Large language models (LLM) significantly advance data augmentation research. However, existing approaches largely overlook two issues: first, empirical evidence that LLM natural language (LLMNL) aligns with human natural language (HNL) remains insufficient, which is a foundational question; second, current methodologies often neglect the variability among LLM-generated texts, potentially constraining the effectiveness of data augmentation. To address the gap, we introduce a comprehensive scaling-law-based framework for examining the congruence between LLMNL and HNL. Through extensive experiments, we uncover a progression of findings: LLMNL fails to achieve congruence with HNL; there is a consistent discrepancy, with Mandelbrot exponents for LLMNL being approximately 0.2 lower than those of HNL; LLMNL exhibits reduced fractal complexity, corroborated our analysis to stylistic factors such as readability, sentiment, and semantics. Furthermore, we propose a new data augmentation approach for text classification, which leverages scaling laws to make decisions on LLM-generated texts. Extensive experiments under real-world scenarios demonstrate the competitiveness and robustness of the approach, outperforming recent methods and consistently maintaining performance advantages across varying LLMs and prompts.
Wang et al. (Mon,) studied this question.