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
|
// $Id$
#include <list>
#include "TranslationOptionCollectionLattice.h"
#include "ConfusionNet.h"
#include "WordLattice.h"
#include "DecodeGraph.h"
#include "DecodeStepTranslation.h"
#include "DecodeStepGeneration.h"
#include "FactorCollection.h"
#include "FF/InputFeature.h"
#include "TranslationModel/PhraseDictionaryTreeAdaptor.h"
#include "util/exception.hh"
using namespace std;
namespace Moses
{
/** constructor; just initialize the base class */
TranslationOptionCollectionLattice
::TranslationOptionCollectionLattice
( ttasksptr const& ttask, const WordLattice &input,
size_t maxNoTransOptPerCoverage, float translationOptionThreshold)
: TranslationOptionCollection(ttask, input, maxNoTransOptPerCoverage,
translationOptionThreshold)
{
UTIL_THROW_IF2(StaticData::Instance().GetUseLegacyPT(),
"Not for models using the legqacy binary phrase table");
const InputFeature *inputFeature = InputFeature::InstancePtr();
UTIL_THROW_IF2(inputFeature == NULL, "Input feature must be specified");
size_t maxPhraseLength = StaticData::Instance().GetMaxPhraseLength();
size_t size = input.GetSize();
// 1-word phrases
for (size_t startPos = 0; startPos < size; ++startPos) {
const std::vector<size_t> &nextNodes = input.GetNextNodes(startPos);
const ConfusionNet::Column &col = input.GetColumn(startPos);
for (size_t i = 0; i < col.size(); ++i) {
const Word &word = col[i].first;
UTIL_THROW_IF2(word.IsEpsilon(), "Epsilon not supported");
size_t nextNode = nextNodes[i];
size_t endPos = startPos + nextNode - 1;
WordsRange range(startPos, endPos);
if (range.GetNumWordsCovered() > maxPhraseLength) {
continue;
}
const NonTerminalSet &labels = input.GetLabelSet(startPos, endPos);
Phrase subphrase;
subphrase.AddWord(word);
const ScorePair &scores = col[i].second;
ScorePair *inputScore = new ScorePair(scores);
InputPath *path
= new InputPath(ttask, subphrase, labels, range, NULL, inputScore);
path->SetNextNode(nextNode);
m_inputPathQueue.push_back(path);
// recursive
Extend(*path, input);
}
}
}
void TranslationOptionCollectionLattice::Extend(const InputPath &prevPath, const WordLattice &input)
{
size_t nextPos = prevPath.GetWordsRange().GetEndPos() + 1;
if (nextPos >= input.GetSize()) {
return;
}
size_t startPos = prevPath.GetWordsRange().GetStartPos();
const Phrase &prevPhrase = prevPath.GetPhrase();
const ScorePair *prevInputScore = prevPath.GetInputScore();
UTIL_THROW_IF2(prevInputScore == NULL,
"Null previous score");
const std::vector<size_t> &nextNodes = input.GetNextNodes(nextPos);
const ConfusionNet::Column &col = input.GetColumn(nextPos);
for (size_t i = 0; i < col.size(); ++i) {
const Word &word = col[i].first;
UTIL_THROW_IF2(word.IsEpsilon(), "Epsilon not supported");
size_t nextNode = nextNodes[i];
size_t endPos = nextPos + nextNode - 1;
WordsRange range(startPos, endPos);
size_t maxPhraseLength = StaticData::Instance().GetMaxPhraseLength();
if (range.GetNumWordsCovered() > maxPhraseLength) {
continue;
}
const NonTerminalSet &labels = input.GetLabelSet(startPos, endPos);
Phrase subphrase(prevPhrase);
subphrase.AddWord(word);
const ScorePair &scores = col[i].second;
ScorePair *inputScore = new ScorePair(*prevInputScore);
inputScore->PlusEquals(scores);
InputPath *path = new InputPath(prevPath.ttask, subphrase, labels,
range, &prevPath, inputScore);
path->SetNextNode(nextNode);
m_inputPathQueue.push_back(path);
// recursive
Extend(*path, input);
}
}
void TranslationOptionCollectionLattice::CreateTranslationOptions()
{
GetTargetPhraseCollectionBatch();
VERBOSE(2,"Translation Option Collection\n " << *this << endl);
const vector <DecodeGraph*> &decodeGraphs = StaticData::Instance().GetDecodeGraphs();
UTIL_THROW_IF2(decodeGraphs.size() != 1, "Multiple decoder graphs not supported yet");
const DecodeGraph &decodeGraph = *decodeGraphs[0];
UTIL_THROW_IF2(decodeGraph.GetSize() != 1, "Factored decomposition not supported yet");
const DecodeStep &decodeStep = **decodeGraph.begin();
const PhraseDictionary &phraseDictionary = *decodeStep.GetPhraseDictionaryFeature();
for (size_t i = 0; i < m_inputPathQueue.size(); ++i) {
const InputPath &path = *m_inputPathQueue[i];
TargetPhraseCollection::shared_ptr tpColl
= path.GetTargetPhrases(phraseDictionary);
const WordsRange &range = path.GetWordsRange();
if (tpColl && tpColl->GetSize()) {
TargetPhraseCollection::const_iterator iter;
for (iter = tpColl->begin(); iter != tpColl->end(); ++iter) {
const TargetPhrase &tp = **iter;
TranslationOption *transOpt = new TranslationOption(range, tp);
transOpt->SetInputPath(path);
transOpt->EvaluateWithSourceContext(m_source);
Add(transOpt);
}
} else if (path.GetPhrase().GetSize() == 1) {
// unknown word processing
ProcessOneUnknownWord(path, path.GetWordsRange().GetStartPos(), path.GetWordsRange().GetNumWordsCovered() , path.GetInputScore());
}
}
// Prune
Prune();
Sort();
// future score matrix
CalcFutureScore();
// Cached lex reodering costs
CacheLexReordering();
}
void
TranslationOptionCollectionLattice::
ProcessUnknownWord(size_t sourcePos)
{
UTIL_THROW(util::Exception, "ProcessUnknownWord() not implemented for lattice");
// why??? UG
}
bool
TranslationOptionCollectionLattice::
CreateTranslationOptionsForRange
(const DecodeGraph &decodeStepList, size_t startPosition, size_t endPosition,
bool adhereTableLimit, size_t graphInd)
{
UTIL_THROW(util::Exception,
"CreateTranslationOptionsForRange() not implemented for lattice");
}
} // namespace
|