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
|
/***********************************************************************
Moses - factored phrase-based language decoder
Copyright (C) 2009 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 <cstdio>
#include <iostream>
#include <fstream>
#include <vector>
#include <string>
#include <cstdlib>
#include <cstring>
#include "tables-core.h"
#include "SafeGetline.h"
#include "InputFileStream.h"
#include "OutputFileStream.h"
#define LINE_MAX_LENGTH 10000
using namespace std;
bool hierarchicalFlag = false;
bool onlyDirectFlag = false;
bool phraseCountFlag = true;
bool lowCountFlag = false;
bool goodTuringFlag = false;
bool kneserNeyFlag = false;
bool logProbFlag = false;
bool outputNTLengths = false;
inline float maybeLogProb( float a ) { return logProbFlag ? log(a) : a; }
char line[LINE_MAX_LENGTH];
void processFiles( char*, char*, char*, char* );
void loadCountOfCounts( char* );
bool getLine( istream &fileP, vector< string > &item );
vector< string > splitLine();
int main(int argc, char* argv[])
{
cerr << "Consolidate v2.0 written by Philipp Koehn\n"
<< "consolidating direct and indirect rule tables\n";
if (argc < 4) {
cerr << "syntax: consolidate phrase-table.direct phrase-table.indirect phrase-table.consolidated [--Hierarchical] [--OnlyDirect] [--OutputNTLengths] \n";
exit(1);
}
char* &fileNameDirect = argv[1];
char* &fileNameIndirect = argv[2];
char* &fileNameConsolidated = argv[3];
char* fileNameCountOfCounts;
for(int i=4; i<argc; i++) {
if (strcmp(argv[i],"--Hierarchical") == 0) {
hierarchicalFlag = true;
cerr << "processing hierarchical rules\n";
} else if (strcmp(argv[i],"--OnlyDirect") == 0) {
onlyDirectFlag = true;
cerr << "only including direct translation scores p(e|f)\n";
} else if (strcmp(argv[i],"--NoPhraseCount") == 0) {
phraseCountFlag = false;
cerr << "not including the phrase count feature\n";
} else if (strcmp(argv[i],"--GoodTuring") == 0) {
goodTuringFlag = true;
if (i+1==argc) {
cerr << "ERROR: specify count of count files for Good Turing discounting!\n";
exit(1);
}
fileNameCountOfCounts = argv[++i];
cerr << "adjusting phrase translation probabilities with Good Turing discounting\n";
} else if (strcmp(argv[i],"--KneserNey") == 0) {
kneserNeyFlag = true;
if (i+1==argc) {
cerr << "ERROR: specify count of count files for Kneser Ney discounting!\n";
exit(1);
}
fileNameCountOfCounts = argv[++i];
cerr << "adjusting phrase translation probabilities with Kneser Ney discounting\n";
} else if (strcmp(argv[i],"--LowCountFeature") == 0) {
lowCountFlag = true;
cerr << "including the low count feature\n";
} else if (strcmp(argv[i],"--LogProb") == 0) {
logProbFlag = true;
cerr << "using log-probabilities\n";
} else if (strcmp(argv[i],"--OutputNTLengths") == 0) {
outputNTLengths = true;
} else {
cerr << "ERROR: unknown option " << argv[i] << endl;
exit(1);
}
}
processFiles( fileNameDirect, fileNameIndirect, fileNameConsolidated, fileNameCountOfCounts );
}
vector< float > countOfCounts;
vector< float > goodTuringDiscount;
float kneserNey_D1, kneserNey_D2, kneserNey_D3, totalCount = -1;
void loadCountOfCounts( char* fileNameCountOfCounts )
{
Moses::InputFileStream fileCountOfCounts(fileNameCountOfCounts);
if (fileCountOfCounts.fail()) {
cerr << "ERROR: could not open count of counts file " << fileNameCountOfCounts << endl;
exit(1);
}
istream &fileP = fileCountOfCounts;
countOfCounts.push_back(0.0);
while(1) {
if (fileP.eof()) break;
SAFE_GETLINE((fileP), line, LINE_MAX_LENGTH, '\n', __FILE__);
if (fileP.eof()) break;
if (totalCount < 0)
totalCount = atof(line); // total number of distinct phrase pairs
else
countOfCounts.push_back( atof(line) );
}
fileCountOfCounts.Close();
// compute Good Turing discounts
if (goodTuringFlag) {
goodTuringDiscount.push_back(0.01); // floor value
for( size_t i=1; i<countOfCounts.size()-1; i++ ) {
goodTuringDiscount.push_back(((float)i+1)/(float)i*((countOfCounts[i+1]+0.1) / ((float)countOfCounts[i]+0.1)));
if (goodTuringDiscount[i]>1)
goodTuringDiscount[i] = 1;
if (goodTuringDiscount[i]<goodTuringDiscount[i-1])
goodTuringDiscount[i] = goodTuringDiscount[i-1];
}
}
// compute Kneser Ney co-efficients [Chen&Goodman, 1998]
float Y = countOfCounts[1] / (countOfCounts[1] + 2*countOfCounts[2]);
kneserNey_D1 = 1 - 2*Y * countOfCounts[2] / countOfCounts[1];
kneserNey_D2 = 2 - 3*Y * countOfCounts[3] / countOfCounts[2];
kneserNey_D3 = 3 - 4*Y * countOfCounts[4] / countOfCounts[3];
// sanity constraints
if (kneserNey_D1 > 0.9) kneserNey_D1 = 0.9;
if (kneserNey_D2 > 1.9) kneserNey_D2 = 1.9;
if (kneserNey_D3 > 2.9) kneserNey_D3 = 2.9;
}
void processFiles( char* fileNameDirect, char* fileNameIndirect, char* fileNameConsolidated, char* fileNameCountOfCounts )
{
if (goodTuringFlag || kneserNeyFlag)
loadCountOfCounts( fileNameCountOfCounts );
// open input files
Moses::InputFileStream fileDirect(fileNameDirect);
Moses::InputFileStream fileIndirect(fileNameIndirect);
if (fileDirect.fail()) {
cerr << "ERROR: could not open phrase table file " << fileNameDirect << endl;
exit(1);
}
istream &fileDirectP = fileDirect;
if (fileIndirect.fail()) {
cerr << "ERROR: could not open phrase table file " << fileNameIndirect << endl;
exit(1);
}
istream &fileIndirectP = fileIndirect;
// open output file: consolidated phrase table
Moses::OutputFileStream fileConsolidated;
bool success = fileConsolidated.Open(fileNameConsolidated);
if (!success) {
cerr << "ERROR: could not open output file " << fileNameConsolidated << endl;
exit(1);
}
// loop through all extracted phrase translations
int i=0;
while(true) {
i++;
if (i%100000 == 0) cerr << "." << flush;
vector< string > itemDirect, itemIndirect;
if (! getLine(fileIndirectP,itemIndirect) ||
! getLine(fileDirectP, itemDirect ))
break;
// direct: target source alignment probabilities
// indirect: source target probabilities
// consistency checks
if (itemDirect[0].compare( itemIndirect[0] ) != 0) {
cerr << "ERROR: target phrase does not match in line " << i << ": '"
<< itemDirect[0] << "' != '" << itemIndirect[0] << "'" << endl;
exit(1);
}
if (itemDirect[1].compare( itemIndirect[1] ) != 0) {
cerr << "ERROR: source phrase does not match in line " << i << ": '"
<< itemDirect[1] << "' != '" << itemIndirect[1] << "'" << endl;
exit(1);
}
// output hierarchical phrase pair (with separated labels)
fileConsolidated << itemDirect[0] << " ||| " << itemDirect[1];
// SCORES ...
fileConsolidated << " |||";
vector<string> directCounts = tokenize(itemDirect[4].c_str());
vector<string> indirectCounts = tokenize(itemIndirect[4].c_str());
float countF = atof(directCounts[0].c_str());
float countE = atof(indirectCounts[0].c_str());
float countEF = atof(indirectCounts[1].c_str());
float n1_F, n1_E;
if (kneserNeyFlag) {
n1_F = atof(directCounts[2].c_str());
n1_E = atof(indirectCounts[2].c_str());
}
// Good Turing discounting
float adjustedCountEF = countEF;
if (goodTuringFlag && countEF+0.99999 < goodTuringDiscount.size()-1)
adjustedCountEF *= goodTuringDiscount[(int)(countEF+0.99998)];
float adjustedCountEF_indirect = adjustedCountEF;
// Kneser Ney discounting [Foster et al, 2006]
if (kneserNeyFlag) {
float D = kneserNey_D3;
if (countEF < 2) D = kneserNey_D1;
if (countEF < 3) D = kneserNey_D2;
if (D > countEF) D = countEF - 0.01; // sanity constraint
float p_b_E = n1_E / totalCount; // target phrase prob based on distinct
float alpha_F = D * n1_F / countF; // available mass
adjustedCountEF = countEF - D + countF * alpha_F * p_b_E;
// for indirect
float p_b_F = n1_F / totalCount; // target phrase prob based on distinct
float alpha_E = D * n1_E / countE; // available mass
adjustedCountEF_indirect = countEF - D + countE * alpha_E * p_b_F;
}
// prob indirect
if (!onlyDirectFlag) {
fileConsolidated << " " << maybeLogProb(adjustedCountEF_indirect/countE);
fileConsolidated << " " << itemIndirect[2];
}
// prob direct
fileConsolidated << " " << maybeLogProb(adjustedCountEF/countF);
fileConsolidated << " " << itemDirect[2];
// phrase count feature
if (phraseCountFlag) {
fileConsolidated << " " << maybeLogProb(2.718);
}
// low count feature
if (lowCountFlag) {
fileConsolidated << " " << maybeLogProb(exp(-1.0/countEF));
}
// alignment
fileConsolidated << " ||| " << itemDirect[3];
// counts, for debugging
fileConsolidated << "||| " << countE << " " << countF; // << " " << countEF;
if (outputNTLengths)
{
fileConsolidated << " ||| " << itemDirect[5];
}
fileConsolidated << endl;
}
fileDirect.Close();
fileIndirect.Close();
fileConsolidated.Close();
}
bool getLine( istream &fileP, vector< string > &item )
{
if (fileP.eof())
return false;
SAFE_GETLINE((fileP), line, LINE_MAX_LENGTH, '\n', __FILE__);
if (fileP.eof())
return false;
item = splitLine();
return true;
}
vector< string > splitLine()
{
vector< string > item;
bool betweenWords = true;
int start=0;
int i=0;
for(; line[i] != '\0'; i++) {
if (line[i] == ' ' &&
line[i+1] == '|' &&
line[i+2] == '|' &&
line[i+3] == '|' &&
line[i+4] == ' ') {
if (start > i) start = i; // empty item
item.push_back( string( line+start, i-start ) );
start = i+5;
i += 3;
}
}
item.push_back( string( line+start, i-start ) );
return item;
}
|