis "<< m_id <<" from "<< m_prevHypo->m_id<<" ( ";
int end = m_prevHypo->m_targetPhrase.GetSize()-1;
int start = end-1;
if ( start < 0 ) start = 0;
if ( m_prevHypo->m_currTargetWordsRange.GetStartPos() == NOT_FOUND ) {
cout << "<s> ";
}
else {
cout << "... ";
}
if (end>=0) {
WordsRange range(start, end);
cout << m_prevHypo->m_targetPhrase.GetSubString(range) << " ";
}
cout<<")"<<endl;
cout<<"\tbase score "<< (m_prevHypo->m_totalScore - m_prevHypo->m_futureScore) <<endl;
cout<<"\tcovering "<<m_currSourceWordsRange.GetStartPos()<<"-"<<m_currSourceWordsRange.GetEndPos()<<": "<< source.GetSubString(m_currSourceWordsRange) <<endl;
cout<<"\ttranslated as: "<<m_targetPhrase; // <<" => translation cost "<<m_score[ScoreType::PhraseTrans];
if (m_wordDeleted) cout <<" word_deleted";
cout<<endl;
cout<<"\tdistance: "<<GetCurrSourceWordsRange().CalcDistortion(m_prevHypo->GetCurrSourceWordsRange()); // << " => distortion cost "<<(m_score[ScoreType::Distortion]*weightDistortion)<<endl;
cout<<"\tlanguage model cost "; // <<m_score[ScoreType::LanguageModelScore]<<endl;
cout<<"\tword penalty "; // <<(m_score[ScoreType::WordPenalty]*weightWordPenalty)<<endl;
cout<<"\tscore "<<m_totalScore - m_futureScore<<" + future cost "<<m_futureScore<<" = "<<m_totalScore<<endl;
cout << "\tunweighted feature scores: " << m_scoreBreakdown << endl;
//PrintLMScores();
}
TO_STRING_BODY(Hypothesis)
// friend
ostream& operator<<(ostream& out, const Hypothesis& hypothesis)
{
hypothesis.ToStream(out);
// words bitmap
out << "[" << hypothesis.m_sourceCompleted << "] ";
// scores
out << " [0.12, " << hypothesis.GetTotalScore() << "]";
out << " " << hypothesis.GetScoreBreakdown();
return out;
}
std::string Hypothesis::GetSourcePhraseStringRep() const
{
if(m_sourcePhrase) {
assert(m_sourcePhrase->ToString()==m_sourcePhrase->GetStringRep(WordsRange(0,m_sourcePhrase->GetSize()-1)));
return m_sourcePhrase->ToString();
}
else
return m_sourceInput.GetStringRep(m_currSourceWordsRange);
}
std::string Hypothesis::GetTargetPhraseStringRep() const
{
return m_targetPhrase.GetStringRep(m_currTargetWordsRange);
}
|