Elamur Research⌕
☰
home / papers / 2023 / Probing Pre-Trained Language Models for Cross-Cultural Differences in Values

Probing Pre-Trained Language Models for Cross-Cultural Differences in Values

Summary

The paper asks whether multilingual pre-trained language models (PLMs) absorb culture-specific values from their Web training text, and whether those values match what social scientists have measured in real populations. The authors build probes from two established instruments: Hofstede's six cultural dimensions and the World Values Survey (WVS). They semi-automatically rewrite the survey questions as cloze-style mask probes and translate them from English into 13 languages. Each language is chosen because it is largely tied to a single country or region, so it can stand in for that culture. They then query mBERT, XLM and XLM-R and compare the models' predicted answers against the surveys' population-level scores.

The models do produce noticeably different value profiles across languages. However, these differences correlate only weakly with the survey data. This matters for anyone deploying multilingual models across cultures: such models may quietly misrepresent local values or impose a single normative view. The authors argue for culturally diverse training data and for auditing that takes the downstream use into account.

Key points

  • Three multilingual models are probed: mBERT, XLM and XLM-R. Probes cover 13 languages, each mostly confined to one country or region.
  • Hofstede's framework has six dimensions: pdi, idv, uai, mas, lto and ivr. Each is computed by a fixed formula from 4 of the 24 survey questions. The original survey covered 116,000 IBM employees in 40 countries.
  • The WVS data comes from wave 7 (2017–2020) and covers 57 countries, with results reported per question in 13 categories. Two categories, Economic Values and the Postmaterialist Index, were dropped because their questions could not be turned into mask probes without losing information.
  • The models do differ in the values they express across cultures, but their agreement with the surveys is weak.
  • The probes are released at github.com/copenlu/value-probing.

Interesting ideas

  • I1. Turning questionnaires into cloze probes. Converting validated survey items into fill-the-mask sentences reuses instruments that experts already designed and tested. The same approach could audit models against other questionnaires, such as opinion polls or attitude scales.
  • I2. Language as a proxy for culture. Selecting languages spoken mostly in one country makes it possible to link model outputs to country-level survey scores. The authors are open that this mapping is imperfect, and the same trade-off applies to any country-level comparison of models.
  • I3. Using two frameworks instead of one. Pairing the coarse, widely criticised Hofstede dimensions with the more detailed per-question WVS protects conclusions from depending on one theory's simplifications. Any evaluation built on a contested framework could benefit from this kind of cross-check.
  • I4. Choosing a scoring method that fits the probes. The authors prefer cloze-style probing over pseudo-log-likelihood scoring. Their reasoning is that the probe wording was deliberately crafted by social scientists, so wording-induced bias is less of a concern than PLL's assumption that masked tokens are independent. This shows that the scoring method should follow from where the probe text comes from.
  • I5. Testing robustness with synonyms, and reporting why it failed. The authors tried expanding each label word into a set of synonyms to make results more robust. This proved hard: substitutions broke grammar, matching senses were missing in some languages, and many synonyms were multi-word expressions. These pitfalls are worth anticipating in any single-mask probing setup, especially for morphologically rich languages.

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

Fields: Natural language processing · Society & ethics

Cited by

98 papers of our library cite it (of the 140884 whose bibliographies we read); 10 of them have a page here, 88 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.c3nlp-1.12

@inproceedings{arora2023probing,
  title = {{Probing Pre-Trained Language Models for Cross-Cultural Differences in Values}},
  author = {Arora, Arnav and Kaffee, Lucie-aimée and Augenstein, Isabelle},
  booktitle = {Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP)},
  year = {2023},
  doi = {10.18653/v1/2023.c3nlp-1.12},
  url = {https://doi.org/10.18653/v1/2023.c3nlp-1.12}
}

References

70 references, 16 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jacob Levy Abitbol, Márton Karsai, Jean-Philippe Magué, Jean-Pierre Chevrot, and Eric Fleury. 2018. Socioeconomic dependencies of linguistic patterns in twitter: A multivariate analysis. In Proceedings of the 2018 World Wide Web Conference, WWW ’18, page 1125–1134, Republic and Canton of Geneva, CHE… find
  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, FAccT ’21, page 610–623, New York, NY, USA. Associat… find
  3. Aylin Caliskan, Joanna J. Bryson, and Arvind Narayanan. 2017. Semantics derived automatically from language corpora contain human-like biases. Science, 356(6334):183–186. find
  4. Sky CH-Wang and David Jurgens. 2021. Using sociolinguistic variables to reveal changing attitudes towards sexuality and gender. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9918–9938, Online and Punta Cana, Dominican Republic. Association for Compu… find
  5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association fo… find
  6. Alexis Conneau and Guillaume Lample. 2019. Cross-lingual language model pretraining. In Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc. find
  7. Daniel de Vassimon Manela, David Errington, Thomas Fisher, Boris van Breugel, and Pasquale Minervini. 2021. Stereotype and skew: Quantifying gender bias in pre-trained and fine-tuned language models. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational L… find
  8. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. find
  9. Zi-Yi Dou and Graham Neubig. 2021. Word alignment by fine-tuning embeddings on parallel corpora. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, EACL 2021, Online, April 19 - 23, 2021, pages 2112–2128. Association for Compu… find
  10. Jacob Eisenstein, Brendan O’Connor, Noah A. Smith, and Eric P. Xing. 2014. Diffusion of lexical change in social media. PLOS ONE, 9(11):1–13. find
  11. Denis Emelin, Ronan Le Bras, Jena D. Hwang, Maxwell Forbes, and Yejin Choi. 2021. Moral stories: Situated reasoning about norms, intents, actions, and their consequences. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 698–718, Online and Punta Cana,… find
  12. Maxwell Forbes, Jena D. Hwang, Vered Shwartz, Maarten Sap, and Yejin Choi. 2020. Social chemistry 101: Learning to reason about social and moral norms. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 653–670, Online. Association for Computatio… find
  13. Nikhil Garg, Londa Schiebinger, Dan Jurafsky, and James Zou. 2018. Word embeddings quantify 100 years of gender and ethnic stereotypes. Proceedings of the National Academy of Sciences, 115(16):E3635– E3644. find
  14. Seraphina Goldfarb-Tarrant, Rebecca Marchant, Ricardo Muñoz Sánchez, Mugdha Pandya, and Adam Lopez. 2021. Intrinsic bias metrics do not correlate with application bias. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Confere… find our card
  15. Wei Guo and Aylin Caliskan. 2021. Detecting emergent intersectional biases: Contextualized word embeddings contain a distribution of human-like biases. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, AIES ’21, page 122–133, New York, NY, USA. Association for Computing Mach… find
  16. C. Haerpfer, R. Inglehart, A. Moreno, C. Welzel, K. Kizilova, M. Lagos Diez-Medrano J., P. Norris, E. Ponarin, and B. Puranen. 2022. World values survey: Round seven - country-pooled datafile version 3.0. Madrid, Spain & Vienna, Austria: JD Systems Institute & WVSA Secretariat. find
  17. Nina Hansen and Luzia Heu. 2020. All human, yet different. Social Psychology, 51(6):361–369. find
  18. Noriko Hara, Pnina Shachaf, and Khe Foon Hew. 2010. Cross-cultural analysis of the wikipedia community. J. Assoc. Inf. Sci. Technol., 61(10):2097–2108. find
  19. Dan Hendrycks, Collin Burns, Steven Basart, Andrew Critch, Jerry Li, Dawn Song, and Jacob Steinhardt. 2021. Aligning {ai} with shared human values. In International Conference on Learning Representations. find our card
  20. Karl Moritz Hermann, Tomás Kociský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. 2015. Teaching machines to read and comprehend. CoRR, abs/1506.03340. find our card
  21. 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 n… find our card
  22. John Hewitt and Christopher D. Manning. 2019. A structural probe for finding syntax in word representations. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4129… find our card
  23. Geert Hofstede. 1984. Culture’s consequences: International differences in work-related values, volume 5. sage. find
  24. Geert Hofstede. 2005. Culture’s recent consequences. In Designing for Global Markets 7, IWIPS 2005, Bridging Cultural Differences, 7-9 July 2005, Amsterdam, The Netherlands, Proceedings of the Seventh International Workshop on Internationalisation of Products and Systems, pages 3–4. Product & System… find
  25. Joe Hoover, Gwenyth Portillo-Wightman, Leigh Yeh, Shreya Havaldar, Aida Mostafazadeh Davani, Ying Lin, Brendan Kennedy, Mohammad Atari, Zahra Kamel, Madelyn Mendlen, Gabriela Moreno, Christina Park, Tingyee E. Chang, Jenna Chin, Christian Leong, Jun Yen Leung, Arineh Mirinjian, and Morteza Dehghani.… find
  26. Dirk Hovy, Anders Johannsen, and Anders Søgaard. 2015. User review sites as a resource for large-scale sociolinguistic studies. In Proceedings of the 24th International Conference on World Wide Web, WWW ’15, page 452–461, Republic and Canton of Geneva, CHE. International World Wide Web Conferences S… find
  27. Ronald Inglehart. 2006. Inglehart-welzel cultural map of the world. World Values Survey. find
  28. Languages and Cultures. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10467–10485, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics. find
  29. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. ArXiv, abs/1907.11692. find
  30. Terence Jackson. 2020. The legacy of geert hofstede. find
  31. Alexandra Jaffe. 2009. Stance: Sociolinguistic Perspectives. Oxford University Press. find
  32. Zhengbao Jiang, Antonios Anastasopoulos, Jun Araki, Haibo Ding, and Graham Neubig. 2020. X-FACTR: Multilingual Factual Knowledge Retrieval from Pre-trained Language Models. arXiv:2010.06189 [cs]. ArXiv: 2010.06189. — [arXiv:2010.06189](https://arxiv.org/abs/2010.06189) find
  33. Rebecca L Johnson, Giada Pistilli, Natalia Menédez-González, Leslye Denisse Dias Duran, Enrico Panai, Julija Kalpokiene, and Donald Jay Bertulfo. 2022. The ghost in the machine has an american accent: value conflict in gpt-3. find
  34. Michael L Jones. 2007. Hofstede-culturally questionable? find
  35. Kenneth Joseph, Sarah Shugars, Ryan J. Gallagher, Jon Green, Alexi Quintana Mathé, Zijian An, and David Lazer. 2021. (mis)alignment between stance expressed in social media data and public opinion surveys. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, EMN… find
  36. David Jurgens, Yulia Tsvetkov, and Dan Jurafsky. 2017. Writer profiling without the writer’s text. In Social Informatics, pages 537–558, Cham. Springer International Publishing. find
  37. Masahiro Kaneko and Danushka Bollegala. 2022. Unmasking the mask – evaluating social biases in masked language models. Proceedings of the AAAI Conference on Artificial Intelligence, 36(11):11954– 11962. find
  38. Johannes Kiesel, Milad Alshomary, Nicolas Handke, Xiaoni Cai, Henning Wachsmuth, and Benno Stein. 2022. Identifying the human values behind arguments. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4459–4471, Dublin, Ireland.… find
  39. Alexandra Luccioni and Joseph Viviano. 2021. What’s in the box? an analysis of undesirable content in the Common Crawl corpus. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume… find
  40. Chandler May, Alex Wang, Shikha Bordia, Samuel R. Bowman, and Rachel Rudinger. 2019. On measuring social biases in sentence encoders. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and… find our card
  41. Marc Miquel-Ribé and David Laniado. 2019. Wikipedia cultural diversity dataset: A complete cartography for 300 language editions. In Proceedings of the Thirteenth International Conference on Web and Social Media, ICWSM 2019, Munich, Germany, June 11-14, 2019, pages 620–629. AAAI Press. find
  42. Marius Mosbach, Anna Khokhlova, Michael A. Hedderich, and Dietrich Klakow. 2020. On the interplay between fine-tuning and sentence-level probing for linguistic knowledge in pre-trained transformers. In Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NL… find
  43. Moin Nadeem, Anna Bethke, and Siva Reddy. 2021. StereoSet: Measuring stereotypical bias in pretrained language models. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long… find our card
  44. William Labov. 1963. The social motivation of a sound change. \<i\>WORD\</i\>, 19(3):273–309. find
  45. Nikita Nangia, Clara Vania, Rasika Bhalerao, and Samuel R. Bowman. 2020. CrowS-pairs: A challenge dataset for measuring social biases in masked language models. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1953–1967, Online. Association for… find our card
  46. Fangyu Liu, Emanuele Bugliarello, Edoardo Maria Ponti, Siva Reddy, Nigel Collier, and Desmond Elliott. 2021. Visually Grounded Reasoning across Ercan G Nasif, Hamad Al-Daeaj, Bahman Ebrahimi, and Mary S Thibodeaux. 1991. Methodological problems in cross-cultural research: An updated review. find
  47. MIR: Management International Review, pages 79– 91. find
  48. Bonny Norton. 1997. Language, identity, and the ownership of english. TESOL Quarterly, 31(3):409–429. find
  49. Nedjma Ousidhoum, Xinran Zhao, Tianqing Fang, Yangqiu Song, and Dit-Yan Yeung. 2021. Probing toxic content in large pre-trained language models. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language… find
  50. Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Te-jani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and… find our card
  51. Fabio Petroni, Tim Rocktäschel, Sebastian Riedel, Patrick Lewis, Anton Bakhtin, Yuxiang Wu, and Alexander Miller. 2019. Language models as knowledge bases? In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural… find our card
  52. Inioluwa Deborah Raji, Andrew Smart, Rebecca N. White, Margaret Mitchell, Timnit Gebru, Ben Hutchinson, Jamila Smith-Loud, Daniel Theron, and Parker Barnes. 2020. Closing the ai accountability gap: Defining an end-to-end framework for internal algorithmic auditing. In Proceedings of the 2020 Confere… find
  53. Anna Rogers, Olga Kovaleva, and Anna Rumshisky. 2020. A primer in BERTology: What we know about how BERT works. Transactions of the Association for Computational Linguistics, 8:842–866. find our card
  54. Milton Rokeach. 2008. Understanding human values. Simon and Schuster. find
  55. Shamik Roy, Maria Leonor Pacheco, and Dan Goldwasser. 2021. Identifying morality frames in political tweets using relational learning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9939–9958, Online and Punta Cana, Dominican Republic. Association fo… find
  56. Julian Salazar, Davis Liang, Toan Q. Nguyen, and Katrin Kirchhoff. 2020. Masked language model scoring. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 2699–2712, Online. Association for Computational Linguistics. find our card
  57. Maarten Sap, Saadia Gabriel, Lianhui Qin, Dan Jurafsky, Noah A. Smith, and Yejin Choi. 2020. Social bias frames: Reasoning about social and power implications of language. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5477–5490, Online. Association… find our card
  58. Maarten Sap, Marcella Cindy Prasettio, Ari Holtzman, Hannah Rashkin, and Yejin Choi. 2017. Connotation frames of power and agency in modern films. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2329–2334, Copenhagen, Denmark. Association for Computat… find
  59. Shalom H. Schwartz, Jan Cieciuch, Michele Vecchione, Eldad Davidov, Ronald Fischer, Constanze Beierlein, Alice Ramos, Markku Verkasalo, Jan-Erik Lönnqvist, Kursad Demirutku, Ozlem Dirilen-Gumus, and Mark Konty. 2012. Refining the theory of basic individual values. Journal of Personality and Social P… find
  60. Irene Solaiman and Christy Dennison. 2021. Process for adapting language models to society (palms) with values-targeted datasets. In Advances in Neural Information Processing Systems, volume 34, pages 5861– 5873. Curran Associates, Inc. find
  61. Karolina Stanczak and Isabelle Augenstein. 2021. A survey on gender bias in natural language processing. CoRR, abs/2112.14168. — [arXiv:2112.14168](https://arxiv.org/abs/2112.14168) find
  62. Karolina Stanczak, Sagnik Ray Choudhury, Tiago Pimentel, Ryan Cotterell, and Isabelle Augenstein. 2021. Quantifying gender bias towards politicians in cross-lingual language models. CoRR, abs/2104.07505. — [arXiv:2104.07505](https://arxiv.org/abs/2104.07505) find
  63. Karolina Stańczak, Sagnik Ray Choudhury, Tiago Pimentel, Ryan Cotterell, and Isabelle Augenstein. 2021. Quantifying Gender Bias Towards Politicians in Cross-Lingual Language Models. arXiv:2104.07505 [cs, stat]. ArXiv: 2104.07505. — [arXiv:2104.07505](https://arxiv.org/abs/2104.07505) find
  64. Gail M Sullivan and Richard Feinn. 2012. Using effect size—or why the p value is not enough. Journal of graduate medical education, 4(3):279–282. find
  65. Zeerak Talat, Hagen Blix, Josef Valvoda, Maya Indira Ganesh, Ryan Cotterell, and Adina Williams. 2021. A Word on Machine Ethics: A Response to Jiang et al. (2021). arXiv:2111.04158 [cs]. ArXiv: 2111.04158. — [arXiv:2111.04158](https://arxiv.org/abs/2111.04158) find
  66. Wilson L Taylor. 1953. “cloze procedure”: A new tool for measuring readability. Journalism Quarterly, 30(4):415–433. find
  67. Peter Trudgill. 2002. Sociolinguistic Variation and Change. Edinburgh University Press. find
  68. Ivan Vulić, Edoardo Maria Ponti, Robert Litschko, Goran Glavaš, and Anna Korhonen. 2020. Probing pretrained language models for lexical semantics. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 7222–7240, Online. Association for Computationa… find our card
  69. Eric Wallace, Yizhong Wang, Sujian Li, Sameer Singh, and Matt Gardner. 2019. Do NLP models know numbers? probing numeracy in embeddings. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing… find our card
  70. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lho… find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/arora-2023-probing-pre-trained-language-models/