Can LLMs Control Readability? A Multi-Dimensional Evaluation Framework for CEFR-Controlled Arabic Generation

2026-06-20

AI Summary

This study addresses the challenge that large language models (LLMs) face in effectively controlling the readability of Arabic text aligned with the Common European Framework of Reference for Languages (CEFR), which limits their utility in adaptive language learning. To tackle this, the authors propose a multidimensional readability assessment framework that integrates CEFR-guided generation, lexical constraints, and syntactic complexity analysis. Combining the Taha-19 automated readability prediction model with structured prompt engineering, they systematically evaluate LLMs’ controllability. Experimental results demonstrate that the proposed CEFR-guided prompting strategy with lexical constraints significantly enhances generation controllability, achieving a corpus-profile cosine similarity of 0.91 and a readability prediction consistency of 0.99—substantially outperforming unconstrained generation. These findings underscore the critical role of prompt design in regulating text readability.

Abstract

While Large Language Models (LLMs) can generate fluent Arabic text, their ability to reliably control readability levels remains unclear. We propose a multi-dimensional evaluation framework for Common European Framework of Reference for Language (CEFR)-controlled Arabic text generation, assessing whether instruction-following LLMs can serve as reliable generators for adaptive language learning. Our framework integrates controlled prompting, automatic readability prediction using a validated Taha-19 model, lexical constraint validation, and syntactic complexity profiling. Results show that structured prompting substantially improves CEFR alignment. In particular, CEFR-guided prompting with lexical constraints achieves the highest conformity to reference linguistic profiles (0.91 cosine similarity) and near-perfect agreement with predicted readability levels (0.99), while unconstrained prompting exhibits weak control. These findings establish an empirical foundation for integrating readability-aware Arabic text generation into adaptive educational systems.

Problem

• readability control
• CEFR
• Arabic text generation
• Large Language Models
• adaptive language learning

Innovation

• CEFR-controlled generation
• readability prediction
• controlled prompting
• lexical constraints
• Arabic text generation