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
|
// $Id$
#include <cassert>
#include "StaticData.h"
#include "InputFileStream.h"
#include "LexicalDistortionCost.h"
#include "LexicalReorderingTable.h"
#include "WordsRange.h"
#include "TranslationOption.h"
/******** LexicalDistortionCost: static member functions ************/
LexicalDistortionCost *LexicalDistortionCost::CreateModel(const std::string &modelType,
const std::string &filePath,
Direction direction,
std::vector< FactorType >& f_factors,
std::vector< FactorType >& e_factors)
{
if(modelType == "beta-binomial-phrase")
return new LDCBetaBinomial(filePath, direction, PhrasePair, f_factors, e_factors);
else if(modelType == "beta-geometric-phrase")
return new LDCBetaGeometric(filePath, direction, PhrasePair, f_factors, e_factors);
else {
UserMessage::Add("Lexical distortion model type not implemented: " + modelType);
return NULL;
}
}
/******** LexicalDistortionCost ************/
LexicalDistortionCost::LexicalDistortionCost(const std::string &filePath,
Direction direction,
Condition condition,
std::vector< FactorType >& f_factors,
std::vector< FactorType >& e_factors,
size_t numParametersPerDirection) :
m_modelFileName(filePath),
m_direction(direction), m_condition(condition),
m_srcfactors(f_factors), m_tgtfactors(e_factors),
m_numParametersPerDirection(numParametersPerDirection)
{
m_distortionTable = LexicalReorderingTable::LoadAvailable(filePath, f_factors, e_factors, std::vector<FactorType>(), false);
}
LexicalDistortionCost::~LexicalDistortionCost()
{
delete m_distortionTable;
}
const std::vector<float> LexicalDistortionCost::GetDistortionParameters(const Phrase &src, const Phrase &tgt) const
{
std::vector<float> params = m_distortionTable->GetScore(src, tgt, Phrase(Output));
// if we can't find find the phrase, return zeroes - the prior will be used
if(params.size() == 0)
return std::vector<float>(GetNumParameterSets() * GetNumParametersPerDirection(), .0f);
assert(params.size() == GetNumParameterSets() * GetNumParametersPerDirection());
return params;
}
std::vector<float> LexicalDistortionCost::CalcScore(Hypothesis *h) const
{
std::vector<float> scores;
if(m_direction == Forward || m_direction == Bidirectional) {
std::vector<float> params_cur = h->GetTranslationOption().GetDistortionParamsForModel(this);
assert(params_cur.size() == GetNumParameterSets() * GetNumParametersPerDirection());
std::vector<float> params_pre(params_cur.begin(), params_cur.begin() + GetNumParametersPerDirection());
scores.push_back(CalculateDistortionScore(
h->GetPrevHypo()->GetCurrSourceWordsRange(),
h->GetCurrSourceWordsRange(), ¶ms_pre));
}
if(m_direction == Backward || m_direction == Bidirectional) {
if(h->GetPrevHypo()->GetPrevHypo() == NULL)
scores.push_back(.0f);
else {
std::vector<float> params_prev = h->GetPrevHypo()->GetTranslationOption().GetDistortionParamsForModel(this);
assert(params_prev.size() == GetNumParameterSets() * GetNumParametersPerDirection());
std::vector<float> params_post(params_prev.begin() + GetNumParametersPerDirection(), params_prev.end());
scores.push_back(CalculateDistortionScore(
h->GetPrevHypo()->GetCurrSourceWordsRange(),
h->GetCurrSourceWordsRange(), ¶ms_post));
}
}
return scores;
}
std::vector<float> LexicalDistortionCost::GetProb(const Phrase &src, const Phrase &tgt) const {
assert(m_condition == PhrasePair); // everything else isn't implemented right now
return GetDistortionParameters(src, tgt);
}
/******** LDCBetaBinomial ************/
float LDCBetaBinomial::CalculateDistortionScore(const WordsRange &prev, const WordsRange &curr,
const std::vector<float> *params) const
{
const std::vector<float> &prior = GetPrior();
float p = params->at(0) + prior[0];
float q = params->at(1) + prior[1];
int x = StaticData::Instance().GetInput()->ComputeDistortionDistance(prev, curr);
if(x < -m_distortionRange) x = -m_distortionRange;
if(x > m_distortionRange) x = m_distortionRange;
x += m_distortionRange;
return beta_binomial(p, q, x);
}
float LDCBetaBinomial::beta_binomial(float p, float q, int x) const
{
const long double n = (long double) 2 * m_distortionRange;
long double n1 = lgammal(n + 1);
long double n2 = lgammal(q + n - x);
long double n3 = lgammal(x + p);
long double n4 = lgammal(p + q);
long double d1 = lgammal(x + 1);
long double d2 = lgammal(n - x + 1);
long double d3 = lgammal(p + q + n);
long double d4 = lgammal(p);
long double d5 = lgammal(q);
// std::cerr << n1 << " + " << n2 << " + " << n3 << " + " << n4 << std::endl;
// std::cerr << d1 << " + " << d2 << " + " << d3 << " + " << d4 << " + " << d5 << std::endl;
long double part1 = n1 + n4 - d3 - d4 - d5;
// std::cerr << part1 << std::endl;
long double part2 = n2 + n3 - d1 - d2;
// std::cerr << part2 << std::endl;
long double score = part1 + part2;
// std::cerr << "x: " << x << " score: " << score << std::endl;
if(!finite(score)) {
std::cerr << p << " ; " << q << std::endl;
std::cerr << n1 << " + " << n2 << " + " << n3 << " + " << n4 << std::endl;
std::cerr << d1 << " + " << d2 << " + " << d3 << " + " << d4 << " + " << d5 << std::endl;
std::cerr << part1 << std::endl;
std::cerr << part2 << std::endl;
std::cerr << "x: " << x << " score: " << score << std::endl;
assert(false);
}
return FloorScore((float) score);
}
std::vector<float> LDCBetaBinomial::GetDefaultPrior() const
{
return std::vector<float>(4, 12);
}
/******** LDCBetaGeometric ************/
float LDCBetaGeometric::CalculateDistortionScore(const WordsRange &prev, const WordsRange &curr,
const std::vector<float> *params) const
{
const std::vector<float> &prior = GetPrior();
float p_dir = params->at(0) + prior[0];
float q_dir = params->at(1) + prior[1];
float p_pos = params->at(2) + prior[2];
float q_pos = params->at(3) + prior[3];
float p_neg = params->at(4) + prior[4];
float q_neg = params->at(5) + prior[5];
int distance = StaticData::Instance().GetInput()->ComputeDistortionDistance(prev, curr);
float score = .0f;
if(distance >= 0) {
score += log(p_dir / (p_dir + q_dir));
int x = distance;
score += beta_geometric(p_pos, q_pos, x);
} else {
score += log(1.0f - p_dir / (p_dir + q_dir));
int x = -distance - 1;
score += beta_geometric(p_neg, q_neg, x);
}
assert(finite(score));
return FloorScore(score);
}
float LDCBetaGeometric::beta_geometric(float p, float q, int x) const
{
assert(x >= 0);
float result = log(p);
for(int i = 0; i <= x - 1; i++)
result += log(q + i);
for(int i = 0; i <= x; i++)
result -= log(p + q + i);
return result;
}
std::vector<float> LDCBetaGeometric::GetDefaultPrior() const
{
return std::vector<float>(12, 2);
}
|
