Who Gets a Token, and What Does It Carry? Unequal Name Support and Concept Access in Large Language Models

2026-09-28

AI Summary

This study addresses the implicit biases in large language models arising from uneven name tokenization, which leads to disparities in concept accessibility. To investigate this, we propose NameTrace, a framework introducing a novel model-native, fine-grained pre-behavioral measurement paradigm. By integrating tokenizer analysis, continuous weight probability estimation, and hidden state intervention techniques, it quantifies how lexical support shapes internal representations. This work establishes the principle that lexical comparability underpins behavioral comparability, elucidating the mechanism by which input-level inequities translate into task-specific biases. Furthermore, it demonstrates the persistence and transferability of cross-racial and cross-gender biases in scenarios such as recruitment, achieving precise measurement of lexical comparability.

Abstract

Names are personal identifiers, but they also carry social meaning and are widely used to evaluate how language models treat different people. Such evaluations typically assume that matched names are comparable model inputs. We show that this assumption often fails at the lexical interface: matched names are not necessarily matched inputs. Some names receive direct single-token access, while others are assembled from multiple subwords, creating unequal name-surface support. Across nearly half a million first names and 12 LLM-associated tokenizers, direct lexical access is highly selective, model dependent, and uneven across race- and gender-associated name metadata. We introduce NameTrace, a model-native, fine-grained, pre-behavioral framework for measuring whether unequal name-surface support remains a vocabulary property or becomes visible in task-relevant internal representations. NameTrace measures concept accessibility from the model's own probabilities over task-specific adjective axes with continuous task-aligned weights. On matched atomic and short-fragmented names within the same race/ethnicity--gender-associated strata, support predicts systematic differences in concept accessibility across fellowship, hiring, clinical assessment, and lending. These differences persist across all eight matched strata, extend across model families, and transfer to unseen names. Hidden-state interventions further show that the measured task directions have downstream leverage, shifting later constrained choices. Unequal lexical support is therefore demographically structured at the input and remains visible in task-relevant model computation. NameTrace makes lexical comparability measurable, supporting a broader principle: behavioral comparability begins with lexical comparability.

Problem

• tokenization bias
• name representation
• large language models
• fairness
• concept accessibility

Innovation

• tokenizer bias
• NameTrace
• concept accessibility
• hidden-state intervention
• lexical comparability