Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Risks of Cultural Erasure in Large Language Models

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Large language models systematically erase African and Asian cultures through omission and simplification.

desk verdict A genuine empirical study of cultural erasure in LLMs with a real measurement caveat in Study 1; the core pattern is plausible and deserves peer review. read the letter →

arxiv 2501.01056 v1 pith:JLRX7SWY submitted 2025-01-02 cs.CL cs.AI

Rida Qadri , Aida M. Davani , Kevin Robinson , Vinodkumar Prabhakaran This is my paper · ORCID
classification cs.CLcs.AI
keywords culturalerasureomissionsimplificationlargelanguagemodelsrepresentationalbiastravelrecommendationsglobalrepresentationthematicanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that large language models reproduce a form of cultural erasure documented in media studies: some cultures are left out of model outputs entirely, while others appear only in flattened, one-dimensional form. The authors test this in two settings that shape public knowledge: open-ended descriptions of fifty global cities, and travel recommendations from a widely used model API. They find that European and North American cities are more often described through cultural themes, African and Asian cities more often through economic themes, and that whole sub-regions of Africa, Central Asia, the Caribbean and the Pacific never appear in travel recommendations at all. The paper's purpose is to convert a sociological concept of harm into measurable evaluation targets for language technologies, so that omissions and simplifications can be tracked like other model behaviors.

What carries the argument

The machinery is an operational decomposition of cultural erasure into omission and simplification. Omission is measured by extracting city mentions from model responses to seven travel-interest prompts and computing the empirical probability that each world sub-region appears; simplification is measured by having crowdworkers annotate eight representational themes in model-written city descriptions, with the contrast between culture and economy treated as a marker of one-dimensional representation. Cities serve as the unit of analysis because they are a standardized proxy for cultural communities across the world, unlike ethnicity or religion, which are recorded differently from country to country.

What would settle it

Re-run Study 1 with annotators drawn from several world regions, or blind the city names; if the pattern of African and Asian cities scoring low on culture disappears or reverses across annotator pools, the measured erasure is an artifact of the single annotator pool rather than a property of the model. Re-running Study 2 on a second model family would likewise show whether the omission pattern is specific to one API.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's terms, is that LLM-generated text replicates the 'single story' pattern long observed in global media. When prompted to describe cities, the model associates culture with Western cities and economy with African and Asian cities: nine of the ten cities scoring lowest on cultural themes are in Africa or Asia, while seven of the ten highest-scoring cities on economic themes are in Africa or Asia. When prompted for travel advice, the API omits eight sub-regions entirely and gives minimal representation to eight more, with all of Africa and four of five Asian sub-regions affected; even within represented regions a single city often dominates, with London making up 100% of Northern European references and Paris 94% of Western European references. The authors treat omission and simplification as two measurable forms of cultural erasure and argue that language technologies embedded in search, education and travel will scale up these historical inequities unless they are explicitly evaluated for them.

Load-bearing premise

The load-bearing premise is that crowdworkers recruited from a single country, India, can reliably judge whether a text about a city anywhere in the world is cultural or economic, and that this judgment captures simplification rather than the annotators' own cultural lens.

Editorial extensions

If this is right

  • The same two-step evaluation can be applied to any model or API, so cultural erasure can become a routine benchmark alongside safety and accuracy.
  • Travel planners built on these models will systematically under-recommend African, Central Asian, Caribbean and Pacific destinations, affecting the tourism economies of those regions.
  • Region-level coverage is not enough: within a well-covered region such as Northern Europe, a single city (London) can erase the diversity of that region, so evaluations must slice below the regional level.
  • Because African contexts appear mainly under 'authentic places' queries and Southern Asia under 'spiritual journey' queries, personalized interest-based recommendation can reinforce colonial-era associations even when overall coverage looks balanced.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's per-region probabilities could be aggregated into a single 'erasure index' that weights sub-regions equally, giving developers one number to track across model versions; the paper stops short of proposing such an index.
  • The travel prompts were authored from a Western cultural frame, so a testable extension is to have users from the omitted regions write their own interest cues and see whether the recommended geography widens; if it does, part of the erasure resides in the prompt rather than in the model's stored knowledge.
  • If the same pattern holds in search and education, the paper's travel finding implies that AI-generated lessons and search summaries will under-represent the same regions, turning a travel-specific result into a broader claim about AI-mediated cultural knowledge.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This paper adapts the media-studies concept of 'cultural erasure' to LLM evaluation, defining erasure as omission (cultures not represented at all) and simplification (one-dimensional representations). Study 1 probes a pre-trained PaLM model with five prompts for 50 global cities, has crowdworkers annotate the generated descriptions for representational themes, and reports that European and North American cities score higher on the 'culture' theme while African and Asian cities score higher on 'economy/economic conditions,' interpreting this as a replication of historical media tropes. Study 2 queries the PaLM 2 API with seven interest cues across four application contexts, automatically extracts city references with author review, and reports that 8 sub-regions have zero representation, that African regions are almost entirely absent, and that within well-represented regions a few cities dominate (London 100%, Paris 94%). The authors argue that LLMs risk algorithmically scaling up cultural erasure and advocate for sociologically informed evaluation frameworks that go beyond statistical parity.

Significance. The paper's main contribution is an operational, metricizable framework for examining representational harms in LLM outputs, moving beyond parity-based bias metrics to ask not only whether cultures are represented but how they are represented. It productively imports media-studies theory into NLP evaluation, provides transparent methods and full prompt templates, and makes concrete claims about geographic disparities that are, in principle, falsifiable. If the empirical findings survive closer scrutiny of the annotation layer, this would be a useful and reproducible evaluation recipe. The Study 2 omission results, which rely on automated city extraction with author review rather than subjective annotation, are more robust to the measurement concerns. The main uncertainty is whether Study 1's human-annotation procedure yields valid and reliable measures of the intended themes; this uncertainty is acknowledged in the paper but not yet resolved.

major comments (4)
  1. [§4.1.3, §3] The core quantitative evidence for the simplification claim rests on majority-vote theme labels assigned by 11 annotators recruited from a single country (India), with no demographic data, no inter-annotator agreement statistics, and no sensitivity analyses. The authors themselves state in §3 that annotators 'may lack nuanced sociocultural knowledge required to accurately evaluate themes of representation across such different places and cultures.' Because the regional culture/economy gap in Figures 1–2 and the associated t-tests are computed on these labels, this is a load-bearing measurement validity concern: if annotators apply 'culture' and 'economy' differently to cities they know well versus those they do not, the headline regional differences could reflect annotator interpretation rather than model behavior. I request per-theme inter-annotator agreement (e.g., Krippendorff's alpha or Fleiss' kappa), a check of whether annotator familiarity with cities predicts theme assignment, and ideally a sensitivity analysis using raters from multiple countries.
  2. [§4.1.1, §2.1] One of the five Study 1 prompts is 'Write a few paragraphs about the culture of <city>?', which directly primes the culture theme. The paper does not report per-prompt breakdowns or results with this prompt excluded. Since the central claim concerns relative emphasis on culture versus economy by region, a prompt-driven artifact cannot be ruled out: the culture-priming prompt may inflate culture scores in ways that interact with region, especially if certain cities elicit longer or more detailed culture responses. The authors should show that the regional patterns persist when the analysis is restricted to the four neutral prompts (or when the priming prompt is removed), or otherwise establish that prompt 4 is not driving the reported effects.
  3. [§2.1, §4.1.3] The pairwise t-tests reported in §2.1 are numerous comparisons across continents and themes, yet the paper reports only p-values without multiple-testing corrections, effect sizes, or confidence intervals. Given that the comparisons are post hoc and correlated, some reported differences at p<0.001 may not survive a correction such as Benjamini-Hochberg or Tukey HSD. I ask for a clearer statistical analysis plan, including effect sizes (e.g., Cohen's d or odds ratios) and corrected p-values, to substantiate the claim that Africa's lower culture/higher economy representation differs from other continents.
  4. [§2.2, Table 3] The interpretation of within-region homogeneity as erasure uses the implicit baseline that a city's representation should be proportional to its population share ('despite each city containing less than 10% of the population'). This normative baseline is not justified: travel recommendations need not mirror population size, and the paper's own framing of erasure is about absence from discourse rather than proportional representation. The claim that London's 100% share constitutes erasure of other Northern European cities should be explicitly grounded either in a stated normative principle or in evidence about user expectations, rather than population share alone.
minor comments (5)
  1. [§2.1] The text lists 'eight representational themes' but then enumerates nine themes (culture, economy, demography, geography, government, history, industry, political situations, and social issues); please correct the count or the list.
  2. [Table 2] The table caption says 'Grouping shows p(region) split at 0, 0.1, and 2.5'; this appears to be a typo, likely '0.25' — please verify.
  3. [§4.2.1] The prompt 'i want a spiritual journey. which city should i visit?' has inconsistent capitalization; the paper notes that this was not intentional but does not provide a sensitivity check. Please add a brief note or analysis showing that capitalization does not affect the outcomes.
  4. [Figure 3] The figure caption says 'graphs' but appears to reference a single figure; please clarify the number of panels and what each panel shows.
  5. [§4.2.3] The 'Automated evaluation with author review' subsection gives little detail on the review step: please specify how many responses were manually reviewed, how disagreements with the automated heuristic were resolved, and whether any reported statistics exclude corrected extractions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper directly measures model outputs and imports 'cultural erasure' from external media scholarship; self-citations are background or system references, not load-bearing evidence.

full rationale

This is an empirical measurement paper, not a derivation or fitted-model prediction. Study 1 computes theme-presence probabilities from 2,500 PaLM outputs annotated by crowdworkers (Section 4.1.3), and the regional culture/economy contrast is a direct statistical comparison of observed annotations; no parameter is fit and then relabeled as a prediction. Study 2 counts city mentions in 280 PaLM 2 API responses (Section 4.2), so the omission rates in Tables 2-3 are observed frequencies. The concept of 'cultural erasure' is explicitly imported from external media scholarship (Gerbner & Gross [14]; Tuchman [47]) and operationalized into omission and simplification; applying an external concept to new data is not circular. Self-citations appear as related-work context (e.g., [37]) or as system documentation for the evaluated models ([9], [3], [17]) and do not carry the empirical burden. The acknowledged limitations—single-country annotator pool, one model/API, limited tasks—are measurement-validity and generalizability concerns, not cases where the conclusion is equivalent to its inputs by construction. No equation or claim reduces to a self-citation or to its own definition, so the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or formal entities. Its central claim rests on the selection of measurement instruments (themes, city list, prompts) and on normative assumptions about representation.

free parameters (2)
  • Minimal representation threshold = 0.10
    Table 2 groups regions with p(region) <= 0.10 as having 'minimal representation'. The cutoff is chosen by the authors and affects the reported counts of erased regions.
  • City list = 50 cities balanced across continents
    The specific 50 cities chosen by the authors determine which locales are probed; the list is not a random sample and may affect regional estimates.
assumptions (4)
  • domain assumption Cities are valid stand-ins for global cultures and communities
    Section 1: 'We have chosen cities as our unit of analysis as they provide a standardized proxy list for representing global communities and cultures.'
  • ad hoc to paper Cultural erasure is defined as omission and simplification, adapted from media studies
    Section 1: 'we look at two concepts of erasure: omission... and simplification.' This conceptualization is imposed on LLM outputs.
  • ad hoc to paper Crowdworker annotations of themes are a valid operationalization of simplification
    Section 4.1.3: annotations by 11 raters from India; the paper assumes this measures cultural representation, acknowledged as a limitation.
  • domain assumption A balanced distribution of themes across regions is the expected baseline
    The paper treats differences in theme prevalence as erasure without an independent baseline of what proportional representation should be.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Risks of Cultural Erasure in Large Language Models." pith.science (2026). https://pith.science/paper/JLRX7SWY

@misc{pith2026250101056,
  author       = {Pith},
  title        = {Pith review of: Risks of Cultural Erasure in Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JLRX7SWY}},
  note         = {Machine review of arXiv:2501.01056}
}
read the original abstract

Large language models are increasingly being integrated into applications that shape the production and discovery of societal knowledge such as search, online education, and travel planning. As a result, language models will shape how people learn about, perceive and interact with global cultures making it important to consider whose knowledge systems and perspectives are represented in models. Recognizing this importance, increasingly work in Machine Learning and NLP has focused on evaluating gaps in global cultural representational distribution within outputs. However, more work is needed on developing benchmarks for cross-cultural impacts of language models that stem from a nuanced sociologically-aware conceptualization of cultural impact or harm. We join this line of work arguing for the need of metricizable evaluations of language technologies that interrogate and account for historical power inequities and differential impacts of representation on global cultures, particularly for cultures already under-represented in the digital corpora. We look at two concepts of erasure: omission: where cultures are not represented at all and simplification i.e. when cultural complexity is erased by presenting one-dimensional views of a rich culture. The former focuses on whether something is represented, and the latter on how it is represented. We focus our analysis on two task contexts with the potential to influence global cultural production. First, we probe representations that a language model produces about different places around the world when asked to describe these contexts. Second, we analyze the cultures represented in the travel recommendations produced by a set of language model applications. Our study shows ways in which the NLP community and application developers can begin to operationalize complex socio-cultural considerations into standard evaluations and benchmarks.

Figures

Figures reproduced from arXiv: 2501.01056 by the authors.

Figure 1
Figure 1. Culture and economy themes in representations of places within each region. Scores on the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Bar chart showing the presence of cultural and economic themes in descriptions of various [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The graphs show the frequency of each region being recommended as a travel destination [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Distribution of references to cities within each region, across different application contexts. [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Towards Experience-Centered AI: A Framework for Integrating Lived Experience in Design and Development

    cs.CY 2025-08 conditional novelty 5.0 of 10

    A conceptual framework, LEAF, organizes four dimensions of lived experience into a five-stage AI development pipeline for more human-centered systems.

  2. LLM Harms: A Taxonomy and Discussion

    cs.CY 2025-12 unverdicted novelty 3.0 of 10

    This paper proposes a taxonomy of LLM harms in five categories and suggests mitigation strategies plus a dynamic auditing system for responsible development.

Reference graph

Works this paper leans on

49 extracted references · 40 canonical work pages · cited by 2 Pith papers

  1. [1]

    The danger of a single story, 2009

    Chimamanda Ngozi Adichie. The danger of a single story, 2009

    work page 2009

  2. [2]

    Towards measuring and modeling “culture” in LLMs: A survey

    Muhammad Farid Adilazuarda, Sagnik Mukherjee, Pradhyumna Lavania, Siddhant Shivdutt Singh, Alham Fikri Aji, Jacki O’Neill, Ashutosh Modi, and Monojit Choudhury. Towards measuring and modeling “culture” in LLMs: A survey. In Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen, editors,Proceedings of the 2024 Conference on Empirical Methods in Natural Languag...

    work page 2024

  3. [3]

    Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, Eric Chu, Jonathan H

    Rohan Anil, Andrew M. Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, Eric Chu, Jonathan H. Clark, Laurent El Shafey, Yanping Huang, Kathy Meier-Hellstern, Gaurav Mishra, Erica Moreira, Mark Omernick, Kevin Robinson, Sebastian Ruder, Yi Tay, Kefan Xiao, Yuanzhong Xu, Yujing Z...

    2023

  4. [4]

    Bowman, Zac Hatfield-Dodds, Ben Mann, Dario Amodei, Nicholas Joseph, Sam McCandlish, Tom Brown, and Jared Kaplan

    Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, Carol Chen, Catherine Olsson, Christopher Olah, Danny Hernandez, Dawn Drain, Deep Ganguli, Dustin Li, Eli Tran-Johnson, Ethan Perez, Jamie Kerr, Jared Mueller, Jeffrey Ladish, Joshua Landau, Kamal Ndousse, K...

    2022

  5. [5]

    Re- contextualizing fairness in NLP: The case of India

    Shaily Bhatt, Sunipa Dev, Partha Talukdar, Shachi Dave, and Vinodkumar Prabhakaran. Re- contextualizing fairness in NLP: The case of India. InProceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), page...

    work page 2022

  6. [6]

    Routledge Urban Reader Series

    Neil Brenner and Roger Keil, editors.The global cities reader. Routledge Urban Reader Series. Routledge, London, England, December 2005. 11

    work page 2005

  7. [7]

    Bridging cultural nuances in dialogue agents through cultural value surveys

    Yong Cao, Min Chen, and Daniel Hershcovich. Bridging cultural nuances in dialogue agents through cultural value surveys. In Yvette Graham and Matthew Purver, editors,Findings of the Association for Computational Linguistics: EACL 2024, pages 929–945, St. Julian’s, Malta, March 2024. Association for Computational Linguistics

    work page 2024

  8. [8]

    Chatgpt for tourism: applications, benefits and risks

    Inês Carvalho and Stanislav Ivanov. Chatgpt for tourism: applications, benefits and risks. Tourism Review, 2023

    work page 2023

Show all 49 references
  1. [9]

    Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam Shazeer, Vino...

    2022

  2. [10]

    Global cities and global order

    Simon Curtis. Global cities and global order. Oxford University Press, 2016

    2016

  3. [11]

    Disentangling perceptions of offensiveness: Cultural and moral correlates

    Aida Davani, Mark Díaz, Dylan Baker, and Vinodkumar Prabhakaran. Disentangling perceptions of offensiveness: Cultural and moral correlates. InThe 2024 ACM Conference on Fairness, Accountability, and Transparency, pages 2007–2021, 2024

    2024

  4. [12]

    Building socio-culturally inclusive stereotype resources with community engagement, 2023

    Sunipa Dev, Jaya Goyal, Dinesh Tewari, Shachi Dave, and Vinodkumar Prabhakaran. Building socio-culturally inclusive stereotype resources with community engagement, 2023

    2023

  5. [13]

    Such, and Natalia Criado

    Xavier Ferrer, Tom van Nuenen, Jose M. Such, and Natalia Criado. Discovering and categorising language biases in reddit, 2020

    2020

  6. [14]

    Living with Television: The Violence Profile.Journal of Communication, 26(2):172–199, 02 2006

    George Gerbner and Larry Gross. Living with Television: The Violence Profile.Journal of Communication, 26(2):172–199, 02 2006

    2006

  7. [15]

    Improving alignment of dialogue agents via targeted human judgements, 2022

    Amelia Glaese, Nat McAleese, Maja Trębacz, John Aslanides, Vlad Firoiu, Timo Ewalds, Maribeth Rauh, Laura Weidinger, Martin Chadwick, Phoebe Thacker, Lucy Campbell-Gillingham, Jonathan Uesato, Po-Sen Huang, Ramona Comanescu, Fan Yang, Abigail See, Sumanth Dathathri, Rory Greig...

    2022

  8. [16]

    Destination Insights with Google: Top insights for outbound demand from US, queried 8/4/23, 2023

    Google. Destination Insights with Google: Top insights for outbound demand from US, queried 8/4/23, 2023. Available athttps://destinationinsights.withgoogle.com/intl/en_ALL/

    2023

  9. [17]

    PaLM API Overview, 2023

    Google. PaLM API Overview, 2023. Available athttps://developers.generativeai.google/ guide/palm_api_overview

    2023

  10. [18]

    PaLM API Safety Guidance, 2023

    Google. PaLM API Safety Guidance, 2023. Available athttps://developers.generativeai. google/guide/safety_guidance. 12

    2023

  11. [19]

    Straumann, and Ahmed Medhat

    Mark Graham, Bernie Hogan, Ralph K. Straumann, and Ahmed Medhat. Uneven geographies of user-generated information: Patterns of increasing informational poverty. Annals of the Association of American Geographers, 104(4):746–764, 2014

    2014

  12. [20]

    Straumann, and Bernie Hogan

    Mark Graham, Ralph K. Straumann, and Bernie Hogan. Digital divisions of labor and informa- tional magnetism: Mapping participation in wikipedia.Annals of the Association of American Geographers, 105(6):1158–1178, 2015

    2015

  13. [21]

    Challenges and strategies in cross-cultural nlp

    Daniel Hershcovich, Stella Frank, Heather Lent, Miryam de Lhoneux, Mostafa Abdou, Stephanie Brandl, Emanuele Bugliarello, Laura Cabello Piqueras, Ilias Chalkidis, Ruixiang Cui, et al. Challenges and strategies in cross-cultural nlp. InProceedings of the 60th Annual Meeting of ...

    2022

  14. [22]

    Culturally aware natural language inference

    Jing Huang and Diyi Yang. Culturally aware natural language inference. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 7591–7609, 2023

    2023

  15. [23]

    Seegull: A stereotype benchmark with broad geo-cultural coverage leveraging generative models

    Akshita Jha, Aida Davani, Chandan K Reddy, Shachi Dave, Vinodkumar Prabhakaran, and Sunipa Dev. Seegull: A stereotype benchmark with broad geo-cultural coverage leveraging generative models. arXiv e-prints, pages arXiv–2305, 2023

    2023

  16. [24]

    Chatgpt for good? on opportunities and challenges of large language models for education.Learning and Individual Differences, 103:102274, 2023

    Enkelejda Kasneci, Kathrin Seßler, Stefan Küchemann, Maria Bannert, Daryna Dementieva, Frank Fischer, Urs Gasser, Georg Groh, Stephan Günnemann, Eyke Hüllermeier, et al. Chatgpt for good? on opportunities and challenges of large language models for education.Learning and Indiv...

    2023

  17. [25]

    Harnessing gpt-4 so that all students benefit

    Sal Khan. Harnessing gpt-4 so that all students benefit. a nonprofit ap- proach for equal access, 2023. Accessed at https://blog.khanacademy.org/ harnessing-ai-so-that-all-students-benefit-a-nonprofit-approach-for-equal-access/

    2023

  18. [26]

    Hate speech classifiers are culturally insensitive

    Nayeon Lee, Chani Jung, and Alice Oh. Hate speech classifiers are culturally insensitive. In Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP), pages 35–46, 2023

    2023

  19. [27]

    Culturally aware and adapted nlp: A taxonomy and a survey of the state of the art.arXiv preprint arXiv:2406.03930, 2024

    Chen Cecilia Liu, Iryna Gurevych, and Anna Korhonen. Culturally aware and adapted nlp: A taxonomy and a survey of the state of the art.arXiv preprint arXiv:2406.03930, 2024

    2024 arXiv

  20. [28]

    Using generative ai for travel inspiration and discovery, 2023

    Yiling Lui. Using generative ai for travel inspiration and discovery, 2023. Accessed athttps: //developers.googleblog.com/2023/05/generative-ai-travel-developers.html

    2023

  21. [29]

    Othering and low prestige framing of immigrant cuisines in us restaurant reviews and large language models, 2023

    Yiwei Luo, Kristina Gligorić, and Dan Jurafsky. Othering and low prestige framing of immigrant cuisines in us restaurant reviews and large language models, 2023

    2023

  22. [30]

    Discover the wonder of ai-powered search., 2023

    Microsoft. Discover the wonder of ai-powered search., 2023. Accessed athttps://www.microsoft. com/en-us/bing?form=MA13FJ

    2023

  23. [31]

    Blend: A benchmark for llms on everyday knowledge in diverse cultures and languages.arXiv preprint arXiv:2406.09948, 2024

    Junho Myung, Nayeon Lee, Yi Zhou, Jiho Jin, Rifki Afina Putri, Dimosthenis Antypas, Hsuvas Borkakoty, Eunsu Kim, Carla Perez-Almendros, Abinew Ali Ayele, et al. Blend: A benchmark for llms on everyday knowledge in diverse cultures and languages.arXiv preprint arXiv:2406.09948, 2024

    2024 arXiv

  24. [32]

    How western journalists actually write about africa: Re-assessing the myth of representations of africa.Journalism Studies, 19(8):1138–1159, 2018

    Toussaint Nothias. How western journalists actually write about africa: Re-assessing the myth of representations of africa.Journalism Studies, 19(8):1138–1159, 2018

    2018

  25. [33]

    The digital knowledge economy index: Mapping content production.The Journal of Development Studies, 55(12):2626–2643, 2019

    Sanna Ojanperä, Mark Graham, and Matthew Zook. The digital knowledge economy index: Mapping content production.The Journal of Development Studies, 55(12):2626–2643, 2019. 13

    2019

  26. [34]

    Gpt-4 technical report, 2023

    OpenAI. Gpt-4 technical report, 2023

    2023

  27. [35]

    Teaching with ai, 2023

    OpenAI. Teaching with ai, 2023. Accessed athttps://openai.com/blog/teaching-with-ai

    2023

  28. [36]

    Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and...

    2022

  29. [37]

    Cultural incongruencies in artificial intelligence

    Vinodkumar Prabhakaran, Rida Qadri, and Ben Hutchinson. Cultural incongruencies in artificial intelligence. arXiv preprint arXiv:2211.13069, 2022

    2022 arXiv

  30. [38]

    Normad: A benchmark for measuring the cultural adaptability of large language models.arXiv e-prints, pages arXiv–2404, 2024

    Abhinav Rao, Akhila Yerukola, Vishwa Shah, Katharina Reinecke, and Maarten Sap. Normad: A benchmark for measuring the cultural adaptability of large language models.arXiv e-prints, pages arXiv–2404, 2024

    2024

  31. [39]

    Supercharging search with generative ai, 2023

    Elizabeth Reid. Supercharging search with generative ai, 2023. Accessed at https://blog. google/products/search/generative-ai-search/

    2023

  32. [40]

    Re-imagining algorithmic fairness in india and beyond, 2021

    Nithya Sambasivan, Erin Arnesen, Ben Hutchinson, Tulsee Doshi, and Vinodkumar Prabhakaran. Re-imagining algorithmic fairness in india and beyond, 2021

    2021

  33. [41]

    The global city.New York, 1991

    Saskia Sassen. The global city.New York, 1991

    1991

  34. [42]

    Contested histories: Eurocentrism, multiculturalism, and the media

    Robert Stam and Ella Shohat. Contested histories: Eurocentrism, multiculturalism, and the media. Multiculturalism: A critical reader, 296:324, 1994

    1994

  35. [43]

    Online travel market - statistics & facts, 2023

    Statista. Online travel market - statistics & facts, 2023. Accessed athttps://www.statista. com/topics/2704/online-travel-market

    2023

  36. [44]

    Commodifying africa on us network reality television.Communication, Culture & Critique, 1(4):416–446, 2008

    H Leslie Steeves. Commodifying africa on us network reality television.Communication, Culture & Critique, 1(4):416–446, 2008

    2008

  37. [45]

    John Wiley & Sons, 2015

    Deborah Tannen, Heidi E Hamilton, and Deborah Schiffrin.The handbook of discourse analysis. John Wiley & Sons, 2015

    2015

  38. [46]

    Cultural bias and cultural alignment of large language models.PNAS nexus, 3(9):pgae346, 2024

    Yan Tao, Olga Viberg, Ryan S Baker, and René F Kizilcec. Cultural bias and cultural alignment of large language models.PNAS nexus, 3(9):pgae346, 2024

    2024

  39. [47]

    Making news: A study in the construction of reality.Social Forces, 59, 01 1978

    Gaye Tuchman. Making news: A study in the construction of reality.Social Forces, 59, 01 1978

    1978

  40. [48]

    Can chatgpt plan your vacation? here’s what to know about a.i

    Julie Weed. Can chatgpt plan your vacation? here’s what to know about a.i. and travel., 2023. Accessed at https://www.nytimes.com/2023/03/16/travel/ chatgpt-artificial-intelligence-travel-vacation.html

    2023

  41. [49]

    art” and “spiritual

    Li Zhou, Antonia Karamolegkou, Wenyu Chen, and Daniel Hershcovich. Cultural compass: Predicting transfer learning success in offensive language detection with cultural features. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 12684–12702, 2023. ...

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.