Noise Steering for Controlled Text Generation: Improving Diversity and Reading-Level Fidelity in Arabic Educational Story Generation

TL;DR

This paper introduces noise steering, a training-free method that injects Gaussian noise into transformer models to enhance diversity in Arabic educational story generation while maintaining reading level fidelity.

Contribution

It demonstrates that internal representation noise injection improves diversity and stability in constrained Arabic story generation without retraining models.

Findings

Residual stream noise enhances narrative diversity with minimal quality loss.

Attention entropy noise injection stabilizes attention and recovers quality.

High-temperature sampling increases reading level and causes model collapse.

Abstract

Generating diverse, pedagogically valid stories for Arabic early-grade reading assessments requires balancing tight constraints on vocabulary, reading level, and narrative structure against the need to avoid repetitive plots that undermine assessment validity. We investigate noise steering, injecting calibrated Gaussian perturbations into the internal representations of transformer models at inference time, as a training-free diversity method evaluated across five small Arabic-centric language models (7-9B parameters). We compare four injection strategies against high-temperature sampling baselines, measuring diversity, quality, constraint adherence, and reading grade level. Residual stream noise consistently improves narrative diversity with minimal quality or constraint cost and preserves early-grade reading level across all Arabic-centric models. Attention entropy noise injection (AENI) stabilizes the otherwise unreliable attention-logit noise while recovering quality. High-temperature sampling inflates reading grade level and causes catastrophic collapse on several models. We find internal representation-level perturbation to be a more suitable diversity strategy than output-level stochasticity for constrained educational content generation.