[SRILM User List] question about srilm algorithm to build lm (arpa file)

[Andreas Stolcke]

On 10/5/2019 12:30 AM, Muhammad Hermawan wrote:
>  Pardon me to bothering srilm i've send you the my hypothesis how 
> srilm build the lm (arpa file) and i want ask you isn't right ? I read 
> it based on journal srilm toolkit and can you show me simple example 
> to build LM with srilm from n-gram modelling, n-gram probabilitas, 
> discounting methods & frequency smoothing, and arpa file.
>
> especially can show me to how to count with discount methods and 
> frequency smoothing with interpolation, witten-bell, and natural 
> discounting.
>
> i've try i just know how to count n-gram model and n-gram probabilities
>
> Hope you can help me cause i'm so confused,
>
> Muhammad Hermawan

Muhammad,

The basic steps for LM training and evaluation (including perplexity 
computation) are described in

         $SRILM/doc/lm-intro

For more background and in-depth information see the links at 
http://www.speech.sri.com/projects/srilm/manpages/.

If you google "srilm tutorial" you will find some tutorials created by 
other people, such as

https://gcc02.safelinks.protection.outlook.com/?url=https:%2F%2Fwww.cs.brandeis.edu%2F~cs114%2FCS114_docs%2FSRILM_Tutorial_20080512.pdf&data=01%7C01%7Csrilm-user%40speech.sri.com%7Ca33ec4d0d39741b777f108d74a0f5f4b%7C40779d3379c44626b8bf140c4d5e9075%7C1&sdata=mwqu67Cg8fEnAleZrr7zrPdEW7njNcIhV2MfIFb6KtE%3D&reserved=0

Good luck,

Andreas