---
title: "The American Palimpsest — paper by Bharati et al. (2026)"
description: "The paper asks how much a writing assistant strips South Asian English (SAsE) features from user text when asked to fix or… 23 references."
authors: ["Soumedhik Bharati", "Shibam Mandal", "Swarup Kr Ghosh", "Sayani Mondal"]
year: 2026
venue: "Proceedings of the 4th Workshop on Cross-Cultural Considerations in NLP (C3NLP 2026)"
source: "https://aclanthology.org/2026.c3nlp-1.8/"
doi: "10.18653/v1/2026.c3nlp-1.8"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/bharati-2026-american-palimpsest-quantifying-south-asian/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/bharati-2026-american-palimpsest-quantifying-south-asian/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/52/) / The American Palimpsest: Quantifying South Asian English Dialect Erasure in LLMs

# The American Palimpsest: Quantifying South Asian English Dialect Erasure in LLMs

**Authors:** Soumedhik Bharati, Shibam Mandal, Swarup Kr Ghosh and Sayani Mondal · **Year:** 2026 · **Venue:** [Proceedings of the 4th Workshop on Cross-Cultural Considerations in NLP (C3NLP 2026)](https://md.research.elamur.ai/free/en/venues/c3nlp/) ([Workshop on Cross-Cultural Considerations in NLP 2026](https://md.research.elamur.ai/free/en/venues/c3nlp/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.c3nlp-1.8/) · [PDF](https://aclanthology.org/2026.c3nlp-1.8.pdf) · [DOI 10.18653/v1/2026.c3nlp-1.8](https://doi.org/10.18653/v1/2026.c3nlp-1.8) · [OpenAlex](https://openalex.org/W7166888525)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper asks how much a writing assistant strips South Asian English (SAsE) features from user text when asked to fix or polish it. The authors frame this as a cultural harm, not just a stylistic side effect. Three SAsE-speaking authors built a diagnostic benchmark of 500 sentences, and two further annotators checked the feature labels. Each sentence carries exactly one salient marker, either lexical (*prepone*, *do the needful*) or syntactic (*discuss about*, *is knowing*, *isn't it?*). After a model rewrites a sentence, the marker can be checked by exact match or regex. Llama 3.3 70B was tested under three prompts: standard correction, formalization and dialect-aware correction. Correction kept only about a quarter of the markers, and formalization kept even fewer. Most lexical items were swapped for American equivalents. A short instruction to preserve the dialect restored most markers, but article insertion did not yield to it. A harder subset built from the most-erased features pushed retention close to zero. The work matters because it supplies a cheap, auditable protocol and a metric bundle for checking whether rewriting tools impose a single prestige norm.

## Key points
- The benchmark has 500 sentences (320 lexical, 180 syntactic) spread over a 22-way category scheme. Agreement was Cohen's κ 0.86 for the lexical/syntactic split and 0.79 for the category labels.
- Results for Llama 3.3 70B by prompt:
  - **Standard correction:** 26.0% overall retention (lexical 31.2%, syntactic 16.7%), with 56.2% lexical Americanization.
  - **Formalization:** 14.0% retention, with 59.4% lexical Americanization.
  - **Dialect-aware prompt:** 92.0% retention, with lexical Americanization down to 6.2%.
- Phase 2 used a 150-item core subset: the 15 most-erased features, with 10 items each. Under plain correction, retention fell to 6.7%. Typical swaps were *almirah* → *wardrobe*, *airdash* → *rush* and *cousin brother* → *cousin*.
- The Article category (zero-article use) stayed at 0.0% retention even with the dialect-aware prompt. This points to a prescriptive habit that instructions cannot override.
- The proposed metrics are:
  - feature retention, checked by regex or exact match;
  - Americanization, meaning a known Standard American paraphrase appears in the output;
  - edit aggressiveness, measured as unchanged rate or edit distance.

## Interesting ideas
- Build test items that each contain one salient marker, so success or failure can be checked mechanically and blamed on a specific phenomenon rather than lost in aggregate scores.
- Score not only whether a marker survived but whether it was replaced by a known standard paraphrase, which separates deletion from assimilation.
- Report a strict score alongside a relaxed one that tolerates orthographic variants (*batch-mate* vs *batchmate*), so surface normalization is not counted as erasure.
- Derive a stress-test subset from the features that failed most in a broad first pass, concentrating evaluation where the system is weakest.
- Some normalizations resist explicit instructions, so an LLM's output cannot be trusted to preserve input features just because it was told to.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (tree): A small set of single-construction SAsE sentences (*discuss about*, *is knowing*, *today itself*, tag *no?*) could test whether the parser and UD dialect conversion give these constructions sensible trees instead of mis-parsing them as errors.
- **I2** (back): The text → graph → text compressor and generation could be audited for marker retention and for drift toward standard paraphrases, so a round trip does not quietly normalize a dialect.
- **I3** (learn): Evaluation of lemmas and generated text could carry both scores, keeping hyphenation or spelling noise apart from real lexical substitution.
- **I4** (learn): Mova's evaluation loop could automatically assemble hard sets from the most frequent parser or reader failures and track them separately over training runs.
- **I5** (learn): When LLM teacher tiers produce or repair silver data, outputs should be checked for silently inserted articles or Americanized words before acceptance, perhaps with dialect sentences as a negative control.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

Also on the byline: [Sister Nivedita University](https://ror.org/02fv8d015).

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,954 works · [🤝 Safety Research](https://md.research.elamur.ai/free/en/shelves/social-sciences/safety-research/) · 626 works

🧭 Topics (2): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,894 works · [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works

## Cite

Cite as: Soumedhik Bharati, Shibam Mandal, Swarup Kr Ghosh and Sayani Mondal. 2026. The American Palimpsest: Quantifying South Asian English Dialect Erasure in LLMs. Proceedings of the 4th Workshop on Cross-Cultural Considerations in NLP (C3NLP 2026).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{bharati2026american,
  title = {{The American Palimpsest: Quantifying South Asian English Dialect Erasure in LLMs}},
  author = {Soumedhik Bharati, Shibam Mandal, Swarup Kr Ghosh and Sayani Mondal},
  booktitle = {Proceedings of the 4th Workshop on Cross-Cultural Considerations in NLP (C3NLP 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.c3nlp-1.8},
  url = {https://doi.org/10.18653/v1/2026.c3nlp-1.8}
}
```

- [← previous The Algorithmic Unconscious: Structural Mechanisms and Implicit Biases in Large Language Models](https://md.research.elamur.ai/free/en/papers/boisnard-2026-algorithmic-unconscious-structural-mechanisms-implicit/)
- [next → The Anthropology of Food: How NLP can Help us Unravel the Food cultures of the World](https://md.research.elamur.ai/free/en/papers/riabi-2026-anthropology-food-how-nlp-can/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/52/)

## References

23 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Arnav Arora, Lucie-aimée Kaffee, and Isabelle Augenstein. 2023. [Probing pre-trained language models for cross-cultural differences in values](https://md.research.elamur.ai/free/en/papers/arora-2023-probing-pre-trained-language-models/). In Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP), pages 114–130, Dubrovnik, Croatia. Association for Computational Linguistics.
2. Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pages 610–623. [↗](https://doi.org/10.1145/3442188.3445922)
3. Su Lin Blodgett, Solon Barocas, Hal Daumé III, and Hanna Wallach. 2020. [Language (technology) is power: A critical survey of “bias” in NLP](https://md.research.elamur.ai/free/en/papers/blodgett-2020-language-technology-is-power-critical/). In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5454– 5476, Online. Association for Computational Linguistics. [↗](https://doi.org/10.18653/v1/2020.acl-main.485)
4. Yong Cao, Li Zhou, Seolhwa Lee, Laura Cabello, Min Chen, and Daniel Hershcovich. 2023. Assessing cross-cultural alignment for large language models. In Findings of the Association for Computational Linguistics: ACL 2023, pages 3276–3293, Toronto, Canada. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Yong+Cao%2C+Li+Zhou%2C+Seolhwa+Lee%2C+Laura+Cabello%2C+Min+Chen%2C+and+Daniel+Hershcovich%2E+2023%2E+Assessing+cross%2Dcultural+alignment+for+large+language+models%2E+In+Findings+of+the+Association+for+Computational+Li)
5. Jacob Cohen. 1960. A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20:37 – 46. [↗](https://doi.org/10.1177/001316446002000104)
6. Kate Crawford. 2017. The trouble with bias. In Advances in Neural Information Processing Systems, Keynote. [find](https://scholar.google.com/scholar?q=Kate+Crawford%2E+2017%2E+The+trouble+with+bias%2E+In+Advances+in+Neural+Information+Processing+Systems%2C+Keynote%2E)
7. Christof Demont-Heinrich. 2011. Cultural imperialism versus globalization of culture: Riding the structure-agency dialectic in global communication and media studies. Sociology Compass, 5(8):666–678. [↗](https://doi.org/10.1111/j.1751-9020.2011.00401.x)
8. Sebastian Gehrmann and 1 others. 2021. [The GEM benchmark: Natural language generation, its evaluation and metrics](https://md.research.elamur.ai/free/en/papers/gehrmann-2021-gem-benchmark-natural-language-generation/). In Proceedings of the 1st Workshop on Natural Language Generation, Evaluation, and Metrics (GEM 2021), pages 96–120.
9. Suchin Gururangan, Dallas Card, Sarah Dreier, Emily Gade, Leticia Wang, Zeyu Wang, Luke Zettlemoyer, and Noah A. Smith. 2022. Whose language counts as high quality? measuring language ideologies in text data selection. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2562–2580, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics. [↗](https://doi.org/10.48550/arxiv.2201.10474)
10. Daniel Hershcovich, Stella Frank, Heather Lent, Miryam de Lhoneux, Mostafa Abdou, Stephanie Brandl, Emanuele Bugliarello, Laura Cabello Piqueras, Ilias Chalkidis, Ruixiang Cui, Constanza Fierro, Katerina Margatina, Phillip Rust, and Anders Søgaard. 2022. [Challenges and strategies in cross-cultural NLP](https://md.research.elamur.ai/free/en/papers/hershcovich-2022-challenges-strategies-cross-cultural-nlp/). In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6997–7013, Dublin, Ireland. Association for Computational Linguistics. [↗](https://doi.org/10.48550/arxiv.2203.10020)
11. Valentin Hofmann, Pratyusha Ria Kalluri, Dan Jurafsky, and Su Lin Blodgett. 2024. Dialect prejudice in language models. Nature, 633:893–899. [find](https://scholar.google.com/scholar?q=Valentin+Hofmann%2C+Pratyusha+Ria+Kalluri%2C+Dan+Jurafsky%2C+and+Su+Lin+Blodgett%2E+2024%2E+Dialect+prejudice+in+language+models%2E+Nature%2C+633%3A893%E2%80%93899%2E)
12. Pratik Joshi, Sebastin Santy, Amar Budhiraja, Kalika Bali, and Monojit Choudhury. 2020. [The state and fate of linguistic diversity and inclusion in the NLP world](https://md.research.elamur.ai/free/en/papers/joshi-2020-state-fate-linguistic-diversity-inclusion/). In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6282–6293.
13. Braj B. Kachru. 2005. Asian Englishes: Beyond the Canon. Hong Kong University Press. [find](https://scholar.google.com/scholar?q=Braj+B%2E+Kachru%2E+2005%2E+Asian+Englishes%3A+Beyond+the+Canon%2E+Hong+Kong+University+Press%2E)
14. Nitika Mathur, Timothy Baldwin, and Trevor Cohn. 2020. [Tangled up in BLEU: Reevaluating the evaluation of automatic machine translation evaluation metrics](https://md.research.elamur.ai/free/en/papers/mathur-2020-tangled-up-bleu-reevaluating-evaluation/). In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4984–4997.
15. Moin Nadeem, Anna Bethke, and Aida Cho. 2021. [StereoSet: Measuring stereotypical bias in pretrained language models](https://md.research.elamur.ai/free/en/papers/nadeem-2021-stereoset-measuring-stereotypical-bias-pretrained/). In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics, pages 5356–5371.
16. Tarek Naous, Michael J Ryan, Alan Ritter, and Wei Xu. 2024. [Having beer after prayer? measuring cultural bias in large language models](https://md.research.elamur.ai/free/en/papers/naous-2023-having-beer-after-prayer-measuring/). In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16366–16393, Bangkok, Thailand. Association for Computational Linguistics. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, and 1 others. 2022. Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems, volume 35, pages 27730–27744.
17. Shramay Palta and Rachel Rudinger. 2023. [FORK: A bite-sized test set for probing culinary cultural biases in commonsense reasoning models](https://md.research.elamur.ai/free/en/papers/palta-2023-fork-bite-sized-test-set/). In Findings of the Association for Computational Linguistics: ACL 2023, pages 9952–9962, Toronto, Canada. Association for Computational Linguistics. [↗](https://doi.org/10.48448/pd8s-my91)
18. Marco Tulio Ribeiro, Tongshuang Wu, Carlos Guestrin, and Sameer Singh. 2020. Beyond accuracy: Behavioral testing of NLP models with CheckList. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4902– 4912, Online. Association for Computational Linguistics. [↗](https://doi.org/10.18653/v1/2020.acl-main.442)
19. P. Sailaja. 2009. Indian English. Dialects of English. Edinburgh University Press. [find](https://scholar.google.com/scholar?q=P%2E+Sailaja%2E+2009%2E+Indian+English%2E+Dialects+of+English%2E+Edinburgh+University+Press%2E)
20. Shibani Santurkar, Esin Tariq, Andrew Ilyas, and Aleksander M adry. 2023. [Whose opinions do language models reflect?](https://md.research.elamur.ai/free/en/papers/santurkar-2023-whose-opinions-language-models-reflect/) In Proceedings of the 40th International Conference on Machine Learning.
21. Wenxuan Wang, Wenxiang Jiao, Jingyuan Huang, Ruyi Dai, Jen-tse Huang, Zhaopeng Tu, and Michael Lyu. 2024. [Not all countries celebrate thanksgiving: On the cultural dominance in large language models](https://md.research.elamur.ai/free/en/papers/wang-2024-not-all-countries-celebrate-thanksgiving/). In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6349–6384. Association for Computational Linguistics. [↗](https://doi.org/10.48550/arxiv.2310.12481)
22. Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2018. [Gender bias in coreference resolution: Evaluation and debiasing methods](https://md.research.elamur.ai/free/en/papers/zhao-2018-gender-bias-coreference-resolution-evaluation/). In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 15–20. [↗](https://doi.org/10.48550/arxiv.1804.06876)
23. Caleb Ziems, William Alikaniotis, Luke Zettlemoyer, and Diyi Yang. 2023. [Multi-VALUE: A framework for cross-dialectal English NLP](https://md.research.elamur.ai/free/en/papers/ziems-2023-multi-value-framework-cross-dialectal/). In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, pages 3156–3176.

*Updated 2 November 2026*