To acquire oral skills in a second language (L2), L2 learners need to start with form-focused practice that upgrades their declarative knowledge of syntactic forms to procedural knowledge. Even in form-focused practice, learners should construct a meaningful sentence that is coherent with the conversational context. However, if they cannot construct a meaningful sentence by themselves, they have to start from simply repeating a model answer. In this study, we conducted a comparative evaluation of two training methods: one based on a question-answering (QA) task and the other based on a repetition (RP) task using a robot-assisted language learning system. The experimental results showed that the QA-task-based training led to a higher retention effect than the RP-task-based training 10-30 days after the training period, though the results of a post-test on the fifth day were not significantly different between the two groups.