Retrieving keywords or keyphrases from a text document is one of the essential features in many applications. For example, the dataset search systems may use those keyphrases to suggest more relevant data to users. However, obtaining the keyphrases from a Japanese text is challenging due to its unique sentence structure. Moreover, most existing methods derive an extraction approach incapable of producing new Japanese keyphrases for each document. Thus, we propose a method for generating Japanese keyphrases based on a deep learning-based text-to-text generative model. Our method can accurately generate the existing and the additional relevant keyphrases for each document text. A model architecture used in our method also helps produce the Japanese keyphrases that preserve their original meanings.