*3.3. Evaluation of Questionnaire for Hypotheses (II) and (III)*

Figure 6b,c show the questionnaire results. Medians of each question between the two conditions were compared using Mann–Whitney's U test. Figure 6b shows scores of four questions about the strength of subjective message (Q2-1, Q2-2, Q2-3, and Q2-4), and no significant difference was suggested. Figure 6c shows scores from the two questions about empathic concern from participants toward the androids (Q3-1 and Q3-2). The result of Q3-1 shows that the score of the semi-passive condition (Mdn = 5.5) was higher than that of the passive condition (Mdn = 4.5) with a marginal difference

(U = 57.5, *p* < 0.1). The result of Q3-2 shows that the score of the semi-passive condition (Mdn = 4.5) was significantly higher than that of the passive condition (Mdn = 3) (U = 48.5, *p* < 0.05).

**Figure 6.** The result of (**a**) the recall test, and the (**b**) strength of subjective message, and (**c**) empathic concern questionnaires.

#### **4. Discussion**

During the analysis of the recall test, we found that the participants remembered more words of the conversation in the semi-passive condition than in the passive one, which supports the first hypothesis. In the experiment, the participants attended a 10-min conversation where the androids asked them something roughly once every minute in the semi-passive condition. The questions in the recall test were chosen so that the answer words appeared between these interaction blocks of asking and answering in the semi-passive condition. In other words, the questions were independent of what the androids asked. Therefore, the interaction block was considered to encourage participants to remember not only the words that appeared in the block, but also in the entire conversation.

Regarding participants' evaluation of the strength of message, we did not find any improvement in the semi-passive condition compared to the passive one, which does not support the second hypothesis. In other words, the semi-passive form of the conversation did not effectively contribute to how much the participants were convinced by the subjective message from the android. On the other hand, higher empathic concern toward the androids was observed in the semi-passive condition than in the passive one, supporting the third hypothesis. In other words, the participants were moved to follow the misery situation of the androids. These two potentially inconsistent results may imply that the subjective message from the androids in the semi-passive conversation was successfully conveyed not through being felt that they were strongly requested but rather through spontaneously evoking the subject's empathic concern.

It is worth considering how the current result is limited by the fact that we tested only with one scenario including a subjective message from the androids. The remaining question is whether the improvement in recall performance is also maintained in a scenario without such a message, which should be beneficial for the general purpose of information conveyance. Therefore, a further experiment should use content with only objective information. On the other hand, even if it is limited to cases with subjective aspects, the results of this study should still be good news for information media because the development of affective applications have been considered [21]. However, the current result might be limited to the android robot, which is very human-like. Therefore, it is also worth investigating how much the information media should be human-like so that the effect of the semi-passive form of conversation can be utilized.

#### **5. Conclusions**

In this paper, we proposed efficient media to convey information based on what we call semi-passive conversation. To test the effect, a conversation system using two androids was implemented, which consisted of a scenario-based conversation system having a function of sometimes briefly listening to the user's agreement or disagreement to the conversation. Regarding objective information, the result of the recall test showed that the participants who attended the semi-passive conversation remembered more words in the conversation than those who attended the passive conversation which involved the same content except for the interaction blocks that checked the participants' agreement or disagreement. Regarding subjective information, the participants showed stronger agreement for the action of the androids, which was interpreted as indicating that spontaneously empathic concern was evoked more in the semi-passive conversation than in the passive one. Further experiments changing the subjective quality of the contents and human-likeness of the robots are important for future work.

**Author Contributions:** Conceptualization, T.N., Y.Y., K.O. and H.I.; methodology, T.N., Y.Y. and K.O.; software, T.N.; validation, T.N.; formal analysis, T.N.; investigation, T.N.; resources, Y.Y., K.O. and H.I.; data curation, T.N.; writing—original draft preparation, T.N.; writing—review and editing, Y.Y. and K.O.; visualization, T.N.; supervision, Y.Y. and K.O.; project administration, H.I.; funding acquisition, H.I.

**Funding:** This work was supported by the Japan Science and Technology Agency (JST) Exploratory Research for Advanced Technology: ERATO, grant number JPMJER1401, Japan.

**Conflicts of Interest:** The authors declare no conflict of interest.
