ABSTRACT

We were surprised by the frequency with which children interacted with the robot in the company of other children (see Figures 8 and 9). Sixty-three percent of a first grader’s interaction time with the robot was in the company of one or more friends. Seventy-two percent of a sixth grader’s interaction time

(a)

(c)

(a) (b) with the robot was in the company of one or more friends. Because the pres­ence of friends could have affected each child’s learning of English from the robot, we controlled for the presence of friends in the following analyses. 4.4. Learning English

The analyses we present are analyses of variance in which the dependent variable is the improvement in each child’s English test score from the child’s English pretest score. Although many children did not know English at the be­ginning of the trial, some knew a bit. If they knew any of the phrases on the English test (such as “bye”) their improvement might have been small owing to a ceiling effect. Therefore, the appropriate analysis of the effects of the robot on learning is the change from the pretest to the posttest, controlling for the ini­tial pretest score. The main analyses we ran were standard least squares analy­ses, described as follows:Model (2nd week English score - pretest English score) = intercept + pre­test English score + Week 1 interaction minutes with robot + Week 2 interac­tion minutes with robot + percentage of interaction time with friends.The results of this analysis are shown in Figures 10 and 11. This analysis showed the expected significant ceiling effect of pretest English scores on the change in scores from pretest to posttest, /'(l, 198) = 86, p< .001. That is, the more English the children already knew at the beginning of the trial, the less they learned from the robot. However, the amount of time they interacted

with friends and the robots together did not have an impact on the change in the English scores. The amount of time children spent with the robot during the 1st week also had no effect on their improvement in English by the 2nd week, but the amount of time that children interacted with the robots during the 2nd week did have a significant and positive impact on improvement in English in the 2nd week, / ( l , 198) = 5.6, p = .02, d= .33.Because we found significant improvement in English learning after 2 weeks, we examined whether there was any evidence of improvement after only 1 week with the robot. This analysis showed that time spent with the robot during the 1 st week did not have a significant impact on the change in the Eng­lish scores from the pretest to the 1st week’s scores. Indeed, the trend was slightly negative [p< . 10). The absence of a 1st-week result suggests that learn­