Ten pairs of undergraduates learned a miniature artificial language (MAL) under the following two different conditions.Under the Active condition, one subject learned the MAL through interactions with a personal computer (NEC PC-9801) that acted as language tutor.The subject listened to sample MAL sentences provided by the computer along with their meaning visually presented on the color CRT. Each MAL sentence the subject produced was processed by the speech recognition unit (NEC SR-150) attached to the computer that parsed the sentence and gave a feedback. Under the Passive condition, the other subject learned the same MAL through only observing the interactions between the Active subject and the computer. Although the Behaviorist learning theories would predict the superior performances of the Active learners, the results showed the opposite.The poor performances of the Active learners were attributed to the speech recognition errors.Two additional experiments with ten pairs of subjects each were also performed confirming the same conclusion.