File: LanguageModel.cc

package info (click to toggle)
torch3 3.1-2
  • links: PTS
  • area: main
  • in suites: lenny
  • size: 2,968 kB
  • ctags: 2,743
  • sloc: cpp: 24,245; python: 299; makefile: 153
file content (665 lines) | stat: -rw-r--r-- 26,081 bytes parent folder | download | duplicates (5)
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
// 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 "LanguageModel.h"
#include "string_stuff.h"
#include "DiskXFile.h"


namespace Torch {


LanguageModel::LanguageModel( int order_ , Vocabulary *vocabulary_ , 
                              char *lm_fname , real lm_scaling_factor_ )
{
    FILE *lm_fd ;
    char buf[4] ;
    
    if ( vocabulary_ == NULL )
        error("LanguageModel::LanguageModel - no vocabulary defined\n") ;
    if ( order_ <= 0 )
        error("LanguageModel::LanguageModel - LM order must be > 0\n") ;
    if ( (lm_fname == NULL) || (strcmp(lm_fname,"")==0) )
        error("LanguageModel::LanguageModel - no LM filename specified\n") ;
        
    order = order_ ;
    vocabulary = vocabulary_ ;
    n_words = vocabulary->n_words ;
    lm_scaling_factor = lm_scaling_factor_ ;
    ngram = new LMNGram( order , vocabulary ) ;

    lm_has_start_word = false ;
    lm_has_end_word = false ;

    // Open the LM file
    if ( (lm_fd = fopen( lm_fname , "r" )) == NULL )
        error("LanguageModel::LanguageModel - error opening LM file\n") ;

    // Read the first 4 bytes to see if the file is a Noway binary file
    if ( (int)fread( buf , sizeof(char) , 4 , lm_fd ) != 4 )
        error("LanguageModel::LanguageModel - error reading first 4 bytes\n") ;
    
    if ( (strcmp( buf , "NG3" ) == 0) || (strcmp( buf , "TR2" ) == 0) )
    {
        // The file is in Noway binary format.
        readNowayBin( lm_fd ) ;
    }
    else
    {
        // Assume that the file is in ARPA format.
        fseek( lm_fd , 0 , SEEK_SET ) ;
        readARPA( lm_fd ) ;
    }

    fclose( lm_fd ) ;
 
#ifdef DEBUG
    //outputText() ;
#endif
}


LanguageModel::~LanguageModel()
{
    if ( ngram != NULL )
        delete ngram ;
}


real LanguageModel::calcLMProb( DecodingHypothesis *prev_word_end_hyp , int next_word )
{
    int words[30] , n_wrds ;
    WordChainElem *temp_elem ;
    real prob ;

#ifdef DEBUG
    if ( next_word < 0 )
        error("LanguageModel::calcLMProb(2) - next_word < 0\n") ;
#endif

    // We have a word end hypothesis and a next word.
    // We want to calculate the LM probability for this next word.
    // eg. We have a word end for some w2 and a next word w3.
    //     We want to calculate P(w3|w1,w2)
    
    // If the next word is silence or a sentence marker, don't do a LM lookup.
    if ( (next_word == vocabulary->sent_start_index) && (lm_has_start_word == false) )
        return 0.0 ;
    if ( (next_word == vocabulary->sent_end_index) && (lm_has_end_word == false) )
        return 0.0 ;
    if ( next_word == vocabulary->sil_index )
        return 0.0 ;
    
    // Construct a list of the previous words.
    n_wrds = 0 ;
    if ( prev_word_end_hyp != NULL )
    {    
        temp_elem = prev_word_end_hyp->word_level_info ;
#ifdef DEBUG
        if ( prev_word_end_hyp->word_level_info == NULL )
            error("LanguageModel::calcLMProb(2) - word_level_info is NULL\n") ;
#endif

        while ( temp_elem != NULL )
        {
            if ( (temp_elem->word == vocabulary->sil_index) ||
                 ((lm_has_start_word==false) && (temp_elem->word==vocabulary->sent_start_index)) )
            {
                // skip these words for the purpose of LM lookups
                temp_elem = temp_elem->prev_elem ;
                continue ;
            }

            words[n_wrds++] = temp_elem->word ;
            temp_elem = temp_elem->prev_elem ;
            if ( n_wrds >= (order-1) )
                break ;
        }
    }
    
    words[n_wrds++] = next_word ;
    
    // Find the n-gram probability
    prob = ngram->getLogProbBackoff( n_wrds , words ) ;

    // Scale the n-gram probability using the LM scaling factor.
    // Note that we are multiplying the log LM prob by the scaling factor
    prob *= lm_scaling_factor ;

    return ( prob ) ;
}


real LanguageModel::calcLMProb( DecodingHypothesis *word_end_hyp )
{
    int words[30] , n_wrds , temp , j ;
    WordChainElem *temp_elem ;
    real prob ;
    
    if ( word_end_hyp == NULL )
        return LOG_ZERO ;
        
    // We have a word end hypothesis.  We want to tune this using our language model.
    // eg. we have a word end for some w3 and we want to tune this using P(w3|w1,w2)
        
    // Construct a list of the current word and previous words.
    n_wrds = 0 ;
    temp_elem = word_end_hyp->word_level_info ;
#ifdef DEBUG
    if ( word_end_hyp->word_level_info == NULL )
        error("LanguageModel::calcLMProb - word_level_info is NULL\n") ;
#endif
    // If the most recent word is a sentence marker and the LM does not
    //   have entries for the sentence markers, don't do a LM lookup.
    if ( (temp_elem->word == vocabulary->sent_start_index) && (lm_has_start_word == false) )
        return 0.0 ;
    if ( (temp_elem->word == vocabulary->sent_end_index) && (lm_has_end_word == false) )
        return 0.0 ;
    if ( temp_elem->word == vocabulary->sil_index )
        return 0.0 ;

    while ( temp_elem != NULL )
    {
        if ( (temp_elem->word == vocabulary->sil_index) ||
            ((lm_has_start_word==false) && (temp_elem->word==vocabulary->sent_start_index)) )
        {
            // skip these words for the purpose of LM lookups
            temp_elem = temp_elem->prev_elem ;
            continue ;
        }
        words[n_wrds++] = temp_elem->word ;
        temp_elem = temp_elem->prev_elem ;
        if ( n_wrds >= order )
            break ;
    }

    // The method in LMNGram requires a different ordering of words
    temp = words[0] ;
    for ( j=1 ; j<n_wrds ; j++ )
        words[j-1] = words[j] ;
    words[n_wrds-1] = temp ;
        
    // Find the n-gram probability
    prob = ngram->getLogProbBackoff( n_wrds , words ) ;

    // Scale the n-gram probability using the LM scaling factor.
    // Note that we are multiplying the log LM prob by the scaling factor
    //   IS THIS OK ?
    prob *= lm_scaling_factor ;

    // Tune the word end hypothesis score using the n-gram probability and return
    return ( prob ) ;
}


void LanguageModel::readARPA( FILE *arpa_fd )
{
    int n_exp_entries[30] ; // n_exp_entries[0] is the expected number of unigram entries
                            // n_exp_entries[0] is the expected number of bigram entries
    int n_act_entries[30] ; // the actual number of entries read from the file.
    int words[30] ;    // holds the predecessor words for a given word.
    real curr_prob=0.0 , curr_bow=0.0 ;
    char *curr_word=NULL ;
    int curr_index ;
    real ln_10 = (real)log(10.0) ;
    int tempn=0 , tempn_entries=0 , max_n_in_file=0 ;
    char line[1000] ;
    bool got_begin_data_mark=false , expecting_end=false , got_end=false , error_flag ;
    int curr_gram_data=0 ;
    
    if ( arpa_fd == NULL )
        error("LanguageModel::readARPA - arpa_fd is NULL\n") ;
        
    // discard lines until we get the "beginning of data mark".
    while ( fgets( line , 1000 , arpa_fd ) != NULL )
    {
        // if the new line is empty, get the next line
        if ( (line[0]==' ') || (line[0]=='\r') || (line[0]=='\n') || (line[0]=='\t') ||
             (line[0]=='#') )
            continue ;
            
        if ( line[0] == '\\' )
        {
            strtoupper( line ) ;
            if ( strstr( line , "\\END\\" ) != NULL )
            {
                if ( curr_gram_data < order )
                {
                    // we haven't encountered the n-grams we expected
                    error("LanguageModel::readARPA - not enough data in file\n") ;
                }
            
                // we've reached the end of the ARPA file - we're done
                got_end = true ;
                break ;
            }
            else if ( expecting_end == true )
            {
                // we're expecting the end marker and didn't get it - get the next line
                continue ;
            }
            else if ( strstr( line , "\\DATA\\" ) != NULL )
            {
                if ( got_begin_data_mark == true )
                {
                    // we have already seen the beginning of data marker - error !
                    error("LanguageModel::readARPA - duplicate beginning of data marker\n") ;
                }
                got_begin_data_mark = true ;
            }
            else if ( strstr( line , "-GRAMS:" ) != NULL )
            {
                if ( got_begin_data_mark == true )
                {
                    // find out whether we are at the start of the 1-gram, 
                    //   2-gram, 3-gram, etc data.
                    if ( (curr_gram_data+1) != ( line[1]-0x30 ) )
                        error("LanguageModel::readARPA - N-Gram N out of order\n") ;

                    curr_gram_data = line[1]-0x30 ;
                    n_act_entries[curr_gram_data-1] = 0 ;
                    if ( curr_gram_data > order )
                    {
                        // the order has exceeded the order of our LM - we're done 
                        //   reading probabilties.
                        expecting_end = true ;
                    }
                }
            }
            else
            {
                // we got something else that started with a '\' - error !!
                error("LanguageModel::readARPA - unrecognised marker\n%s\n",line) ;
            }
        }
        else
        {
            if ( (got_begin_data_mark == false) || (expecting_end == true) )
                continue ;

            if ( strstr( line , "<UNK>") != NULL )
            {
                n_exp_entries[curr_gram_data-1]-- ;
                continue ;
            }

            if ( curr_gram_data == 0 )
            {
                // we are just below the \data\ - therefore expecting ngram x=y lines
                sscanf( line , "%*s %d=%d" , &tempn , &tempn_entries ) ;
                if ( tempn != (max_n_in_file+1) )
                    error("LanguageModel::readARPA - ngram n=y -> unexpected n\n") ;
                max_n_in_file = tempn ;
                if ( tempn <= order )
                    n_exp_entries[tempn-1] = tempn_entries ;
            }    
            else if ( (curr_gram_data > 0) && (curr_gram_data < max_n_in_file) )
            {
                // The line should contain (curr_gram_data+2) fields.
                // eg. for 2-gram entry -> p wd_1 wd_2 bo_wt_2
            
                // Read the probability from the first field (in log10 format) and convert
                //   to ln format.
#ifdef USE_DOUBLE
                if ( sscanf( line , "%lf" , &curr_prob ) != 1 )
#else
                if ( sscanf( line , "%f" , &curr_prob ) != 1 )
#endif
                    error("LanguageModel::readARPA - error reading prob\n") ;
                if ( curr_prob < -90.0 )
                    curr_prob = LOG_ZERO/2 ;
                else
                    curr_prob *= ln_10 ;
                
                // get past the prob field so we can read the words
                strtok( line , " \n\r\t" ) ;
                                
                // read wd_1 , ... , wd_n (ie. all predecessor words of wd_n)
                error_flag = false ;
                for ( int i=0 ; i<curr_gram_data ; i++ )
                {
                    // Extract the next word from the line
                    curr_word = strtok( NULL , " \n\r\t" ) ;

                    // determine the index of the word in the vocabulary
                    curr_index = vocabulary->getIndex( curr_word ) ;
                    if ( curr_index < 0 ) 
                    {
                        // The word is not in our vocab - don't add the entry to our LM
                        error_flag = true ;
                        n_exp_entries[curr_gram_data-1]-- ;
                        break ;
                        //error("LanguageModel::readARPA - %s in ARPA file not in vocab\n" ,
                        //                           curr_word ) ;
                    }
                    else
                    {
                        if ( curr_index == vocabulary->sent_start_index )
                            lm_has_start_word = true ;
                        if ( curr_index == vocabulary->sent_end_index )
                            lm_has_end_word = true ;
                    }
                    
                    // Place the word index into the array of predecessor words in
                    //   oldest-word-first order.
                    words[i] = curr_index ;
                }

                if ( error_flag == true )
                    continue ;

                // Extract the back off weight from the last field in the line and
                //   convert from log10 to ln.
#ifdef USE_DOUBLE
                if ( sscanf( strtok( NULL , " \n\r\t" ) , "%lf" , &curr_bow ) != 1 )
#else
                if ( sscanf( strtok( NULL , " \n\r\t" ) , "%f" , &curr_bow ) != 1 )
#endif
                    error("LanguageModel::readARPA - back off weight not found\n") ;
                if ( curr_bow < -90.0 )
                    curr_bow = 0.0 ;
                else
                    curr_bow *= ln_10 ;
                
                // add the entry to the curr_gram_data-gram for the new word
                ngram->addEntry( curr_gram_data, words, curr_prob, curr_bow ) ;
                n_act_entries[curr_gram_data-1]++ ;
            }
            else if ( curr_gram_data == max_n_in_file )
            {
                // The line should contain (curr_gram_data+1) fields because
                //   backoff weights are only required for N-grams that form a prefix of
                //   longer N-grams in the model file (ie. not this one - the longest).
                // eg. for 4-gram entry -> p wd_1 wd_2 wd_3 wd_4
                //    (where 4-gram probabilities are the maximum in the file.
            
                // read the probability from the first field
#ifdef USE_DOUBLE
                sscanf( line , "%lf" , &curr_prob ) ;
#else
                sscanf( line , "%f" , &curr_prob ) ;
#endif
                if ( curr_prob < -90.0 )
                    curr_prob = LOG_ZERO/2 ;
                else
                    curr_prob *= ln_10 ;
                
                // get past the prob field so we can read the words
                strtok( line , " \n\r\t" ) ;
                                
                // read wd_1 , ... , wd_n and insert indices in 'words' array.
                error_flag = false ;
                for ( int i=0 ; i<curr_gram_data ; i++ )
                {
                    // Extract the next word from the line
                    curr_word = strtok( NULL , " \n\r\t" ) ;

                    // determine the index of the word in the vocabulary
                    curr_index = vocabulary->getIndex( curr_word ) ;
                    if ( curr_index < 0 ) 
                    {
                        // The word is not in our vocab - don't add the entry to our LM
                        error_flag = true ;
                        n_exp_entries[curr_gram_data-1]-- ;
                        break ;
                        //error("LanguageModel::readARPA - %s in ARPA file not in vocab\n" ,
                        //                           curr_word ) ;
                    }
                    else
                    {
                        if ( curr_index == vocabulary->sent_start_index )
                            lm_has_start_word = true ;
                        if ( curr_index == vocabulary->sent_end_index )
                            lm_has_end_word = true ;
                    }
                    
                    // Place the word index into the array of predecessor words in
                    //   oldest-word-first order.
                    words[i] = curr_index ;
                }
                    
                if ( error_flag == true )
                    continue ;

                // add the entry to the curr_gram_data-gram for the new word
                n_act_entries[curr_gram_data-1]++ ;
                ngram->addEntry( curr_gram_data , words , curr_prob ) ;
            }
        }
    }   
    
    // make sure that we got the /end/ marker
    if ( got_end == false )
        error("LanguageModel::readARPA - EOF but no end marker\n") ;
        
    // Issue warnings if the number of expected entries for each n-gram did
    //   not match the actual number read from the file
    for ( int i=0 ; i<order ; i++ )
    {
        if ( n_exp_entries[i] != n_act_entries[i] )
        {
            warning("LanguageModel::readARPA - warning - %d-gram entry count mismatch\n" , i+1 ) ;
            warning("\t%d expected != %d actual\n",n_exp_entries[i],n_act_entries[i]) ;
        }
    }
}


void LanguageModel::readNowayBin( FILE *nw_fd )
{
    // This reads a Noway Binary format language model file.
    // Only the TR2 and NG3 types are supported (ie. trigrams).
    char buf[1000] , *bptr , c ;
    int n_xgrams[3] ;
    unsigned short wd_index , log10_prob , n_bigrams , n_trigrams ;
    short log10_backoff ;
    int *vocab_index_map ;
    int total_bigrams=0 , total_trigrams=0 , words[3] ;
    real prob , backoff ;
    real neg_ln_10 = -log(10.0) , ln_10 = log(10.0) ;
    bool uni_invocab=true , bi_invocab=true , tri_invocab=true ;
    DiskXFile *lm_file ;
    
#ifdef DEBUG
    if ( (sizeof(unsigned short) != 2) || (sizeof(int) != 4) )
        error("LanguageModel::readNowayBin - unexpected type sizes\n") ;
#endif

    if ( nw_fd == NULL )
        error("LanguageModel::readNowayBin - nw_fd is NULL\n") ;
    if ( order > 3 )
        error("LanguageModel::readNowayBin - order is greater than 3\n") ;

    lm_file = new DiskXFile( nw_fd ) ;
    
    // Assume that the first 4-bytes have already been read, 
    //   and have been verified to contain TR2 or NG3.
    // Read the number of unigrams, bigrams and trigrams.
    if ( lm_file->read( n_xgrams , sizeof(int) , 3 ) != 3 )
        error("LanguageModel::readNowayBin - error reading number of ngrams\n") ;

    // Allocate memory to hold the mapping between our vocab indices and
    //   the indices in the file.
    vocab_index_map = (int *)Allocator::sysAlloc( n_xgrams[0] * sizeof(int) ) ;

    // Now read in the "vocabulary"
    for ( int i=0 ; i<n_xgrams[0] ; i++ )
    {
        // Read the word string and make sure it is uppercase
        bptr = buf ;
        lm_file->read( &c , 1 , 1 ) ;
        while ( c > 0 )
        {
            *bptr = (unsigned char)c ;
            bptr++ ;
            lm_file->read( &c , 1 , 1 ) ;
        }
        *bptr = '\0' ;
        //strtoupper( buf ) ;
        
        // Read the word index from the file
        if ( lm_file->read( &wd_index , sizeof(unsigned short) , 1 ) != 1 )
            error("LanguageModel::readNowayBin - error reading word index\n") ;

        if ( wd_index >= n_xgrams[0] )
            error("LanguageModel::readNowayBin - word index exceeds num unigrams\n") ;
            
        // Find the index of the word in our vocabulary and store the mapping.
        vocab_index_map[wd_index] = vocabulary->getIndex( buf ) ;

        if ( (vocab_index_map[wd_index] >= 0) && 
             (vocab_index_map[wd_index] == vocabulary->sent_start_index) )
             lm_has_start_word = true ;
        if ( (vocab_index_map[wd_index] >= 0) && 
             (vocab_index_map[wd_index] == vocabulary->sent_end_index) )
             lm_has_end_word = true ;
    }
    
    // Now we read the unigram, bigram and trigram entries.
    for ( int i=0 ; i<n_xgrams[0] ; i++ )
    {
        // Read the word index from file
        if ( lm_file->read( &wd_index , sizeof(unsigned short) , 1 ) != 1 )
            error("LanguageModel::readNowayBin - error reading unigram word index\n") ;
        if ( vocab_index_map[wd_index] < 0 )
            uni_invocab = false ;
        
        // Read the probability ( -(log10(prob)*8192) format ) and convert to ln(prob).
        if ( lm_file->read( &log10_prob , sizeof(unsigned short) , 1 ) != 1 )
            error("LanguageModel::readNowayBin - error reading unigram prob\n") ;
        prob = ((real)log10_prob / 8192.0) * neg_ln_10 ;
        
        // Read the backoff weight
        if ( lm_file->read( &log10_backoff , sizeof(short) , 1 ) != 1 )
            error("LanguageModel::readNowayBin - error reading unigram backoff\n") ;
        backoff = ((real)log10_backoff / 8192.0) * ln_10 ;

#ifdef DEBUG
        if ( prob > 0.0 )
            error("LanguageModel::readNowayBin - prob > 0.0\n") ;
#endif

        // Read the number of bigrams associated with this word
        if ( lm_file->read( &n_bigrams , sizeof(unsigned short) , 1 ) != 1 )
            error("LanguageModel::readNowayBin - error reading number of bigrams\n") ;
        total_bigrams += n_bigrams ;

        if ( uni_invocab == true )
        {
            // Add the new entry to the 1-gram
            words[0] = vocab_index_map[wd_index] ;
            ngram->addEntry( 1 , words , prob , backoff ) ;
        }
        
        // Now read all the bigrams that have the current word as the predecessor
        for ( int j=0 ; j<n_bigrams ; j++ )
        {
            // Read the word index from file
            if ( lm_file->read( &wd_index , sizeof(unsigned short) , 1 ) != 1 )
                error("LanguageModel::readNowayBin - error reading bigram word index\n") ;
            if ( vocab_index_map[wd_index] < 0 )
                bi_invocab = false ;
            
            // Read the probability ( -(log10(prob)*8192) format ) and convert to ln(prob).
            if ( lm_file->read( &log10_prob , sizeof(unsigned short) , 1 ) != 1 )
                error("LanguageModel::readNowayBin - error reading bigram prob\n") ;
            prob = ((real)log10_prob / 8192.0) * neg_ln_10 ;

            // Read the backoff weight
            if ( lm_file->read( &log10_backoff , sizeof(short) , 1 ) != 1 )
                error("LanguageModel::readNowayBin - error reading bigram backoff\n") ;
            backoff = ((real)log10_backoff / 8192.0) * ln_10 ;
#ifdef DEBUG
            if ( prob>0.0 )
                error("LanguageModel::readNowayBin - bigram prob > 0.0\n") ;
#endif

            // Read the number of trigrams associated with this word
            if ( lm_file->read( &n_trigrams , sizeof(unsigned short) , 1 ) != 1 )
                error("LanguageModel::readNowayBin - error reading number of trigrams\n") ;
            total_trigrams += n_trigrams ;

            // Add the new entry to the 2-gram
            if ( (uni_invocab == true) && (bi_invocab == true) )
            {
                words[1] = vocab_index_map[wd_index] ;
                if ( order >= 2 )
                    ngram->addEntry( 2 , words , prob , backoff ) ;
            }
        
            // Now read all the trigrams that have the current bigram as the predecessor
            for ( int k=0 ; k<n_trigrams ; k++ )
            {
                // Read the word index from file
                if ( lm_file->read( &wd_index , sizeof(unsigned short) , 1 ) != 1 )
                    error("LanguageModel::readNowayBin - error reading trigram word index\n") ;
                if ( vocab_index_map[wd_index] < 0 )
                    tri_invocab = false ;
                    
                // Read the probability ( -(log10(prob)*8192) format ) and convert to ln(prob).
                if ( lm_file->read( &log10_prob , sizeof(unsigned short) , 1 ) != 1 )
                    error("LanguageModel::readNowayBin - error reading trigram prob\n") ;
                prob = ((real)log10_prob / 8192.0) * neg_ln_10 ;

#ifdef DEBUG
                if ( prob > 0.0 )
                    error("LanguageModel::readNowayBin - bigram prob or backoff > 0.0\n") ;
#endif

                // Add the new entry to the 3-gram
                if ( (uni_invocab == true) && (bi_invocab == true) && (tri_invocab == true) )
                {
                    words[2] = vocab_index_map[wd_index] ;
                    if ( order >= 3 )
                        ngram->addEntry( 3 , words , prob ) ;
                }
                tri_invocab = true ;
            }
            bi_invocab = true ;
        }
        uni_invocab = true ;
    }

    if ( (total_bigrams != n_xgrams[1]) || (total_trigrams != n_xgrams[2]) )
        error("LanguageModel::readNowayBin - did not read expected number of bi & trigrams\n") ;

    free( vocab_index_map ) ;
    delete lm_file ;
}


#ifdef DEBUG
void LanguageModel::outputText()
{
    ngram->outputText() ;
}
#endif


}