Visual grounding in ParaTutor increased children's reasoning turns during math word problem tutoring compared to conventional and general LLM conditions¶

Claim · All claims
Evidence · 1 study · causal r2 · q2
Currency · newest study 2024 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Children produced more reasoning turns in ParaTutor modes (Mode C and Mode D) than in conventional tutoring (Mode A) and general LLM tutoring (Mode B). → LUO Lan 2024

Evidence¶

LUO Lan 2024¶

LUO Lan, WANG Anqi, ZHOU Muzhi, ZHU Junhua, CAI Jie, YU Ao, and PAN Hui. (2024). ParaTutor: Coordinating Parent-Child Math Tutoring through Role-Separated LLM Scaffolding. https://arxiv.org/abs/2606.18030 doi:10.48550/arxiv.2606.18030

q2 · i? · causal · r2

Interaction coding of 23 parent-child dyads across four counterbalanced tutoring conditions in a Latin Square design. Dyad-by-mode counts aggregated across three problems per condition showed children "produced more reasoning turnsin Mode C" and Mode D than in Mode A and Mode B. No effect size or significance test reported.

"Using dyad-by-mode counts aggregated across the three problems in each condition, childrenproduced more reasoning turnsin Mode C (M= 15.52,SD= 3.13) and Mode D (M= 16.83,SD= 3.39) than in Mode A (M= 8.61,SD= 2.79) and Mode B (M= 9.61,SD= 2.80)."

Discussion¶

Related Claims¶

• Child-initiated questions increased in ParaTutor modes compared to baseline tutoring conditions — related
• Interactive visual grounding helped children interpret word problems by making hidden relational constraints manipulable and visible — related
• ParaTutor's role-separated scaffolding distributed support across parents and children in complementary forms — related
• Three recurring challenges disrupt parent-child math tutoring: cognitive misalignment, emotional escalation, and method mismatch — related