In this paper, we propose a method for grounding move expressions in natural language to game states of Shogi. We introduce the concepts of commented trees and candidate trees. The fist step of grounding is enumerating candidate trees which consist of legal moves appearing in natural language comments using a rule-based algorithm. We then choose the best candidate as a commented tree using evaluation values provided by a computer Shogi program. In experiments, we have generated commented trees for positions annotated with comments by human experts, and conducted error analysis. We have also conducted experiments of commentary generation using commented trees. The results show that we have obtained commented trees with many positions, and about $80\%$ of them are correct. In addition, we have applied a commented tree to our commentary generation system, and succeeded in generating some useful comments.