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
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
|
// $Id$
// vim:tabstop=2
/***********************************************************************
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 <algorithm>
#include <typeinfo>
#include "TranslationOptionCollection.h"
#include "Sentence.h"
#include "DecodeStep.h"
#include "LM/Base.h"
#include "FactorCollection.h"
#include "InputType.h"
#include "LexicalReordering.h"
#include "Util.h"
#include "StaticData.h"
#include "DecodeStepTranslation.h"
#include "DecodeGraph.h"
#include "moses/FF/UnknownWordPenaltyProducer.h"
using namespace std;
namespace Moses
{
/** helper for pruning */
bool CompareTranslationOption(const TranslationOption *a, const TranslationOption *b)
{
return a->GetFutureScore() > b->GetFutureScore();
}
/** constructor; since translation options are indexed by coverage span, the corresponding data structure is initialized here
* This fn should be called by inherited classes
*/
TranslationOptionCollection::TranslationOptionCollection(
InputType const& src, size_t maxNoTransOptPerCoverage, float translationOptionThreshold)
: m_source(src)
,m_futureScore(src.GetSize())
,m_maxNoTransOptPerCoverage(maxNoTransOptPerCoverage)
,m_translationOptionThreshold(translationOptionThreshold)
{
// create 2-d vector
size_t size = src.GetSize();
for (size_t startPos = 0 ; startPos < size ; ++startPos) {
m_collection.push_back( vector< TranslationOptionList >() );
size_t maxSize = size - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for (size_t endPos = 0 ; endPos < maxSize ; ++endPos) {
m_collection[startPos].push_back( TranslationOptionList() );
}
}
}
/** destructor, clears out data structures */
TranslationOptionCollection::~TranslationOptionCollection()
{
RemoveAllInColl(m_unksrcs);
}
void TranslationOptionCollection::Prune()
{
// quit, if max size, threshold
if (m_maxNoTransOptPerCoverage == 0 && m_translationOptionThreshold == -std::numeric_limits<float>::infinity())
return;
// bookkeeping for how many options used, pruned
size_t total = 0;
size_t totalPruned = 0;
// loop through all spans
size_t size = m_source.GetSize();
for (size_t startPos = 0 ; startPos < size; ++startPos) {
size_t maxSize = size - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for (size_t endPos = startPos ; endPos < startPos + maxSize ; ++endPos) {
// consider list for a span
TranslationOptionList &fullList = GetTranslationOptionList(startPos, endPos);
total += fullList.size();
// size pruning
if (m_maxNoTransOptPerCoverage > 0 &&
fullList.size() > m_maxNoTransOptPerCoverage) {
// sort in vector
nth_element(fullList.begin(), fullList.begin() + m_maxNoTransOptPerCoverage, fullList.end(), CompareTranslationOption);
totalPruned += fullList.size() - m_maxNoTransOptPerCoverage;
// delete the rest
for (size_t i = m_maxNoTransOptPerCoverage ; i < fullList.size() ; ++i) {
delete fullList.Get(i);
}
fullList.resize(m_maxNoTransOptPerCoverage);
}
// threshold pruning
if (fullList.size() > 1 && m_translationOptionThreshold != -std::numeric_limits<float>::infinity()) {
// first, find the best score
float bestScore = -std::numeric_limits<float>::infinity();
for (size_t i=0; i < fullList.size() ; ++i) {
if (fullList.Get(i)->GetFutureScore() > bestScore)
bestScore = fullList.Get(i)->GetFutureScore();
}
//std::cerr << "best score for span " << startPos << "-" << endPos << " is " << bestScore << "\n";
// then, remove items that are worse than best score + threshold
for (size_t i=0; i < fullList.size() ; ++i) {
if (fullList.Get(i)->GetFutureScore() < bestScore + m_translationOptionThreshold) {
//std::cerr << "\tremoving item " << i << ", score " << fullList.Get(i)->GetFutureScore() << ": " << fullList.Get(i)->GetTargetPhrase() << "\n";
delete fullList.Get(i);
fullList.Remove(i);
total--;
totalPruned++;
i--;
}
//else
//{
// std::cerr << "\tkeeping item " << i << ", score " << fullList.Get(i)->GetFutureScore() << ": " << fullList.Get(i)->GetTargetPhrase() << "\n";
//}
}
} // end of threshold pruning
}
} // end of loop through all spans
VERBOSE(2," Total translation options: " << total << std::endl
<< "Total translation options pruned: " << totalPruned << std::endl);
}
/** Force a creation of a translation option where there are none for a particular source position.
* ie. where a source word has not been translated, create a translation option by
* 1. not observing the table limits on phrase/generation tables
* 2. using the handler ProcessUnknownWord()
* Call this function once translation option collection has been filled with translation options
*
* This function calls for unknown words is complicated by the fact it must handle different input types.
* The call stack is
* Base::ProcessUnknownWord()
* Inherited::ProcessUnknownWord(position)
* Base::ProcessOneUnknownWord()
*
*/
void TranslationOptionCollection::ProcessUnknownWord()
{
const vector<DecodeGraph*>& decodeGraphList = StaticData::Instance().GetDecodeGraphs();
size_t size = m_source.GetSize();
// try to translation for coverage with no trans by expanding table limit
for (size_t graphInd = 0 ; graphInd < decodeGraphList.size() ; graphInd++) {
const DecodeGraph &decodeGraph = *decodeGraphList[graphInd];
for (size_t pos = 0 ; pos < size ; ++pos) {
TranslationOptionList &fullList = GetTranslationOptionList(pos, pos);
size_t numTransOpt = fullList.size();
if (numTransOpt == 0) {
CreateTranslationOptionsForRange(decodeGraph, pos, pos, false, graphInd);
}
}
}
bool alwaysCreateDirectTranslationOption = StaticData::Instance().IsAlwaysCreateDirectTranslationOption();
// create unknown words for 1 word coverage where we don't have any trans options
for (size_t pos = 0 ; pos < size ; ++pos) {
TranslationOptionList &fullList = GetTranslationOptionList(pos, pos);
if (fullList.size() == 0 || alwaysCreateDirectTranslationOption)
ProcessUnknownWord(pos);
}
}
/** special handling of ONE unknown words. Either add temporarily add word to translation table,
* or drop the translation.
* This function should be called by the ProcessOneUnknownWord() in the inherited class
* At the moment, this unknown word handler is a bit of a hack, if copies over each factor from source
* to target word, or uses the 'UNK' factor.
* Ideally, this function should be in a class which can be expanded upon, for example,
* to create a morphologically aware handler.
*
* \param sourceWord the unknown word
* \param sourcePos
* \param length length covered by this word (may be > 1 for lattice input)
* \param inputScores a set of scores associated with unknown word (input scores from latties/CNs)
*/
void TranslationOptionCollection::ProcessOneUnknownWord(const Word &sourceWord,size_t sourcePos, size_t length, const Scores *inputScores)
{
const StaticData &staticData = StaticData::Instance();
const UnknownWordPenaltyProducer *unknownWordPenaltyProducer = staticData.GetUnknownWordPenaltyProducer();
float unknownScore = FloorScore(TransformScore(0));
// unknown word, add as trans opt
FactorCollection &factorCollection = FactorCollection::Instance();
size_t isDigit = 0;
const Factor *f = sourceWord[0]; // TODO hack. shouldn't know which factor is surface
const StringPiece s = f->GetString();
bool isEpsilon = (s=="" || s==EPSILON);
if (StaticData::Instance().GetDropUnknown()) {
isDigit = s.find_first_of("0123456789");
if (isDigit == 1)
isDigit = 1;
else
isDigit = 0;
// modify the starting bitmap
}
Phrase* m_unksrc = new Phrase(1);
m_unksrc->AddWord() = sourceWord;
m_unksrcs.push_back(m_unksrc);
TranslationOption *transOpt;
TargetPhrase targetPhrase;
targetPhrase.SetSourcePhrase(*m_unksrc);
if (!(staticData.GetDropUnknown() || isEpsilon) || isDigit) {
// add to dictionary
Word &targetWord = targetPhrase.AddWord();
for (unsigned int currFactor = 0 ; currFactor < MAX_NUM_FACTORS ; currFactor++) {
FactorType factorType = static_cast<FactorType>(currFactor);
const Factor *sourceFactor = sourceWord[currFactor];
if (sourceFactor == NULL)
targetWord[factorType] = factorCollection.AddFactor(UNKNOWN_FACTOR);
else
targetWord[factorType] = factorCollection.AddFactor(sourceFactor->GetString());
}
//create a one-to-one alignment between UNKNOWN_FACTOR and its verbatim translation
targetPhrase.SetAlignmentInfo("0-0");
} else {
// drop source word. create blank trans opt
//targetPhrase.SetAlignment();
}
targetPhrase.GetScoreBreakdown().Assign(unknownWordPenaltyProducer, unknownScore);
if (inputScores != NULL) {
targetPhrase.SetInputScore(*inputScores);
}
targetPhrase.Evaluate(*m_unksrc);
transOpt = new TranslationOption(WordsRange(sourcePos, sourcePos + length - 1), targetPhrase);
Add(transOpt);
}
/** compute future score matrix in a dynamic programming fashion.
* This matrix used in search.
* Call this function once translation option collection has been filled with translation options
*/
void TranslationOptionCollection::CalcFutureScore()
{
// setup the matrix (ignore lower triangle, set upper triangle to -inf
size_t size = m_source.GetSize(); // the width of the matrix
for(size_t row=0; row<size; row++) {
for(size_t col=row; col<size; col++) {
m_futureScore.SetScore(row, col, -numeric_limits<float>::infinity());
}
}
// walk all the translation options and record the cheapest option for each span
for (size_t startPos = 0 ; startPos < size ; ++startPos) {
size_t maxSize = m_source.GetSize() - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for (size_t endPos = startPos ; endPos < startPos + maxSize ; ++endPos) {
TranslationOptionList &transOptList = GetTranslationOptionList(startPos, endPos);
TranslationOptionList::const_iterator iterTransOpt;
for(iterTransOpt = transOptList.begin() ; iterTransOpt != transOptList.end() ; ++iterTransOpt) {
const TranslationOption &transOpt = **iterTransOpt;
float score = transOpt.GetFutureScore();
if (score > m_futureScore.GetScore(startPos, endPos))
m_futureScore.SetScore(startPos, endPos, score);
}
}
}
// now fill all the cells in the strictly upper triangle
// there is no way to modify the diagonal now, in the case
// where no translation option covers a single-word span,
// we leave the +inf in the matrix
// like in chart parsing we want each cell to contain the highest score
// of the full-span trOpt or the sum of scores of joining two smaller spans
for(size_t colstart = 1; colstart < size ; colstart++) {
for(size_t diagshift = 0; diagshift < size-colstart ; diagshift++) {
size_t startPos = diagshift;
size_t endPos = colstart+diagshift;
for(size_t joinAt = startPos; joinAt < endPos ; joinAt++) {
float joinedScore = m_futureScore.GetScore(startPos, joinAt)
+ m_futureScore.GetScore(joinAt+1, endPos);
/* // uncomment to see the cell filling scheme
TRACE_ERR( "[" <<startPos<<","<<endPos<<"] <-? ["<<startPos<<","<<joinAt<<"]+["<<joinAt+1<<","<<endPos
<< "] (colstart: "<<colstart<<", diagshift: "<<diagshift<<")"<<endl);
*/
if (joinedScore > m_futureScore.GetScore(startPos, endPos))
m_futureScore.SetScore(startPos, endPos, joinedScore);
}
}
}
IFVERBOSE(3) {
int total = 0;
for(size_t row=0; row<size; row++) {
size_t maxSize = size - row;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for(size_t col=row; col<row+maxSize; col++) {
int count = GetTranslationOptionList(row, col).size();
TRACE_ERR( "translation options spanning from "
<< row <<" to "<< col <<" is "
<< count <<endl);
total += count;
}
}
TRACE_ERR( "translation options generated in total: "<< total << endl);
for(size_t row=0; row<size; row++)
for(size_t col=row; col<size; col++)
TRACE_ERR( "future cost from "<< row <<" to "<< col <<" is "<< m_futureScore.GetScore(row, col) <<endl);
}
}
/** Create all possible translations from the phrase tables
* for a particular input sentence. This implies applying all
* translation and generation steps. Also computes future cost matrix.
*/
void TranslationOptionCollection::CreateTranslationOptions()
{
// loop over all substrings of the source sentence, look them up
// in the phraseDictionary (which is the- possibly filtered-- phrase
// table loaded on initialization), generate TranslationOption objects
// for all phrases
const StaticData &staticData = StaticData::Instance();
// there may be multiple decoding graphs (factorizations of decoding)
const vector <DecodeGraph*> &decodeGraphList = StaticData::Instance().GetDecodeGraphs();
const vector <size_t> &decodeGraphBackoff = StaticData::Instance().GetDecodeGraphBackoff();
// length of the sentence
const size_t size = m_source.GetSize();
// loop over all decoding graphs, each generates translation options
for (size_t graphInd = 0 ; graphInd < decodeGraphList.size() ; graphInd++) {
if (staticData.IsDecodingGraphIgnored( graphInd )) {
std::cerr << "ignoring decoding path " << graphInd << std::endl;
continue;
}
if (decodeGraphList.size() > 1) {
VERBOSE(3,"Creating translation options from decoding graph " << graphInd << endl);
}
const DecodeGraph &decodeGraph = *decodeGraphList[graphInd];
// generate phrases that start at startPos ...
for (size_t startPos = 0 ; startPos < size; startPos++) {
size_t maxSize = size - startPos; // don't go over end of sentence
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
// ... and that end at endPos
for (size_t endPos = startPos ; endPos < startPos + maxSize ; endPos++) {
if (graphInd > 0 && // only skip subsequent graphs
decodeGraphBackoff[graphInd] != 0 && // limited use of backoff specified
(endPos-startPos+1 > decodeGraphBackoff[graphInd] || // size exceeds backoff limit or ...
m_collection[startPos][endPos-startPos].size() > 0)) { // already covered
VERBOSE(3,"No backoff to graph " << graphInd << " for span [" << startPos << ";" << endPos << "]");
VERBOSE(3,", length limit: " << decodeGraphBackoff[graphInd]);
VERBOSE(3,", found so far: " << m_collection[startPos][endPos-startPos].size() << endl);
// do not create more options
continue;
}
// create translation options for that range
CreateTranslationOptionsForRange( decodeGraph, startPos, endPos, true, graphInd);
}
}
}
VERBOSE(2,"Translation Option Collection\n " << *this << endl);
ProcessUnknownWord();
EvaluateWithSource();
// Prune
Prune();
Sort();
// future score matrix
CalcFutureScore();
// Cached lex reodering costs
CacheLexReordering();
}
void TranslationOptionCollection::EvaluateWithSource()
{
const size_t size = m_source.GetSize();
for (size_t startPos = 0 ; startPos < size ; ++startPos) {
size_t maxSize = m_source.GetSize() - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for (size_t endPos = startPos ; endPos < startPos + maxSize ; ++endPos) {
TranslationOptionList &transOptList = GetTranslationOptionList(startPos, endPos);
TranslationOptionList::const_iterator iterTransOpt;
for(iterTransOpt = transOptList.begin() ; iterTransOpt != transOptList.end() ; ++iterTransOpt) {
TranslationOption &transOpt = **iterTransOpt;
transOpt.Evaluate(m_source);
}
}
}
}
void TranslationOptionCollection::Sort()
{
size_t size = m_source.GetSize();
for (size_t startPos = 0 ; startPos < size; ++startPos) {
size_t maxSize = size - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for (size_t endPos = startPos ; endPos < startPos + maxSize; ++endPos) {
TranslationOptionList &transOptList = GetTranslationOptionList(startPos, endPos);
std::sort(transOptList.begin(), transOptList.end(), CompareTranslationOption);
}
}
}
/** create translation options that exactly cover a specific input span.
* Called by CreateTranslationOptions() and ProcessUnknownWord()
* \param decodeGraph list of decoding steps
* \param factorCollection input sentence with all factors
* \param startPos first position in input sentence
* \param lastPos last position in input sentence
* \param adhereTableLimit whether phrase & generation table limits are adhered to
*/
void TranslationOptionCollection::CreateTranslationOptionsForRange(
const DecodeGraph &decodeGraph
, size_t startPos
, size_t endPos
, bool adhereTableLimit
, size_t graphInd)
{
if ((StaticData::Instance().GetXmlInputType() != XmlExclusive) || !HasXmlOptionsOverlappingRange(startPos,endPos)) {
Phrase *sourcePhrase = NULL; // can't initialise with substring, in case it's confusion network
// consult persistent (cross-sentence) cache for stored translation options
bool skipTransOptCreation = false
, useCache = StaticData::Instance().GetUseTransOptCache();
if (useCache) {
const WordsRange wordsRange(startPos, endPos);
sourcePhrase = new Phrase(m_source.GetSubString(wordsRange));
const TranslationOptionList *transOptList = StaticData::Instance().FindTransOptListInCache(decodeGraph, *sourcePhrase);
// is phrase in cache?
if (transOptList != NULL) {
skipTransOptCreation = true;
TranslationOptionList::const_iterator iterTransOpt;
for (iterTransOpt = transOptList->begin() ; iterTransOpt != transOptList->end() ; ++iterTransOpt) {
TranslationOption *transOpt = new TranslationOption(**iterTransOpt, wordsRange);
Add(transOpt);
}
}
} // useCache
if (!skipTransOptCreation) {
// partial trans opt stored in here
PartialTranslOptColl* oldPtoc = new PartialTranslOptColl;
size_t totalEarlyPruned = 0;
// initial translation step
list <const DecodeStep* >::const_iterator iterStep = decodeGraph.begin();
const DecodeStep &decodeStep = **iterStep;
static_cast<const DecodeStepTranslation&>(decodeStep).ProcessInitialTranslation
(m_source, *oldPtoc
, startPos, endPos, adhereTableLimit );
// do rest of decode steps
int indexStep = 1;
for (++iterStep; iterStep != decodeGraph.end() ; ++iterStep, ++indexStep) {
const DecodeStep &decodeStep = **iterStep;
PartialTranslOptColl* newPtoc = new PartialTranslOptColl;
// go thru each intermediate trans opt just created
const vector<TranslationOption*>& partTransOptList = oldPtoc->GetList();
vector<TranslationOption*>::const_iterator iterPartialTranslOpt;
for (iterPartialTranslOpt = partTransOptList.begin() ; iterPartialTranslOpt != partTransOptList.end() ; ++iterPartialTranslOpt) {
TranslationOption &inputPartialTranslOpt = **iterPartialTranslOpt;
decodeStep.Process(inputPartialTranslOpt
, decodeStep
, *newPtoc
, this
, adhereTableLimit
, *sourcePhrase);
}
// last but 1 partial trans not required anymore
totalEarlyPruned += newPtoc->GetPrunedCount();
delete oldPtoc;
oldPtoc = newPtoc;
} // for (++iterStep
// add to fully formed translation option list
PartialTranslOptColl &lastPartialTranslOptColl = *oldPtoc;
const vector<TranslationOption*>& partTransOptList = lastPartialTranslOptColl.GetList();
vector<TranslationOption*>::const_iterator iterColl;
for (iterColl = partTransOptList.begin() ; iterColl != partTransOptList.end() ; ++iterColl) {
TranslationOption *transOpt = *iterColl;
Add(transOpt);
}
// storing translation options in persistent cache (kept across sentences)
if (useCache) {
if (partTransOptList.size() > 0) {
TranslationOptionList &transOptList = GetTranslationOptionList(startPos, endPos);
StaticData::Instance().AddTransOptListToCache(decodeGraph, *sourcePhrase, transOptList);
}
}
lastPartialTranslOptColl.DetachAll();
totalEarlyPruned += oldPtoc->GetPrunedCount();
delete oldPtoc;
// TRACE_ERR( "Early translation options pruned: " << totalEarlyPruned << endl);
} // if (!skipTransOptCreation)
if (useCache)
delete sourcePhrase;
} // if ((StaticData::Instance().GetXmlInputType() != XmlExclusive) || !HasXmlOptionsOverlappingRange(startPos,endPos))
if (graphInd == 0 && StaticData::Instance().GetXmlInputType() != XmlPassThrough && HasXmlOptionsOverlappingRange(startPos,endPos)) {
CreateXmlOptionsForRange(startPos, endPos);
}
}
/** Check if this range overlaps with any XML options. This doesn't need to be an exact match, only an overlap.
* by default, we don't support XML options. subclasses need to override this function.
* called by CreateTranslationOptionsForRange()
* \param startPos first position in input sentence
* \param lastPos last position in input sentence
* \param adhereTableLimit whether phrase & generation table limits are adhered to
*/
bool TranslationOptionCollection::HasXmlOptionsOverlappingRange(size_t, size_t) const
{
return false;
//not implemented for base class
}
/** Populates the current Collection with XML options exactly covering the range specified. Default implementation does nothing.
* called by CreateTranslationOptionsForRange()
* \param startPos first position in input sentence
* \param lastPos last position in input sentence
*/
void TranslationOptionCollection::CreateXmlOptionsForRange(size_t, size_t)
{
//not implemented for base class
};
/** add translation option to the list
* \param translationOption translation option to be added */
void TranslationOptionCollection::Add(TranslationOption *translationOption)
{
const WordsRange &coverage = translationOption->GetSourceWordsRange();
CHECK(coverage.GetEndPos() - coverage.GetStartPos() < m_collection[coverage.GetStartPos()].size());
m_collection[coverage.GetStartPos()][coverage.GetEndPos() - coverage.GetStartPos()].Add(translationOption);
}
TO_STRING_BODY(TranslationOptionCollection);
std::ostream& operator<<(std::ostream& out, const TranslationOptionCollection& coll)
{
size_t size = coll.GetSize();
for (size_t startPos = 0 ; startPos < size ; ++startPos) {
size_t maxSize = size - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for (size_t endPos = startPos ; endPos < startPos + maxSize ; ++endPos) {
const TranslationOptionList& fullList = coll.GetTranslationOptionList(startPos, endPos);
size_t sizeFull = fullList.size();
for (size_t i = 0; i < sizeFull; i++) {
out << *fullList.Get(i) << std::endl;
}
}
}
//std::vector< std::vector< TranslationOptionList > >::const_iterator i = coll.m_collection.begin();
//size_t j = 0;
//for (; i!=coll.m_collection.end(); ++i) {
//out << "s[" << j++ << "].size=" << i->size() << std::endl;
//}
return out;
}
const std::vector<Phrase*>& TranslationOptionCollection::GetUnknownSources() const
{
return m_unksrcs;
}
void TranslationOptionCollection::CacheLexReordering()
{
size_t size = m_source.GetSize();
const std::vector<const StatefulFeatureFunction*> &ffs = StatefulFeatureFunction::GetStatefulFeatureFunctions();
std::vector<const StatefulFeatureFunction*>::const_iterator iter;
for (iter = ffs.begin(); iter != ffs.end(); ++iter) {
const StatefulFeatureFunction &ff = **iter;
if (typeid(ff) == typeid(LexicalReordering)) {
const LexicalReordering &lexreordering = static_cast<const LexicalReordering&>(ff);
for (size_t startPos = 0 ; startPos < size ; startPos++) {
size_t maxSize = size - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
for (size_t endPos = startPos ; endPos < startPos + maxSize; endPos++) {
TranslationOptionList &transOptList = GetTranslationOptionList( startPos, endPos);
TranslationOptionList::iterator iterTransOpt;
for(iterTransOpt = transOptList.begin() ; iterTransOpt != transOptList.end() ; ++iterTransOpt) {
TranslationOption &transOpt = **iterTransOpt;
//Phrase sourcePhrase = m_source.GetSubString(WordsRange(startPos,endPos));
const Phrase *sourcePhrase = transOpt.GetSourcePhrase();
if (sourcePhrase) {
Scores score = lexreordering.GetProb(*sourcePhrase
, transOpt.GetTargetPhrase());
if (!score.empty())
transOpt.CacheLexReorderingScores(lexreordering, score);
} // if (sourcePhrase) {
} // for(iterTransOpt
} // for (size_t endPos = startPos ; endPos < startPos + maxSize; endPos++) {
} // for (size_t startPos = 0 ; startPos < size ; startPos++) {
} // if (typeid(ff) == typeid(LexicalReordering)) {
} // for (iter = ffs.begin(); iter != ffs.end(); ++iter) {
}
//! list of trans opt for a particular span
TranslationOptionList &TranslationOptionCollection::GetTranslationOptionList(size_t startPos, size_t endPos)
{
size_t maxSize = endPos - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
CHECK(maxSize < m_collection[startPos].size());
return m_collection[startPos][maxSize];
}
const TranslationOptionList &TranslationOptionCollection::GetTranslationOptionList(size_t startPos, size_t endPos) const
{
size_t maxSize = endPos - startPos;
size_t maxSizePhrase = StaticData::Instance().GetMaxPhraseLength();
maxSize = std::min(maxSize, maxSizePhrase);
CHECK(maxSize < m_collection[startPos].size());
return m_collection[startPos][maxSize];
}
}
|