We believe that fluency of English translated sentences is related to language model 4#4. In general, when we used a higher order 2#2-gram, the number of parameters dramatically increases, and the reliability for each parameter decreases. Thus we used a normal 5-gram model and did not use a higher 2#2-gram model. This model was the best language model among 2#2-gram from our experiments at the previous dry-run Intrinsic-JE task.