1SLMBUILD(1)           User Contributed Perl Documentation          SLMBUILD(1)
2
3
4

NAME

6       slmbuild - generate language model from idngram file
7

SYNOPSIS

9       slmbuild [option]... idngram_file...
10

DESCRIPTION

12       slmbuild generates a back-off smoothing language model from a given
13       idngram file. Generally, the idngram_file is created by ids2ngram.
14

OPTIONS All the following options are mandatory.

16       -n,--NMax N
17           1 for unigram, 2 for bigram, 3 for trigram. Any number not in the
18           range of 1..3 is not valid.
19
20       -o, --out output-file
21           Specify the output xfilei name.
22
23       -l, --log
24           using -log(pr), use pr directly by default.
25
26       -w, --wordcount N
27           Lexican size, number of different words.
28
29       -b, --brk id...
30           Set the ids which should be treated as breaker.
31
32       -e, --e id...
33           Set the ids which should not be put into LM.
34
35       -c, --cut c...
36           k-grams whose freq <= c[k] are dropped.
37
38       -d, --discount method, param...
39           The k-th -d parm specifies the discount method
40
41           For k-gram, possibble values for method/param are:
42
43                 B<GT>,I<R>,I<dis>  : B<GT> discount for r E<lt>= I<R>, r is the freq of a ngram.
44                             Linear discount for those r E<gt> I<R>, i.e. r'=r*dis
45                             0 E<lt>E<lt> dis E<lt> 1.0, for example 0.999
46                 B<ABS>,[I<dis>] : Absolute discount r'=r-I<dis>. And I<dis> is optional
47                             0 E<lt>E<lt> I<dis> E<lt> cut[k]+1.0, normally I<dis> E<lt> 1.0.
48                 LIN,[I<dis>] : Linear discount r'=r*dis. And dis is optional
49                             0 E<lt> dis E<lt> 1.0
50

NOTE

52       -n must be given before -c -b. And -c must give right number of cut-
53       off, also -ds must appear exactly N times specifying the discounts for
54       1-gram, 2-gram..., respectively.
55
56       BREAKER-IDs could be SentenceTokens or ParagraphTokens. Conceptually,
57       these ids have no meaning when they appeared in the middle of n-gram.
58
59       EXCLUDE-IDs could be ambiguious-ids. Conceptually, n-grams which
60       contain those ids are meaningless.
61
62       We can not erase ngrams according to BREAKER-IDS and EXCLUDE-IDs
63       directly from IDNGRAM file, because some low-level information is still
64       useful in it.
65

EXAMPLE

67       Following example read 'all.id3gram' and write trigram model 'all.slm'.
68
69       At 1-gram level, use Good-Turing discount with cut-off 0, i<R>=8,
70       dis=0.9995. At 2-gram level, use Absolute discount with cut-off 3, dis
71       auto-calc. At 3-gram level, use Absolute discount with cut-off 2, dis
72       auto-calc. Word id 10,11,12 are breakers (sentence/para/paper breaker,
73       etc). Exclude-ID is 9. Lexicon contains 200000 words. The result
74       languagme model uses -log(pr).
75
76       slmbuild -l -n 3 -o all.slm -w 200000 -c 0,3,2 -d GT,8,0.9995 -d ABS -d
77       ABS -b 10,11,12 -e 9 all.id3gram
78

AUTHOR

80       Originally written by Phill.Zhang <phill.zhang@sun.com>.  Currently
81       maintained by Kov.Chai <tchaikov@gmail.com>.
82

SEE ALSO

84       ids2ngram(1), slmprune(1).
85
86
87
88perl v5.28.1                      2016-03-01                       SLMBUILD(1)
Impressum