Special-R1: disability-aware pedagogical RL coupling a two-dimensional adaptive prompt with a persona-aware Thinking Reward

Design · All designs
Evidence · 3 claims (3 for) · 1 study (1 design), q2 · 0 of 1 report an effect size · 3 claims rest on one study

Description

Special-R1 is a framework extending pedagogical multi-turn RL to special education. It factors pedagogical support into two orthogonal dimensions: a support level (amount) mapped from the learner's baseline accuracy to Maximum, Moderate, or Minimum support following the I-Do / We-Do / You-Do progression, and a support style (manner) drawn from the learner's disability profile, so that "thewhatcomes from the support level, thehowcomes from the disability profile." Its second component is a persona-aware Thinking Reward whose judge is prompted with the learner's disability profile and scores whether the thinking trace reflects disability-appropriate reasoning, which the authors describe as "the first reward we are aware of that explicitly scoresdisability-appropriatereasoning."

Design Implications

Context

Requirements

• Requires a pre-measured baseline accuracy of the student simulator on the given problem type to select the support level, and a structured disability rubric with learning characteristics, communication patterns, and recommended tutor strategies.

Constraints

• Every student in the framework is a disability-conditioned LLM simulator; improvements on judge-based metrics have not been validated with real learners with disabilities.
• Experiments use 7-8B tutors; behavior at larger scales remains to be studied.

Target Learners

• Learners with disabilities across five profiles: intellectual disability, specific learning disability, autism spectrum disorder, ADHD, and emotional/behavioral disorder

Learning Goals

• Multi-domain tutoring covering mathematics, reading comprehension, social reasoning, and life skills

Claims

• Special R1 Raises Persona Fit And Sped Helpfulness [+M]
• Thinking Reward Effective Only With Adaptive Prompting [+M]
• Sped Adaptive Shifts Tutor Behavior Codebooks [+M]

Related Designs

Examples

-

Key Sources

• Unggi Lee, Jihoi Na, Yeil Jeong, Haeun Park, Yeonju Jang. (2026). Reinforcement Learning for Special Education: Aligning LLM Tutors to Diverse Learners through Disability-Adaptive Training. https://arxiv.org/abs/2605.30670

Page metadata

Field	Value
type	design
id	special-r1-framework
title	Special-R1: disability-aware pedagogical RL coupling a two-dimensional adaptive prompt with a persona-aware Thinking Reward
description	Special-R1 is a framework extending pedagogical multi-turn RL to special education.
status	review
generated	by process:settle-candidates on 2026-10-11
trust	unverified — no one has confirmed this page's content
sources	1 — unggi-lee-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.