Language Model ($N$ -gram).

We calculated the $5$ -gram model using the n-gram-count in the SRILM toolkit [Stolcke2002] and used ``-ukndiscount -interpolate'' as the smoothing parameter.