Role-play scenarios have been considered a successful learning space for children to develop their social and emotional abilities. In this paper, we investigate whether socially assistive robots in role-playing settings are as effective with small groups of children as they are with a single child and whether individual factors such as gender, grade level (first vs. second), perception of the robots (peer vs. adult), and empathy level (low vs. high) play a role in these two interaction contexts. We conducted a three-week repeated exposure experiment where 40 children interacted with socially assistive robotic characters that acted out interactive stories around words that contribute to expanding children's emotional vocabulary. Our results showed that although participants who interacted alone with the robots recalled the stories better than participants in the group condition, no significant differences were found in children's emotional interpretation of the narratives. With regard to individual differences, we found that a single child setting appeared more appropriate to first graders than a group setting, empathy level is an important predictor for emotional understanding of the narratives, and children's performance varies depending on their perception of the robots (peer vs. adult) in the two conditions.