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
|
// Copyright (C) 2003--2004 Darren Moore (moore@idiap.ch)
//
// This file is part of Torch 3.1.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
// 1. Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// 2. Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
// 3. The name of the author may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
// OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
// IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
// NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "Allocator.h"
#include "LMNGram.h"
namespace Torch {
LMNGram::LMNGram( int n_ , Vocabulary *vocab_ )
{
#ifdef DEBUG
if ( n_ < 1 )
error("LMNGram::LMNGram - n cannot be < 1\n") ;
#endif
n = n_ ;
vocab = vocab_ ;
// Create a unigram entry for every word in our vocabulary.
unigrams = (real *)Allocator::sysAlloc( vocab->n_words * sizeof(real) ) ;
for ( int i=0 ; i<vocab->n_words ; i++ )
unigrams[i] = LOG_ZERO ;
if ( n > 1 )
next_level = new LMInteriorLevelWordList( n-1 ) ;
else
next_level = NULL ;
// Configure the cache - TODO make the cache size configurable (?)
cache = new LMCache( 10 , n , vocab->n_words ) ;
}
LMNGram::~LMNGram()
{
if ( unigrams != NULL )
free(unigrams) ;
if ( next_level != NULL )
delete next_level ;
if ( cache != NULL )
delete cache ;
}
void LMNGram::addEntry( int order , int *words , real prob , real bo_wt )
{
// We assume that the 'words' array is in oldest-word-first order.
// ie. [w1 w2 w3] for trigram entry.
int r_words_1[30] ; // rearranged words [w4,w3,w2,w1]
int r_words_2[30] ; // rearranged words [w3,w2,w1,w4]
#ifdef DEBUG
if ( (order < 1) || (order > n) )
error("LMNGram::addEntry - order out of range\n") ;
#endif
if ( order == 1 )
{
if ( unigrams[*words] > LOG_ZERO )
error("LMNGram::addEntry - duplicate unigram entry encountered\n") ;
unigrams[*words] = prob ;
if ( n > 1 )
{
// We are adding a unigram entry into a language model
// of higher order, so the backoff weight is important.
// Add the backoff weight.
next_level->addBackoffEntry( order , words , bo_wt ) ;
}
}
else
{
// The first thing we want to do is to rearrange the words in the entry
// so that the order is straight forward and matches the architecture
// of the language model, for adding both probs and backoffs.
for ( int i=0 ; i<(order-1) ; i++ )
{
r_words_1[i] = words[order-1-i] ; // backoff ordering
r_words_2[i] = words[order-2-i] ; // prob ordering
}
r_words_1[order-1] = words[0] ;
r_words_2[order-1] = words[order-1] ;
if ( prob > LOG_ZERO )
next_level->addProbEntry( order , r_words_2 , prob ) ;
if ( order < n )
{
// Add the backoff - we don't have/need, eg, trigram backoffs in our
// trigram language model.
next_level->addBackoffEntry( order , r_words_1 , bo_wt ) ;
}
}
}
real LMNGram::getLogProbBackoff( int order , int *words )
{
// There are 'order' entries in 'words'.
// The ordering in words is W3,W2,W1,W4 for a 4-gram query.
// ie. for the query : what is P(W4|W1,W2,W3) ?
real temp , prob ;
#ifdef DEBUG
if ( order < 1 )
error("LMNGram::getNextWordList - order out of range\n") ;
bool output_debug=false ;
if ( output_debug == true )
{
printf( "P( %s | " , vocab->words[words[order-1]] ) ;
for ( int i=0 ; i<(order-1) ; i++ )
printf( "%s " , vocab->words[words[i]] ) ;
printf(") = ") ;
}
#endif
if ( order > n )
order = n ;
if ( order == 1 )
{
// Just return the unigrams prob - no need to backoff or to use the cache.
#ifdef DEBUG
if ( output_debug == true )
printf("%f\n",unigrams[*words]);
#endif
return unigrams[*words] ;
}
else
{
// look in the cache ...
prob = cache->getProb( order , words ) ;
if ( prob >= (-LOG_ZERO) )
{
// The n-gram entry is not in the cache, so we need to search for it
// and then add it to the cache.
if ( next_level->getProbWithBackoff( order , words , &prob ) == false )
{
// No bigram, trigram, etc entries so backoff to unigram
temp = unigrams[words[order-1]] ;
if ( temp <= LOG_ZERO )
prob = LOG_ZERO ;
else if ( prob <= LOG_ZERO )
prob = temp ;
else
prob += temp ;
}
cache->addEntry( order , words , prob ) ;
#ifdef DEBUG
if ( output_debug == true )
printf("%f\n",prob) ;
#endif
return prob ;
}
else
{
// The entry was in the cache. Just return it
#ifdef DEBUG
if ( output_debug == true )
printf("(c)%f\n",prob);
#endif
return prob ;
}
}
}
#ifdef DEBUG
void LMNGram::outputText()
{
int words[30] ;
// Print the unigrams
printf("\\1-gram\\\n") ;
for ( int i=0 ; i<vocab->n_words ; i++ )
printf("%s %f\n",vocab->words[i],unigrams[i]) ;
// Print the rest
if ( next_level != NULL )
next_level->outputText( vocab , words , 0 ) ;
}
#endif
}
|
