When AI Writes, Whose Voice Remains? Quantifying Cultural Marker Erasure Across World English Varieties in Large Language Models

Satyam Kumar Navneet; Joydeep Chandra; Yong Zhang

cs.HCFeb 25, 2026

When AI Writes, Whose Voice Remains? Quantifying Cultural Marker Erasure Across World English Varieties in Large Language Models

Authors: Satyam Kumar Navneet, Joydeep Chandra, Yong Zhang

Organizations: Independent Researcher Bihar, India · BNRIST, Dept. of CST, Tsinghua University Beijing, China

Abstract

Large Language Models (LLMs) are increasingly used to ``professionalize'' workplace communication, often at the cost of linguistic identity. We introduce "Cultural Ghosting", the systematic erasure of linguistic markers unique to non-native English varieties during text processing. Through analysis of 22,350 LLM outputs generated from 1,490 culturally marked texts (Indian, Singaporean,& Nigerian English) processed by five models under three prompt conditions, we quantify this phenomenon using two novel metrics: Identity Erasure Rate (IER) & Semantic Preservation Score (SPS). Across all prompts, we find an overall IER of 10.26%, with model-level variation from 3.5% to 20.5% (5.9x range). Crucially, we identify a Semantic Preservation Paradox: models maintain high semantic similarity (mean SPS = 0.748) while systematically erasing cultural markers. Pragmatic markers (politeness conventions) are 1.9x more vulnerable than lexical markers (71.5% vs. 37.1% erasure). Our experiments demonstrate that explicit cultural-preservation prompts reduce erasure by 29% without sacrificing semantic quality.

Figures & tables

Figure 1. Conceptual illustration of cultural ghosting: LLM-based writing assistants transform culturally marked expressions (e.g., "Kindly…", "Lah!", "Respected Sir…") into semantically preserved but culturally flattened outputs (e.g., "Please…", "Hello", "Respond"), demonstrating how meaning is retained while identity-linked linguistic markers are systematically erased. A central LLM figure converts culturally rich expressions from diverse speakers into standardized professional text, illustrating semantic preservation alongside cultural marker erasure.

Figure 2. End-to-end experimental pipeline for measuring cultural ghosting. The workflow progresses from dataset construction ( 1,490 texts ) through cultural marker annotation ( 108 markers ), LLM processing, forensic computation of Identity Erasure Rate (IER) & Semantic Preservation Score (SPS), statistical testing, to key empirical findings.

Type	Original	After LLM
Pragmatic	“Kindly do X”	“Please do X”
Pragmatic	“Respected sir”	“Hello”
Lexical	“Do the needful”	“Take necessary action”
Lexical	“Revert back”	“Respond”
Syntactic	“Discuss about”	“Discuss”

Table 1. Representative Cultural Markers & Erasure Examples

Figure 3. Illustrative vignette of cultural ghosting in AI-mediated rewriting. Given culturally marked Indian English input, standard professionalism prompts lead models to remove or replace regional markers (left), while preservation-oriented prompts sometimes retain surface forms but alter their pragmatic meaning (right).

Model	IER M (SD)	SPS M (SD)	Rank
Mistral-7B ( Team, 2024b )	0.205 (0.389)	0.857 (0.089)	Worst
Apertus-8B ( Hernández-Cano et al., 2025 )	0.152 (0.343)	0.805 (0.132)	Poor
DeepHat-7B ( Team, 2024a )	0.145 (0.337)	0.764 (0.147)	Fair
MiMo-7B ( Xiaomi, 2025 )	0.073 (0.249)	0.662 (0.257)	Good
Qwen3-8B ( Team, 2025 )	0.035 (0.176)	0.589 (0.204)	Best

Table 2. Model Performance (Baseline Prompt, n=1,490/model)

Type	Instances	Possible	Erased	Rate
Pragmatic	198	990	708	71.5%
Syntactic	166	830	467	56.3%
Lexical	260	1,300	482	37.1%
Total	624	3,120	1,657	53.1%

Table 3. Erasure by Marker Category (All Models, Baseline Prompt)

Mode	Example (Input $\rightarrow$ Output)
Direct Removal	Kindly do the needful $\rightarrow$ Please complete the task
	Respected Sir $\rightarrow$ Hello
Paraphrastic Assimilation	Revert back $\rightarrow$ Respond
	Do the needful $\rightarrow$ Take necessary action
False Preservation	Kindly [+ do X] $\rightarrow$ Kindly [+ do X] (flattened)

Table 4. Ghosting Modes (Mistral-7B)

Figure 4. Erasure rates by marker category under the baseline prompt. Pragmatic markers (71.5%) show the highest vulnerability, followed by syntactic (56.3%) & lexical (37.1%). Erasure rates by marker category under the baseline prompt. Pragmatic markers (71.5\%) show the highest vulnerability, followed by syntactic (56.3\%) \& lexical (37.1\%). — Figure 4. Erasure rates by marker category under the baseline prompt. Pragmatic markers (71.5%) show the highest vulnerability, followed by syntactic (56.3%) & lexical (37.1%). Erasure rates by marker category under the baseline prompt. Pragmatic markers (71.5%) show the highest vulnerability, followed by syntactic (56.3%) & lexical (37.1%).

Prompt	IER M (SD)	SPS M (SD)	$\Delta$ IER
Baseline	0.1156 (0.298)	0.7421 (0.204)
Neutral	0.0934 (0.276)	0.7498 (0.198)	$-19\%$
Preservation	0.0821 (0.267)	0.7548 (0.196)	$-29\%$

Table 5. Prompt Effects (All Models, n=7,450/condition)

Figure 5. The Semantic Preservation Paradox. High semantic similarity (SPS > 0.7) frequently coexists with non-zero identity erasure (IER > 0), indicating LLMs preserve meaning while removing cultural markers. The Semantic Preservation Paradox. High semantic similarity (SPS > 0.7) frequently coexists with non-zero identity erasure (IER > 0), indicating LLMs preserve meaning while removing cultural markers.

Linguistic Bias in Language Models

Linguistic bias in language models studies how language varieties, identity cues, and cultural context shape model outputs and judgments.

See all 87 papers

When English Rewrites Local Knowledge: Global Narrative Dominance in Large Language Models
May 28, 2026Md Arid Hasan, Ruwad Naswan, Farhan Samir +2Multilingual Language Model Evaluation LLM Grounding
University of Toronto · BUET · University of Illinois Urbana-Champaign
Mitigating Cross-Lingual Cultural Inconsistencies in LLMs via Consensus-Driven Preference Optimisation
Apr 2, 2026Lucas Resck, Isabelle Augenstein, Anna KorhonenCross-Lingual Consistency in Language Models Cross-Cultural Language Model Evaluation
Language Technology Lab, University of Cambridge · University of Copenhagen
Beyond "AI Language": The case for the idiolectal nature of LLM output
Aug 6, 2026Karolina Rudnicka, Thomas Stephan JuzekLanguage Model Fingerprinting Output Diversity in Language Models
Faculty of Languages, University of Gdańsk, Poland · Department of Scientific Computing, Florida State University, United States

May 28, 2026J/K move · Enter open · S save

When English Rewrites Local Knowledge: Global Narrative Dominance in Large Language Models

Md Arid Hasan, Ruwad Naswan, Farhan Samir, Sharifa Sultana+1

University of Toronto · BUET · University of Illinois Urbana-Champaign

Multilingual Language Model Evaluation LLM Grounding Cross-Lingual Reasoning in Language Models Cross-Cultural Language Model Evaluation Question Answering Cultural Bias in Language Models+3

Open paper AskPDF arXiv

Large language models (LLMs) are widely used as cross-lingual knowledge interfaces. However, culturally grounded questions often reflect globally dominant narratives rather than local contexts. We study this failure mode as global narrative dominance in Bangla, a low-resource cultural context. We introduce CulturalNB, a dataset of 717 manually curated Bengali cultural instances with parallel Bangla--English question--answer pairs and supporting evidence, metadata, and sociocultural annotations. Using question-only and evidence-based prompting, we evaluate nine state-of-the-art LLMs with human and two independent LLM judges across metrics for cross-lingual consistency, language anchoring, global substitution, institutional bias, and epistemic perspective coverage. Results show that questions asked in English systematically increase global substitution and institutional framing while reducing local perspective coverage. Local evidence improves factual consistency and perspective coverage, but does not eliminate language-induced epistemic shifts. These findings suggest that cultural failures in LLMs are not only missing-knowledge errors but also failures of grounding and narrative prioritization.

Abstract

Linguistic Bias in Language Models

Explore similar work

When English Rewrites Local Knowledge: Global Narrative Dominance in Large Language Models

Mitigating Cross-Lingual Cultural Inconsistencies in LLMs via Consensus-Driven Preference Optimisation

Beyond "AI Language": The case for the idiolectal nature of LLM output