The Truncation Blind Spot: How Decoding Strategies Systematically Exclude Human-Like Token Choices
This addresses the problem of machine text detectability for AI safety and natural language processing, revealing a fundamental mismatch between human and machine token selection, though it is incremental in analyzing existing methods.
The study found that standard decoding strategies for text generation systematically exclude 8-18% of human-selected tokens due to reliance on likelihood, making machine-generated text more detectable by simple classifiers.
Standard decoding strategies for text generation, including top-k, nucleus sampling, and contrastive search, select tokens based on likelihood, restricting selection to high-probability regions. Human language production operates differently: tokens are chosen for communicative appropriateness rather than statistical frequency. This mismatch creates a truncation blind spot: contextually appropriate but statistically rare tokens remain accessible to humans yet unreachable by likelihood-based decoding. We hypothesize this contributes to the detectability of machine-generated text. Analyzing over 1.8 million texts across eight language models, five decoding strategies, and 53 hyperparameter configurations, we find that 8-18% of human-selected tokens fall outside typical truncation boundaries. Simple classifiers trained on predictability and lexical diversity achieve remarkable detection rates. Crucially, neither model scale nor architecture correlates strongly with detectability; truncation parameters account for most variance. Configurations achieving low detectability often produce incoherent text, indicating that evading detection and producing natural text are distinct objectives. These findings suggest detectability is enhanced by likelihood-based token selection, not merely a matter of model capability.