Scripted interventions versus reciprocal teaching in collaborative learning: A comparison of pedagogical and teachable agents using a cognitive architecture

Abstract

Background

From a social constructivist viewpoint, collaborative learning—where learners engage through explanatory activities—is deemed effective. Research on computer-supported collaborative learning explored effective scaffolding provisions during learner interactions, revealing both the positive and negative effects of pedagogical interventions.

Aims

We posit that tutor interventions using a pedagogical conversational agent through scripted pedagogy can enhance collaborative learning. Learning-science studies have shown that reciprocal teaching between learners can improve their learning performance. Method superiority, implementation, and effects of combining scripted and reciprocal teaching strategies were investigated.

Sample

Participants comprised 134 university students.

Methods

We created a collaborative-learning environment wherein learner pairs utilized concept maps to elucidate psychological concepts, and the analysis employed a 2 × 2 between-subject factorial design. Based on the notions derived in previous studies on metacognitive reflections and learning-by-teaching, we examined two agent types: a virtual teacher (pedagogical agent) providing metacognitive reflections through scripts (scripted condition) and a virtual student (teachable agent) generating knowledge from concept maps created collaboratively by learners (reciprocal condition).

Results

Although both scripted and reciprocal teaching improved learning performance, scripted learning further enhanced the collaborative process; however, the effectiveness of combining these methods was only partially evident. A synergistic effect emerged through the generation of concept-map links, suggesting that mutual learning with reciprocal interactions among virtual students is as beneficial as scripted methods. The integration of these methods further enhances the learning process.

Conclusions

This study provides valuable insights for advancing the use of artificial intelligence in developing collaborative-learning support systems, particularly for integrating different scaffolding methods.