<html>
<head>
<style><!--
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 10pt;
font-family:Tahoma
}
--></style>
</head>
<body class='hmmessage'><div dir='ltr'>
I am producing some language models (3-grams) for HTK.<br>What is the best configuration for produce the best language models using SRILM?<br>My configuration is:<br>$SRILM/ngram-count -memuse -order ${trigram} -interpolate -kndiscount -unk -vocab $wordlist -limit-vocab -text ${training} -lm ${train}-lm<br>${trigram}<br><br><br>The script line is above and I am using -kndiscount<br>Is there a better type of discount or parameters to produce better language models using SRILM?<br><br>Number of words (unique): 38k<br>Size: 93Mbytes<br>Number of lines: 550656<br>Number of words (total): 17166049 (17M)<br><br>Thanks.<br><br><br>Luis<br><br><br> </div></body>
</html>