Sentence-final particles serve an important role in (spoken) Japanese, because they express the speaker's mental attitudes toward the proposition and/or the interlocutor. They are acquired at early ages and occur very frequently in everyday conversation. There has been, however, little proposal for the computational model of the acquisition of sentence-final particles. The purpose of this study is to get a robot to learn how to act upon the utterance with a sentence-final particle. The robot learns appropriate responses based on the rewards given by the interlocutor. The experimental results show that the robot learns to behave correctly in response to `yo,' which expresses the speaker's intention to communicate new information, and to `ne,' which denotes the speaker's desire to confirm that some information is shared. Using the learned actions as a lead, the acquisition of inner information processing such as word learning is the next research target.