Saved in:
Bibliographic Details
Main Authors: Awad, Samer, Conde, Javier, Arriaga, Carlos, Fu, Tairan, Coronado-Blázquez, Javier, Reviriego, Pedro
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.27268
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910261290467328
author Awad, Samer
Conde, Javier
Arriaga, Carlos
Fu, Tairan
Coronado-Blázquez, Javier
Reviriego, Pedro
author_facet Awad, Samer
Conde, Javier
Arriaga, Carlos
Fu, Tairan
Coronado-Blázquez, Javier
Reviriego, Pedro
contents Modern Large Language Models (LLMs) are often criticized for producing repetitive and homogeneous text, despite possessing vast latent vocabularies. While previous research has focused on model knowledge and training data, we investigate the role of decoding mechanics in suppressing linguistic diversity. We introduce the Word Coverage Score (WCS), a metric that quantifies the extent to which contextually appropriate human vocabulary is mathematically pruned by standard sampling filters (e.g., Top-$p$, Top-$k$, and Min-$p$). Rather than assessing static knowledge, the WCS measures the lexical survival rate of low-frequency, high-information human words as a function of sampling parameters. By auditing open-weight models on human-authored corpus fragments, we identify which logical lexical choices are rendered unreachable by the decoder, even when they reside within the probability space. Our results provide quantitative evidence that industry-standard sampling defaults act as unintended censorship mechanisms, smoothing the unique textures of human expression into a homogenized discourse. The WCS offers a rigorous framework for optimizing the trade-off between text coherence and lexical richness, providing a diagnostic tool for preserving the diversity of human language in generative models.
format Preprint
id arxiv_https___arxiv_org_abs_2605_27268
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Lost in Sampling: Assessing Lexical Reachability in LLMs via the Word Coverage Score (WCS)
Awad, Samer
Conde, Javier
Arriaga, Carlos
Fu, Tairan
Coronado-Blázquez, Javier
Reviriego, Pedro
Computation and Language
Artificial Intelligence
Modern Large Language Models (LLMs) are often criticized for producing repetitive and homogeneous text, despite possessing vast latent vocabularies. While previous research has focused on model knowledge and training data, we investigate the role of decoding mechanics in suppressing linguistic diversity. We introduce the Word Coverage Score (WCS), a metric that quantifies the extent to which contextually appropriate human vocabulary is mathematically pruned by standard sampling filters (e.g., Top-$p$, Top-$k$, and Min-$p$). Rather than assessing static knowledge, the WCS measures the lexical survival rate of low-frequency, high-information human words as a function of sampling parameters. By auditing open-weight models on human-authored corpus fragments, we identify which logical lexical choices are rendered unreachable by the decoder, even when they reside within the probability space. Our results provide quantitative evidence that industry-standard sampling defaults act as unintended censorship mechanisms, smoothing the unique textures of human expression into a homogenized discourse. The WCS offers a rigorous framework for optimizing the trade-off between text coherence and lexical richness, providing a diagnostic tool for preserving the diversity of human language in generative models.
title Lost in Sampling: Assessing Lexical Reachability in LLMs via the Word Coverage Score (WCS)
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2605.27268