2.2.1. Participants

The current study was approved by The ethics committee for research involving human subjects at the Graduate School of Engineering Science, Osaka University. The participants were 28 university students (14 male and 14 female students), whose ages ranged from 19 to 22 years old. They were randomly assigned to two conditions. In the passive condition, seven males and seven females participated in conversation with two androids that talked and directed their gaze to each other but did not do so to the participant. In the semi-passive one, the remaining seven males and seven females participated in the conversation with two androids that talked and directed their gaze not only to each other but also to the participant. No participants had any experience directly interacting with the androids before the experiment.

#### 2.2.2. Apparatus

In the experimental room, the participants faced two androids across a table. The distances between the two androids and the participant formed an equilateral triangle with 1.2-m sides (Figure 3). Figure 4 shows a scene of the experiment.

The topic of the conversation was related to android robots and the Intelligent Robotics Laboratory where they were developed. We designed an approximately ten-minute scenario for the conversation to introduce both the subjective and objective information. To verify how efficiently the subjective information was conveyed, we included utterances by which the androids explained their situation and how they felt. For example, "Visitors touch my body as they like. I would like them to refrain from it a little more." and "I am seldom quickly repaired (by the laboratory members) when I am broken. Don't you think it's terrible treatment?" Note that, in the semi-passive social condition, when they mentioned "you," they looked at the participant. In contrast, in the passive one, they did not look at him or her. Meanwhile, to verify how efficiently the objective information was conveyed, we included utterances with numbers or proper nouns to introduce facts about the laboratory and the androids. For example, "In fact, our bodies are driven by air pressure at about 0.7 MPa." Figure 5 shows a part of the conversation used for the experiment. In addition to the script of the passive social condition, nine utterances toward the participant were included in the conversation used for the semi-passive condition. Note that, to avoid providing further information by these nine utterances, the androids were limited to utterances asking about the participant's agreement or opinions in relation to the androids' conversation.

The responses from the participants were classified into the four categories: no answer, positive answer, negative answer, and other. This allowed the androids to choose an appropriate but brief response without providing any further information. To categorize the participant's answer as one of the four types, the system was trained before the experiment. For this, we constructed another system that could select a response for the androids using the Wizard of Oz method and collected morphemes as training data via conversations between eight males and the system.

**Figure 3.** The environment of the experiment.

**Figure 4.** A scene of the experiment.

**Figure 5.** Example of the conversation.

#### 2.2.3. Procedure

The experiment was conducted using the following procedure, one participant at a time. First, a participant waited in an anteroom until the start of the experiment. At a specified time, an experimenter explained an overview of the experiment in the anteroom and got the participant's consent regarding participation. After that, the experimenter provided instruction regarding the procedure of the experiment. In the instruction, the participant was requested to listen to the conversation between the androids. However, if the androids asked them questions during the conversation, the participant was permitted to answer it. Next, the participant put on a headset and configured the microphone using a function of Dragon Speech, which is an automatic speech recognition software. The software can adjust the volume level and check the quality of voice input by having the participant read a few prepared sentences aloud. Then, the participant moved to an experimental room containing the two androids. The participant sat across the table from the two androids. When the participant took their seat, the two androids started to talk each other. The conversation lasted for around 10 min, and when the conversation finished, the experimenter called out to the participant, directing them to go back to the anteroom. The participant removed the headset and answered a questionnaire and a recall test about the conversation. Lastly, the experimenter interviewed the participant.
