The following guideline is used for specifying the language model
filename.

  <language>[_<territory>][.<codeset>].<format>

<language> is lowercase and taken from ISO 639-1.  If ISO 639-1 does
not define a two-letter language code, a three-letter code defined by
ISO 639-2 is used.

The <territory> field is optional, uppercase, and taken from ISO
3166-1.  If ISO 3166-1 does not define a two-letter country code, use
two or three lowercase letters and if possible, use the top-level
domain for the country.

The <codeset> field is only optional if there is only one codeset
present for a language.  It should be specified using a lowercase
representation of the preferred MIME name for that codeset.

The <format> is "lm" to specify the original language model format and
"ln" to specify the new language model format.

--

The original language models in "lm" format are part of the TextCat
program located at http://odur.let.rug.nl/~vannoord/TextCat/ and were
originally authored by Gertjan van Noord <vannoord@let.rug.nl>.

tr.iso-8859-9.ln and ja.iso-2022-jp.ln were collated by Daniel Quinlan.