Machine Translation from Japanese to Robot Language for Human-Friendly Communication

Humanoid robots should be capable of communicating naturally with humans to act as good partners to humans. However, there are several challenges associated with ensuring that humanoid robots are able to speak human languages. For example, to enable the humanoid robot NAO/Pepper of SoftBank Robotics Corp. to speak human-friendly Japanese, sentences in Japanese must first be translated into robot language. The target language is manually converted to the robot language to ensure that humanoid robots can communicate appropriately using human languages. Two facets ought to be incorporated into such translations: a dictionary for correct pronunciation must be created and another dictionary for expressing emotions must be created. This paper describes an approach that enables the humanoid robot NAO/Pepper to execute human-friendly communication. We manually prepared 32,293 pairs of sentences in Japanese and robot language as training data. Further, we created dictionaries to associate the correct pronunciations and emotional expressions with each pair of sentences using the morphemic analyzer MeCab. Finally, we developed a machine translation system by incorporating these dictionaries to translate Japanese to robot language. The performance of the machine translation system was verified using test sentences, and the results demonstrate the effectiveness of the proposed machine translation system, although some challenges still remain.