Gathering players is unavoidably costly when we run large-scale "Langrid Gaming", which is a Language Grid-linked multilingual and multinational gaming simulation. If people can be replaced in the gaming simulations, the cost decreases. Thus, a player agent has been introduced. First, game experiments were conducted in natural language, and parallel texts and player agents were developed by analyzing the experiments. Second, experiments were conducted in Japanese Experiment 1) and in Japanese and Chinese (Experiment 2). Negotiation protocols were analyzed, and new parallel texts were developed from Experiments 1 and 2. Third, experiments were conducted in Japanese (Experiment 3) once again, and Experiments 1, 2, and 3 were compared and analyzed. The following results were obtained. (1) Using parallel texts enables a player agent to be introduced. (2) Various negotiation protocols can be extracted by repeating an experiment with a greater variety of participants. (3) Valid parallel texts can be extracted by analyzing negotiation protocols and functions of utterances. (4) Adding new parallel texts reduces fruitless utterances and changes the negotiations protocol.