TACT: Taxonomy-Aligned Post-Training for Pedagogically Adaptive English Tutoring

Synthesis: TACT (Taxonomy-Aligned Conversational Tutor) presents a human-grounded framework for training and evaluating pedagogically adaptive ESL tutors powered by LLMs. Built on a Tutor-Strategy Taxonomy (13 strategies) and a Student-Move Taxonomy, TACT produces TACTutor — a model that improves over its Qwen3.5-4B backbone by 20.30% on a strategy-balanced benchmark and outperforms all evaluated proprietary baselines. The framework represents a significant advance in pedagogically grounded AI tutoring, moving beyond generic response generation toward truly adaptive instructional dialogue.

Framework Design

TACT's architecture rests on two complementary taxonomies drawn from human-tutoring research:

Tutor-Strategy Taxonomy (13 strategies):

Strategies span from direct explanation and modeling to guided discovery and Socratic questioning

Each strategy maps to specific pedagogical goals and learner states

Student-Move Taxonomy:

Characterizes learner utterances by move type (question, attempt, confusion signal, etc.)

Tracks correctness status to inform strategy selection

TACTCorpus: 260 authentic teacher-student conversations enriched with 32,379 annotations and augmented training data.

Training Pipeline

TACTutor is produced through a two-stage post-training process:

1. Supervised Fine-Tuning (SFT): Trains the model on taxonomy-annotated tutoring dialogues

2. Taxonomy-Aligned Group Relative Policy Optimization (GRPO): Optimizes for scaffolding quality using strategy-aligned reward signals, rather than simple reference imitation

This approach ensures the model learns when and why to apply specific strategies, not just what to say.

Key Findings

Metric	Improvement
TACTBench performance vs. backbone	+20.30%
vs. proprietary baselines (same protocol)	Outperforms all
Blinded learner study (N=50)	Highest overall mean rating
External educational benchmarks	Maintains backbone performance

Strategy-balanced evaluation: TACTBench contains 78 authentic tutoring contexts balanced across all 13 tutor strategies

Generalization: Maintains backbone performance on established external benchmarks — no regression

Open release: Data, benchmark, and model weights publicly available, providing an open foundation for pedagogically adaptive tutoring

Implications for AI Tutoring

TACT advances the state of AI tutoring by grounding LLM training in established pedagogical theory rather than treating tutoring as generic dialogue generation. The taxonomy-aligned training approach could be adapted to other domains (math, science, programming) and other languages, offering a principled path toward more effective and pedagogically safe AI tutors.

Connected Concepts

LLM

Scaffolding

Intelligent Tutoring

Pedagogical LLM Training

AI Tutoring

Pedagogical Safety

Connected Articles

Learnmate2 LLM Adaptive Learning — LearnMate^2: Design and Evaluation of an LLM-powered Personalized and Adaptive Support System for Online Learning

LLM Educational Simulation Adhd — LLM-Based Educational Simulation: Evaluating Temporal Student Persona Stability Across ADHD Profiles

Hazra Safetutors Pedagogical Safety 2026 — SafeTutors: Pedagogical Safety in AI Tutoring

Conversational AI Tutors Framework — The Path to Conversational AI Tutors: Integrating Tutoring Best Practices and Targeted Technologies to Produce Scalable AI Agents

Xie Hillm Cd 2026 — HiLLM-CD: LLM-Enhanced Hierarchical Cognitive Diagnosis

Codify Socratic Tutoring Programming — Codify: An Intelligent Socratic Tutoring System for Programming Education

Citation

Yang, D., Lin, S., Shen, L., Sheng, R., Qu, H., & Chen, Z. (2026). TACT: Taxonomy-Aligned Post-Training for Pedagogically Adaptive English Tutoring. arXiv:2608.03952v1.