Data Scaling as Progressive Coverage of a Predictive Contribution Spectrum

In three linesarXiv paper showing that real-data scaling laws are governed by progressive coverage of a latent predictive contribution spectrum, beyond token-frequency tails alone. Using suffix-automaton representation, authors define a global-KL spectrum and demonstrate strong correlation (R²≈0.96) between spectrum slope and empirical scaling exponent across 12 corpora.

Summary generated by Claude — human-verified