We have been evaluated English sentences on two points; accuracy and fluency. We believe that fluency is related to language model 3#3. Thus we used a normal 4-gram model and did not use a higher 4#4-gram model. In general, when we used a higher order 4#4-gram, the number of parameters dramatically increases, and the reliability for each parameter decreases. Therefore we chose a 4-gram model. This model is the best language model among 4#4-gram from our previous results at the IWSLT2007 contest.