Uncertainty Estimation for Language Reward Models