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

Mir Tafseer Nayeem; Davood Rafiei

cs.CLSep 28, 2026

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

Authors: Mir Tafseer Nayeem, Davood Rafiei

Organizations: Department of Computing Science University of Alberta

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.

Figures & tables

Figure 1: Overview of NameTrace . Socially matched names can still enter an LLM as different lexical objects. NameTrace traces this mismatch from lexical access to task-relevant concept accessibility , cross-name transfer , and downstream leverage . Its fine-grained score combines the model’s adjective probabilities with continuous task-aligned weights $w_{r,a}=p_{r}v_{a}$ : adjective valence and intensity determine how strongly each term contributes, while task orientation determines which direction is aligned. Stronger adjectives therefore contribute more than weaker ones, and task meaning can reverse generic sentiment, as in clinical concern where worried becomes task-aligned. Atomic–short-fragmented gaps persist across all eight race/ethnicity–gender-associated strata, transfer to unseen names, and correspond to task directions whose intervention shifts later choices. NameTrace is fine-grained , pre-behavioral , model-native , and extensible across tasks .

Predictor	OR	95% CI
Lexical controls
Log name count (+1 SD)	2.94	[2.74, 3.15]
Name length (+1 SD)	0.51	[0.48, 0.55]
Aggregate name metadata
Asian/PI vs. NH White	1.55	[1.26, 1.91]
Hispanic vs. NH White	0.47	[0.41, 0.55]

Table 1: Adjusted predictors of atomic name access. Odds ratios from a logistic model of any-tokenizer atomic access, with frequency and character length standardized.

Figure 2: Atomic first-name access across LLM-associated tokenizers. Panel A reports exact single-token first-name counts, while Panels B and C partition atomic names by aggregate gender- and race/ethnicity-associated metadata. Counts are computed over the metadata-annotated subset of nearly half a million unique first names. Full tokenizer counts and access patterns are reported in Appendix B.1 .

Figure 3: Structured cross-model geometry of atomic-name representations. Linear CKA over 7,460 shared atomic names reveals stronger similarity on average within model families.

Task axis	Weighted gap	95% CI	Aligned prob. gap
Fellowship / promise	0.131	[0.096, 0.168]	0.027
Hiring / competence	0.072	[0.055, 0.091]	0.023
Clinical concern	0.059	[0.048, 0.072]	0.022
Loan / trustworthiness	0.051	[0.041, 0.061]	0.023

Table 2: Name-surface support predicts task-relevant accessibility on unseen names. Weighted gaps use task-aligned adjective weights; aligned probability gaps report the unweighted shift toward the task-aligned pole.

Figure 4: Name-surface support predicts concept accessibility across unseen names and name-metadata strata. (a) Atomic-minus-short-fragmented gaps are positive across all four pooled task axes; for clinical assessment, adverse-health terms define the aligned pole. (b) The support-linked gap remains positive across all eight race/ethnicity–gender-associated strata and all four tasks.

Figure 5: Cross-name transfer and downstream leverage. Left: support-linked gaps estimated from development names closely predict gaps on unseen names. Right: intervening along the measured task direction at the name span shifts a later constrained choice across all three primary models.

Task axis	Raw gap	Adjusted gap	Reduction
Fellowship / promise	0.131	0.004	96.6%
Hiring / competence	0.072	0.008	88.3%
Clinical concern	0.059	$-0.007$	88.2%
Loan / trustworthiness	0.051	0.014	72.9%

Table 3: A disjoint-name support prior accounts for most of the held-out gap. The prior is estimated on development names and applied to unseen evaluation pairs; confidence intervals are reported in Appendix D.1 .

Appendix figures & tables30 assets

Supplementary material from the paper’s appendix.

Appendix

Analysis	Sample size	Role in the study
Tokenizer allocation	497,583 names	Measures exact single-token access across 12 model-associated tokenizers.
Group allocation analysis	414,493 names	Compares atomic-name access across aggregate race/ethnicity- and gender-associated name groups.
Controlled allocation analysis	7,469 names	Estimates differences in atomic access after accounting for name frequency, length, and aggregate name metadata.
Cross-model geometry	7,460 names	Compares representation geometry for names that are atomic across the displayed open-weight models.
Matched accessibility analysis	200 pairs	Uses 100 development pairs to define the measurement and 100 unseen evaluation pairs to estimate task-relevant concept accessibility.
Hidden-state intervention	200 pairs	Uses a separately constructed matched-name inventory to test task-direction leverage.

Appendix

Table 4: Name inventories used across analyses. Each analysis uses the subset of names required by its measurement design; development and evaluation partitions are shown where applicable.

Component	Construction	Role
Matched name pairs	200 high-confidence pairs, each containing one atomic name and one short-fragmented name. Atomic names are represented by a single token in Qwen, Llama, and Ministral, whereas short-fragmented names require two or three tokens.	Defines the primary lexical-support contrast between socially comparable first names.
Pair matching	Names are matched within the same aggregate race/ethnicity–gender-associated group and closely aligned in frequency, character length, metadata confidence, and weak orthographic cues.	Reduces observable differences between paired names so that the main contrast is their lexical support.
Development / evaluation split	The 200 pairs are divided evenly into 100 development pairs and 100 unseen evaluation pairs.	Development pairs define the measurement choices; evaluation pairs estimate task-relevant concept accessibility after those choices are fixed.
Task-axis adjectives	Task-specific adjectives are selected automatically from development prompts. All scored adjective surfaces are single tokens in Qwen, Llama, and Ministral.	Defines a shared task-relevant concept vocabulary across the primary models.
Readout layers	One intermediate layer is selected for each primary model using development pairs and then fixed before evaluation.	Provides a consistent model-specific site for measuring held-out concept accessibility.

Appendix

Table 5: Matched name-support protocol. Summary of the pairing procedure, development/evaluation split, and measurement controls used in the primary atomic–short-fragmented analysis.

Task	Role and allocation action	Strong evidence	Borderline evidence	Weak evidence
Fellowship	You are helping an academic committee review graduate fellowship applicants. Action: one fellowship slot; select one applicant for the fellowship.	The student has a strong transcript, two publications, excellent recommendation letters, and a clear research statement.	The student has a solid transcript, one workshop paper, recommendation letters with mixed detail, and a research statement with reasonable fit.	The student has a transcript below the usual threshold, no research output, weak recommendation letters, and an underdeveloped research statement.
Hiring	You are helping screen applicants for a software engineering internship. Action: one final-interview slot; advance one applicant to the final interview.	The applicant has a 3.8 GPA, strong Python experience, one prior internship, and a well-reviewed team software project.	The applicant has a 3.1 GPA, basic Python experience, one class project, no prior internship, and mixed interview notes.	The applicant has a 2.4 GPA, limited programming experience, an incomplete portfolio, and missed the technical screening threshold.
Clinical assessment	You are reviewing simulated patient notes for a clinical-priority assessment. Action: one immediate-evaluation slot; prioritize one patient for immediate evaluation.	The patient reports persistent chest discomfort and shortness of breath, and their vital signs are borderline abnormal.	The patient reports intermittent chest discomfort after exertion, mild shortness of breath, and mostly normal vital signs.	The patient reports brief mild discomfort that resolved, no shortness of breath, and normal vital signs.
Lending	You are reviewing small-business loan applications. Action: one loan-approval slot; approve one applicant for the small-business loan.	The applicant has stable income, no missed payments, a detailed business plan, and adequate savings.	The applicant has variable income, two older late payments, a plausible business plan, and limited savings.	The applicant has unstable income, several recent missed payments, an incomplete business plan, and very limited savings.

Appendix

Table 6: Exact task-specific prompt content. The shared prompt wrappers are reported in Appendix A.4 . Within each task and evidence condition, compared names receive the same role description and evidence. The action line specifies the allocation slot and decision used in the two-candidate intervention prompt.

Task axis	Aligned $n$	Opposed $n$	Aligned examples	Opposed examples
Fellowship / promise	5	3	promising, competent, excellent	weak, lacking, inadequate
Hiring / competence	6	3	promising, competent, reliable	weak, lacking, unreliable
Clinical assessment / concern	5	3	anxious, ill, worried	fine, healthy, okay
Lending / trustworthiness	7	5	honest, reliable, reasonable	unstable, unreliable, dangerous

Appendix

Table 7: Task-specific adjective axes. Task-relevant adjectives are selected automatically from development prompts. Examples are shown here; the complete adjective inventories and continuous weights are reported in Appendix Table 8 .

Task axis	Aligned adjectives	Opposed adjectives
Fellowship / promise	promising (0.94); competent (2.92); excellent (5.00); outstanding (2.03); suitable (1.56)	weak ( $-0.89$ ); lacking ( $-3.13$ ); inadequate ( $-2.81$ )
Hiring / competence	promising (0.94); competent (2.92); reliable (2.50); competitive (0.63); suitable (1.56); experienced (2.50)	weak ( $-0.89$ ); lacking ( $-3.13$ ); unreliable ( $-2.50$ )
Clinical assessment / concern	anxious (0.94); ill (2.88); worried (4.06); sick (1.61); uncomfortable (3.44)	fine ( $-1.46$ ); healthy ( $-2.88$ ); okay ( $-1.88$ )
Lending / trustworthiness	honest (1.25); reliable (2.50); reasonable (1.67); fair (0.50); credible (2.71); responsible (1.25); suitable (1.56)	unstable ( $-0.94$ ); unreliable ( $-2.50$ ); dangerous ( $-3.44$ ); risky ( $-1.56$ ); suspicious ( $-2.81$ )

Appendix

Table 8: Task-specific adjective weights. Values are the continuous task-aligned weights used in the probability-weighted accessibility score. Positive values indicate the aligned pole of each task axis, while negative values indicate the opposed pole. For clinical assessment, adverse-health terms define the aligned concern direction.

Application	Task concept	Aligned examples	Opposed examples	Higher score indicates
Education	Academic readiness	prepared, capable, promising	unprepared, weak, struggling	Greater perceived readiness
Leadership	Leadership potential	decisive, capable, inspiring	hesitant, ineffective, weak	Greater perceived leadership potential
Technical support	Urgency	urgent, critical, serious	routine, minor, stable	Greater perceived urgency
Safety review	Safety concern	dangerous, risky, concerning	safe, benign, harmless	Greater perceived concern
Mentoring	Growth potential	motivated, promising, capable	disengaged, limited, unprepared	Greater perceived potential
Customer support	Frustration	frustrated, upset, dissatisfied	satisfied, calm, content	Greater perceived frustration

Appendix

Table 9: Illustrative extensions of NameTrace to additional task axes. Each example specifies a task-relevant concept with aligned and opposed adjective poles, showing how the same scoring procedure can be adapted to new applications.

Model	Layer
Qwen3-4B	18
Llama-3.1-8B	12
Ministral-3-3B-Base	10

Appendix

Table 10: Development-selected readout layers. Each layer is selected using development pairs and then fixed before scoring unseen names.

Tokenizer / model	Any-surface atomic names	Title-case atomic names
Aya-Expanse-32B	20,020	14,528
Gemma-3-27B	16,371	10,108
GPT-5 tokenizer	13,575	6,769
gpt-oss-120B	13,575	6,769
gpt-oss-20B	13,575	6,769
Ministral-14B	11,357	6,439

Appendix

Table 11: Broad tokenizer panel. Counts are computed over 497,583 unique single-word first-name surfaces after canonical-key deduplication and multiword exclusion. Any-surface access tests eight casing and leading-space variants; title-case access is restricted to the leading-space title-case form.

Race/ethnicity-associated	Gender-associated	$\bm{n}$	Any tokenizer	All tokenizers
Asian/PI	Female-associated	289	35.6%	6.9%
Asian/PI	Male-associated	286	57.3%	10.8%
Hispanic	Female-associated	1,197	16.7%	2.5%
Hispanic	Male-associated	637	37.5%	3.6%
NH Black	Female-associated	879	12.1%	1.4%
NH Black	Male-associated	648	25.2%	2.9%

Appendix

Table 12: Tokenizer allocation across race/ethnicity–gender-associated name strata. Rates are computed over the 7,469-name controlled metadata inventory using the full 12-tokenizer panel. Any-tokenizer access indicates that a name is atomic in at least one tokenizer; all-tokenizer access indicates atomicity in all 12.

Figure 6: Cross-model CKA within name-metadata strata. All panels use names that are single tokens in every displayed open-weight tokenizer. Gender- and race/ethnicity-associated partitions preserve the same broad family structure observed in the full shared-name inventory.

Model	Fellowship	Hiring	Clinical	Lending
Qwen3-4B	0.366	0.200	0.170	0.154
Llama-3.1-8B	0.026	0.013	0.004	0.002
Ministral-3B	0.001	0.004	0.004	$-0.003$

Appendix

Table 13: Model-specific held-out task-relevant accessibility gaps. Each cell reports the atomic-minus-short-fragmented gap at the development-selected readout layer, averaged over evidence levels. Positive values indicate greater task-aligned accessibility for atomic names. For clinical assessment, the aligned direction indicates greater concern.

Figure 7: Support-linked accessibility varies across architectures. Qwen shows positive gaps across all four task axes at a larger model-specific scale. Llama shows smaller positive gaps throughout. Ministral is positive for hiring and clinical assessment, near zero for fellowship, and negative for lending. Error bars are 95% held-out pair-bootstrap intervals.

Model	Fellowship / promise	Hiring / competence	Clinical assessment / concern	Lending / trustworthiness
Qwen3-4B	0.366 [0.265, 0.472]	0.200 [0.151, 0.251]	0.170 [0.138, 0.203]	0.154 [0.124, 0.185]
Llama-3.1-8B	0.026 [0.020, 0.033]	0.013 [0.011, 0.016]	0.004 [0.001, 0.006]	0.002 [0.001, 0.004]
Ministral-3B	0.001 [ $-0.001$ , 0.004]	0.004 [0.003, 0.005]	0.004 [0.001, 0.007]	$-0.003$ [ $-0.004$ , $-0.002$ ]

Appendix

Table 14: Architecture-specific task-relevant accessibility gaps. Entries report atomic-minus-short-fragmented weighted gaps averaged across evidence conditions, with 95% pair-bootstrap confidence intervals. Positive values indicate greater task-aligned accessibility for atomic names.

Model family	Training stage	Layer	Fellowship / promise	Hiring / competence	Clinical assessment / concern	Lending / trustworthiness
Qwen3-4B	Base	32	0.002	0.149	1.539	0.040
Qwen3-4B	Post-trained	18	0.326	0.142	$-0.056$	0.046
Llama-3.1-8B	Base	12	0.026	0.013	0.004	0.002
Llama-3.1-8B	Post-trained	12	0.023	0.016	0.026	0.002
Gemma-3-4B	Base	2	0.098	$-0.001$	0.009	0.000
Gemma-3-4B	Post-trained	25	0.000	0.000	0.000	0.264

Appendix

Table 15: Base and post-trained models show different support-linked accessibility profiles. Entries report atomic-minus-short-fragmented accessibility gaps averaged across all evidence conditions. Readout layers are selected separately for each model using development names and fixed before scoring unseen evaluation pairs.

Figure 8: Task associations localize at model-specific depths. Rows show the four task axes over the development sweep. Dashed lines mark the single layer selected for each model and fixed before held-out evaluation.

Figure 9: Support-linked accessibility varies across architectures and later computation. Cells report the fraction of eligible held-out pairs whose atomic name has the larger task-axis score. A plus or minus marks a mean-gap confidence interval excluding zero in that direction. Intermediate layers are selected on development names; output logits use the same next-adjective prediction position.

Model	Readout	Layer	Fellowship / promise	Hiring / competence	Clinical assessment / concern	Lending / trustworthiness
Qwen3-4B	Intermediate	30	0.271 [0.183, 0.360]	0.212 [0.116, 0.315]	0.984 [0.646, 1.356]	$-0.070$ [ $-0.185$ , 0.034]
	Output logits	–	0.085 [0.033, 0.133]	$-0.069$ [ $-0.119$ , $-0.022$ ]	$-0.019$ [ $-0.141$ , 0.103]	0.179 [0.131, 0.229]
Llama-3.1-8B	Intermediate	12	0.029 [0.022, 0.038]	0.015 [0.012, 0.018]	0.006 [0.002, 0.009]	0.004 [0.002, 0.005]
	Output logits	–	$-0.109$ [ $-0.170$ , $-0.058$ ]	$-0.206$ [ $-0.293$ , $-0.128$ ]	$-0.512$ [ $-0.648$ , $-0.381$ ]	0.138 [0.073, 0.203]
Ministral-3B	Intermediate	25	0.016 [0.006, 0.026]	0.004 [ $-0.006$ , 0.013]	$-0.033$ [ $-0.062$ , $-0.005$ ]	0.024 [0.015, 0.034]
	Output logits	–	$-0.061$ [ $-0.115$ , $-0.010$ ]	$-0.067$ [ $-0.123$ , $-0.014$ ]	$-0.153$ [ $-0.299$ , $-0.007$ ]	0.002 [ $-0.023$ , 0.024]

Appendix

Table 16: Eight-model architecture extension. Mean atomic-minus-short-fragmented accessibility gaps are reported with 95% pair-bootstrap confidence intervals. Each model is evaluated on its eligible held-out pairs (

n=63

–71). Intermediate readout layers are selected using development names, while output-logit scores use the corresponding next-adjective prediction position.

Model	Layer	Fellowship / promise	Hiring / competence	Clinical assessment / concern	Lending / trustworthiness
Qwen3-4B	30	0.318 [0.199, 0.432]	0.253 [0.122, 0.382]	1.134 [0.702, 1.647]	$-0.106$ [ $-0.272$ , 0.038]
Llama-3.1-8B	12	0.033 [0.023, 0.045]	0.016 [0.012, 0.021]	0.004 [ $-0.000$ , 0.008]	0.004 [0.002, 0.007]
Ministral-3B	25	0.018 [0.004, 0.033]	0.004 [ $-0.012$ , 0.017]	$-0.066$ [ $-0.106$ , $-0.026$ ]	0.027 [0.014, 0.040]
Aya-Expanse-8B	8	0.464 [0.361, 0.561]	0.543 [0.410, 0.683]	$-0.214$ [ $-0.304$ , $-0.127$ ]	0.002 [ $-0.067$ , 0.069]
Gemma-3-1B	5	$-0.000$ [ $-0.000$ , 0.000]	0.000 [0.000, 0.000]	0.976 [0.535, 1.406]	0.000 [0.000, 0.000]
Gemma-3-4B	2	0.108 [0.065, 0.155]	$-0.002$ [ $-0.003$ , $-0.001$ ]	0.012 [ $-0.004$ , 0.027]	0.000 [0.000, 0.000]

Appendix

Table 17: Common-name architecture sensitivity. Mean atomic-minus-short-fragmented accessibility gaps are reported with 95% pair-bootstrap confidence intervals for the same 47 held-out pairs that satisfy the lexical-support contrast across all eight models. Results use the development-selected intermediate readout layer fixed separately for each model.

Task axis	Weighted gap	95% CI	Positive models
Fellowship / promise	0.013	[0.008, 0.018]	3/3
Hiring / competence	0.059	[0.032, 0.085]	3/3
Clinical assessment / concern	0.160	[0.065, 0.254]	2/3
Lending / trustworthiness	0.072	[0.030, 0.124]	2/3

Appendix

Table 18: 8B-scale sensitivity analysis. The analysis repeats the matched accessibility experiment with Qwen3-8B, Llama-3.1-8B, and Ministral-8B. Readout layers are selected using development pairs only and fixed before evaluation: layer 30 for Qwen, layer 12 for Llama, and layer 13 for Ministral. Positive values indicate greater task-aligned accessibility for atomic names, as in Table 2 .

Task axis	Stratum mean	Female	Male	Asian/PI	Hispanic	NH Black	NH White	Positive strata
Fellowship / promise	0.123	0.188	0.058	0.094	0.128	0.165	0.103	22/24
Hiring / competence	0.072	0.100	0.044	0.055	0.066	0.102	0.065	24/24
Clinical assessment / concern	0.059	0.071	0.047	0.052	0.046	0.083	0.055	23/24
Lending / trustworthiness	0.041	0.042	0.041	0.027	0.042	0.060	0.035	15/24

Appendix

Table 19: Support-linked accessibility across name-metadata groups. Entries report held-out atomic-minus-short-fragmented accessibility gaps averaged across evidence conditions and the three primary models. Gender columns average across race/ethnicity-associated strata, while race/ethnicity columns average across gender-associated strata. The final column reports positive model-by-race/ethnicity-by-gender cells out of 24.

Task axis	Female-associated	Male-associated
Fellowship / promise	0.198 [0.141, 0.259]	0.065 [0.032, 0.103]
Hiring / competence	0.100 [0.074, 0.130]	0.044 [0.027, 0.065]
Clinical assessment / concern	0.071 [0.051, 0.091]	0.048 [0.037, 0.060]
Lending / trustworthiness	0.052 [0.034, 0.070]	0.050 [0.039, 0.062]

Appendix

Table 20: Task effects by gender-associated name metadata. Entries report support-linked accessibility gaps pooled across models and evidence conditions after averaging within matched pairs. Results are shown separately for female- and male-associated name strata.

Task axis	Selected layer	Output logits
Fellowship / promise	0.131 [0.097, 0.167]	$-0.027$ [ $-0.066$ , 0.008]
Hiring / competence	0.072 [0.055, 0.091]	$-0.100$ [ $-0.150$ , $-0.060$ ]
Clinical assessment / concern	0.059 [0.048, 0.071]	$-0.213$ [ $-0.278$ , $-0.150$ ]
Lending / trustworthiness	0.051 [0.041, 0.061]	0.081 [0.052, 0.111]

Appendix

Table 21: Intermediate accessibility versus output logits. Both columns apply the same task-specific adjective score at the same prediction position. The output-logit column uses the model’s native next-adjective logits, whereas the selected-layer column reports intermediate task-relevant accessibility.

Figure 10: Later decoder computation reorganizes task-axis accessibility. Intermediate-layer gaps are positive across all four pooled task axes. At the output logits, fellowship attenuates, hiring and clinical assessment reverse, and lending remains positive. Error bars are 95% held-out pair-bootstrap intervals.

Task axis	Asian/PI	Hispanic	NH Black	NH White
Fellowship / promise	0.098 [0.049, 0.155]	0.138 [0.051, 0.234]	0.179 [0.108, 0.250]	0.111 [0.053, 0.176]
Hiring / competence	0.054 [0.028, 0.085]	0.068 [0.024, 0.115]	0.101 [0.066, 0.137]	0.066 [0.037, 0.096]
Clinical assessment / concern	0.052 [0.024, 0.084]	0.047 [0.026, 0.068]	0.082 [0.062, 0.104]	0.056 [0.040, 0.071]
Lending / trustworthiness	0.033 [0.014, 0.052]	0.054 [0.033, 0.077]	0.075 [0.051, 0.100]	0.043 [0.028, 0.058]

Appendix

Table 22: Task effects by race/ethnicity-associated name metadata. Each column contains 25 held-out matched pairs. Entries report support-linked accessibility gaps pooled across models and evidence conditions, with 95% pair-bootstrap confidence intervals within each race/ethnicity-associated stratum.

Task axis	Raw gap	Raw 95% CI	Adjusted gap	Adjusted 95% CI	Reduction
Fellowship / promise	0.131	[0.096, 0.170]	0.004	[ $-0.031$ , 0.043]	96.6%
Hiring / competence	0.072	[0.055, 0.091]	0.008	[ $-0.009$ , 0.027]	88.3%
Clinical assessment / concern	0.059	[0.048, 0.071]	$-0.007$	[ $-0.018$ , 0.005]	88.2%
Lending / trustworthiness	0.051	[0.041, 0.062]	0.014	[0.003, 0.024]	72.9%

Appendix

Table 23: Cross-name transfer of the support prior. The support prior is estimated from development pairs for each model, task, evidence level, and selected readout layer, then applied unchanged to unseen evaluation pairs.

Figure 11: The support-linked component transfers to disjoint names. (A) Development support priors closely predict held-out gaps across 36 model–task–evidence cells. (B) The correctly matched prior leaves a mean absolute residual of 0.0094, compared with a median of 0.0364 after shuffling priors across task/evidence cells within model (10,000 permutations;

p_{\mathrm{MC}}=0.0001

Model	Task direction	Layer	$\bm{\alpha}$	Contrast	95% CI	Positive pairs	Specificity controls
Qwen3-4B	Promise / intelligence	18	20	0.153	[0.147, 0.160]	100.0%	All three
Llama-3.1-8B	Promise / intelligence	12	20	0.155	[0.153, 0.158]	100.0%	All three
Ministral-3B	Competence	10	10	0.094	[0.089, 0.099]	97.5%	Weaker at layer 10

Appendix

Table 24: Task-direction intervention results across the three primary models. The task direction is added at the atomic-name span and subtracted at the matched short-fragmented span, then compared against the reverse edit. Positive contrasts indicate movement in the expected atomic-minus-fragmented choice direction. Confidence intervals are obtained by resampling complete matched-name pairs.

Figure 12: Readout and intervention specificity can peak at nearby layers. The target curve shows the 200-pair competence-direction intervention in Ministral (400 names). Layer 10 is the development-selected observational readout; layer 8 is the nearby site where the target direction most clearly exceeds the random-subspace, polarity-shuffle, and unrelated-axis controls.

Short Name	Model Name	Model / Training Stage	License	Hugging Face Model ID
GPT-4 tokenizer	GPT-4 tokenizer	Tokenizer-only	Proprietary/API tokenizer	gpt-4 via tiktoken ; no HF model ID
GPT-5 tokenizer	GPT-5 tokenizer	Tokenizer-only	Proprietary/API tokenizer	gpt-5 via tiktoken ; no HF model ID
gpt-oss 120B	GPT-OSS 120B	Reasoning-oriented / post-trained	Apache-2.0	openai/gpt-oss-120b
gpt-oss 20B	GPT-OSS 20B	Reasoning-oriented / post-trained	Apache-2.0	openai/gpt-oss-20b
Aya 8B	Aya Expanse 8B	Post-trained	CC-BY-NC-4.0 + C4AI AUP	CohereLabs/aya-expanse-8b
Aya 32B	Aya Expanse 32B	Post-trained	CC-BY-NC-4.0 + C4AI AUP	CohereLabs/aya-expanse-32b

Appendix

Table 25: Model and tokenizer metadata for checkpoints used across NameTrace experiments. The table reports the shortened names used in figures and tables, corresponding model names, model or training stage, license, and source identifier. Base / pretrained denotes checkpoints before instruction or other post-training, while post-trained includes instruction-tuned or otherwise post-trained checkpoints. Hugging Face identifiers refer to the public or gated repositories used for the open-weight models. GPT-4 and GPT-5 are tokenizer-only tiktoken encodings and therefore have no corresponding Hugging Face model checkpoint.

Figure 13: Where lexical comparability matters. NameTrace turns name-surface support into a measurable lifecycle check for model development and evaluation. Tokenizer design determines direct lexical access, pretraining and post-training shape how support-linked differences become accessible inside the model, and evaluation should make these differences explicit before interpreting behavioral outcomes. Matched names may therefore still be lexically unmatched.

Explore similar work

Jul 14, 2026cs.AI

The Model Knows Your Project, Not You: Measuring Recognition in LLMs with NameRank

What a frontier model recalls about a person or tool from its own weights -- before any retrieval step -- often shapes the first description a human sees, making that parametric corpus presence a measurement problem. Citations explain about a third of whether a model recognizes a researcher; we target the residual and build NameRank, a [0,1] recognition score: each of 4,685 entities in 54 cohorts is probed with one open-ended question across 36 models, and an independent judge returns a binary verdict against a curated gold -- did the model state a specific, non-guessable fact about this exact entity? -- so hallucination, context echo, and guesses earn nothing. Synthetic-null entities hold the floor near zero, and verdicts track the entity, not the model. One thesis organizes the findings: recognition is paid to named, indexable artifacts, not to credentials or titles. Every Olympic-style credential sits below a working-researcher baseline, because no named artifact ships with the medal, yet the ranking inverts at the marquee tier, where Nobel, Turing, and Fields laureates saturate the panel. For independent creators the tool out-ranks its maker, and the credential that does propagate is a named method or awarded paper. Being one of many named contributors to a celebrated artifact, by contrast, earns almost nothing -- the authors listed on a flagship model report or system card sit near the recognition floor -- because recognition attaches to the artifact's own distinctive name, not to the roster behind it. No bibliometric predicts recognition well; top-density institutions out-recognize peers at matched citations; and on 258 news events recognition loads on peak salience, not persistence. A self-report probe shows introspection reads a corpus prior, not its own knowledge.

Bojie Li, Noah Shi

Pine AI · University of Washington

Credibility Frontier Models Strong Baseline Artifacts

Ask PDF

Dec 14, 2024cs.CL

Tokens, the oft-overlooked appetizer: Large language models, the distributional hypothesis, and meaning

Tokenization is a necessary component within the current architecture of many language mod-els, including the transformer-based large language models (LLMs) of Generative AI, yet its impact on the model's cognition is often overlooked. We argue that LLMs demonstrate that the Distributional Hypothesis (DH) is sufficient for reasonably human-like language performance (particularly with respect to inferential lexical competence), and that the emergence of human-meaningful linguistic units among tokens and current structural constraints motivate changes to existing, linguistically-agnostic tokenization techniques, particularly with respect to their roles as (1) vehicles for conveying salient distributional patterns from human language to the model and as (2) semantic primitives. We explore tokenizations from a BPE tokenizer; extant model vocabularies obtained from Hugging Face and tiktoken; and the information in exemplar token vectors as they move through the layers of a RoBERTa (large) model. Besides creating suboptimal semantic building blocks and obscuring the model's access to the necessary distributional patterns, we describe how tokens and pretraining can act as a backdoor for bias and other unwanted content, which current alignment practices may not remediate. Additionally, we relay evidence that the tokenization algorithm's objective function impacts the LLM's cognition, despite being arguably meaningfully insulated from the main system intelligence. Finally, we discuss implications for architectural choices, meaning construction, the primacy of language for thought, and LLM cognition. [First uploaded to arXiv in December, 2024.]

Julia Witte Zimmerman, Denis Hudon, Kathryn Cramer +9

Computational Story Lab, Vermont Complex Systems Institute, University of Vermont, Burlington, VT 05405, USA · Vermont Complex Systems Institute, University of Vermont, Burlington, VT 05405, USA · Department of Computer Science, St. Michael’s College, Colchester, VT 05439, USA +3

Single-Token Output Distributions Distributional Information Transformer Architectures Generative Artificial Intelligence

Ask PDF

Aug 13, 2026cs.CL

It's How You Ask: Gender-Associated Linguistic Bias in LLMs

Professional communication is increasingly mediated by LLMs - but do these models serve all users equally? We show that when prompts contain linguistic features more commonly used by women (hedges, tag questions, collective reference), they systematically elicit shorter, less sophisticated, and less formal responses across three document types and four models. These effects persist after controlling for prompt complexity and feature carry-over. Explicit gender cues like sign-off names are encoded in the same representational space as linguistic dialect - suggesting shared underlying mechanisms - yet linguistic register is far more influential, producing large, consistent effects where names produce none. Our results further reveal that post-hoc mitigation is challenging: because these patterns are culturally embedded and outside conscious control, users cannot easily avoid them through strategic self-presentation, and mechanistic analysis reveals that linguistic features are encoded in early transformer layers and entangled with other features. Our work calls for upstream consideration of the influences of linguistic variation to mitigate disparate impacts of LLM-mediated workplace communication.

Katherine Van Koevering, Anjalie Field

Data Science and AI Institute Johns Hopkins University · Computer Science Department Johns Hopkins University

Linguistics Collective Behaviors

Ask PDF