LLM-Driven Adaptive Pictography: A Novel Framework for Personalized Chinese Character Learning

Abstract

The acquisition of Chinese characters (Hanzi) presents a significant challenge for learners worldwide, from young native speakers to second language (L2) adults, primarily because rote memorization is ineffective. This paper introduces the LLM-Driven Adaptive Pictography (LLMDAP) framework, a novel AI-powered system that transforms Hanzi learning into an active, creative, and highly personalized process. LLMDAP leverages a three-stage pipeline. The first stage, Visual Capture, allows learners to sketch concepts using pictorial cues. The second stage, Multimodal Mapping, uses a heuristic/AI-based recognizer to map sketches to characters and trigger learning content. The last stage is the Personalized Reinforcement, where the system dynamically generates a multimodal poster showcasing the character’s etymological chain and provides text-to-speech feedback. We present the complete system architecture and a fully implemented proof-of-concept that demonstrates the technical feasibility of this approach for a core set of pictographic characters. Our work contributes a novel framework grounded in cognitive theory, a functional prototype, and a detailed technical walkthrough. The LLMDAP framework marks a significant step towards highly personalized, AI-powered language education for a global audience, and we provide a clear roadmap for its future expansion and empirical validation.