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

Language embeds information about social, cultural, and political values people hold. Prior work has explored potentially harmful social biases encoded in Pre-trained Language Models (PLMs). However, there has been no systematic study investigating how values embedded in these models vary across cultures.In this paper, we introduce probes to study which cross-cultural values are embedded in these models, and whether they align with existing theories and cross-cultural values surveys. We find that PLMs capture differences in values across cultures, but those only weakly align with established values surveys. We discuss implications of using mis-aligned models in cross-cultural settings, as well as ways of aligning PLMs with values surveys.

[1]  Miryam de Lhoneux,et al.  Challenges and Strategies in Cross-Cultural NLP , 2022, ACL.

[2]  Rebecca Lynn Johnson,et al.  The Ghost in the Machine has an American accent: value conflict in GPT-3 , 2022, ArXiv.

[3]  Isabelle Augenstein,et al.  A Survey on Gender Bias in Natural Language Processing , 2021, ArXiv.

[4]  Maya Indira Ganesh,et al.  A Word on Machine Ethics: A Response to Jiang et al. (2021) , 2021, ArXiv.

[5]  Nigel Collier,et al.  Visually Grounded Reasoning across Languages and Cultures , 2021, EMNLP.

[6]  David Jurgens,et al.  Using Sociolinguistic Variables to Reveal Changing Attitudes Towards Sexuality and Gender , 2021, EMNLP.

[7]  Dan Goldwasser,et al.  Identifying Morality Frames in Political Tweets using Relational Learning , 2021, EMNLP.

[8]  David Lazer,et al.  (Mis)alignment Between Stance Expressed in Social Media Data and Public Opinion Surveys , 2021, EMNLP.

[9]  Christy Dennison,et al.  Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasets , 2021, NeurIPS.

[10]  Alexandra Luccioni,et al.  What’s in the Box? An Analysis of Undesirable Content in the Common Crawl Corpus , 2021, ACL.

[11]  Isabelle Augenstein,et al.  Quantifying gender bias towards politicians in cross-lingual language models , 2021, PloS one.

[12]  Masahiro Kaneko,et al.  Unmasking the Mask - Evaluating Social Biases in Masked Language Models , 2021, AAAI.

[13]  Emily M. Bender,et al.  On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜 , 2021, FAccT.

[14]  Daniel de Vassimon Manela,et al.  Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Models , 2021, EACL.

[15]  Graham Neubig,et al.  Word Alignment by Fine-tuning Embeddings on Parallel Corpora , 2021, EACL.

[16]  Adam Lopez,et al.  Intrinsic Bias Metrics Do Not Correlate with Application Bias , 2020, ACL.

[17]  Yejin Choi,et al.  Moral Stories: Situated Reasoning about Norms, Intents, Actions, and their Consequences , 2020, EMNLP.

[18]  Yejin Choi,et al.  Social Chemistry 101: Learning to Reason about Social and Moral Norms , 2020, EMNLP.

[19]  Graham Neubig,et al.  X-FACTR: Multilingual Factual Knowledge Retrieval from Pretrained Language Models , 2020, EMNLP.

[20]  Goran Glavas,et al.  Probing Pretrained Language Models for Lexical Semantics , 2020, EMNLP.

[21]  Marius Mosbach,et al.  On the Interplay Between Fine-tuning and Sentence-Level Probing for Linguistic Knowledge in Pre-Trained Transformers , 2020, BLACKBOXNLP.

[22]  Samuel R. Bowman,et al.  CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Models , 2020, EMNLP.

[23]  D. Song,et al.  Aligning AI With Shared Human Values , 2020, ICLR.

[24]  Aylin Caliskan,et al.  Detecting Emergent Intersectional Biases: Contextualized Word Embeddings Contain a Distribution of Human-like Biases , 2020, AIES.

[25]  Siva Reddy,et al.  StereoSet: Measuring stereotypical bias in pretrained language models , 2020, ACL.

[26]  T. Jackson.  The legacy of Geert Hofstede , 2020 .

[27]  Anna Rumshisky,et al.  A Primer in BERTology: What We Know About How BERT Works , 2020, Transactions of the Association for Computational Linguistics.

[28]  Inioluwa Deborah Raji,et al.  Closing the AI accountability gap: defining an end-to-end framework for internal algorithmic auditing , 2020, FAT*.

[29]  Natalia Gimelshein,et al.  PyTorch: An Imperative Style, High-Performance Deep Learning Library , 2019, NeurIPS.

[30]  Noah A. Smith,et al.  Social Bias Frames: Reasoning about Social and Power Implications of Language , 2019, ACL.

[31]  Myle Ott,et al.  Unsupervised Cross-lingual Representation Learning at Scale , 2019, ACL.

[32]  Davis Liang,et al.  Masked Language Model Scoring , 2019, ACL.

[33]  Thomas Wolf,et al.  HuggingFace's Transformers: State-of-the-art Natural Language Processing , 2019, ArXiv.

[34]  Sameer Singh,et al.  Do NLP Models Know Numbers? Probing Numeracy in Embeddings , 2019, EMNLP.

[35]  Sebastian Riedel,et al.  Language Models as Knowledge Bases? , 2019, EMNLP.

[36]  Omer Levy,et al.  RoBERTa: A Robustly Optimized BERT Pretraining Approach , 2019, ArXiv.

[37]  Christopher D. Manning,et al.  A Structural Probe for Finding Syntax in Word Representations , 2019, NAACL.

[38]  Aida Mostafazadeh Davani,et al.  Moral Foundations Twitter Corpus: A Collection of 35k Tweets Annotated for Moral Sentiment , 2019, Social Psychological and Personality Science.

[39]  Chandler May,et al.  On Measuring Social Biases in Sentence Encoders , 2019, NAACL.

[40]  David Laniado,et al.  Wikipedia Cultural Diversity Dataset: A Complete Cartography for 300 Language Editions , 2019, ICWSM.

[41]  Guillaume Lample,et al.  Cross-lingual Language Model Pretraining , 2019, NeurIPS.

[42]  Eric Fleury,et al.  Socioeconomic Dependencies of Linguistic Patterns in Twitter: a Multivariate Analysis , 2018, WWW.

[43]  Daniel Jurafsky,et al.  Word embeddings quantify 100 years of gender and ethnic stereotypes , 2017, Proceedings of the National Academy of Sciences.

[44]  Yulia Tsvetkov,et al.  Writer Profiling Without the Writer's Text , 2017, SocInfo.

[45]  Yejin Choi,et al.  Connotation Frames of Power and Agency in Modern Films , 2017, EMNLP.

[46]  Joanna Bryson,et al.  Semantics derived automatically from language corpora contain human-like biases , 2016, Science.

[47]  Phil Blunsom,et al.  Teaching Machines to Read and Comprehend , 2015, NIPS.

[48]  Dirk Hovy,et al.  User Review Sites as a Resource for Large-Scale Sociolinguistic Studies , 2015, WWW.

[49]  Brendan T. O'Connor,et al.  Diffusion of Lexical Change in Social Media , 2012, PloS one.

[50]  Gail M. Sullivan,et al.  Using Effect Size-or Why the P Value Is Not Enough. , 2012, Journal of graduate medical education.

[51]  Eldad Davidov,et al.  Refining the theory of basic individual values. , 2012, Journal of personality and social psychology.

[52]  K. Hew,et al.  Cross-cultural analysis of the Wikipedia community , 2010, J. Assoc. Inf. Sci. Technol..

[53]  A. Jaffe.  Stance: Sociolinguistic Perspectives , 2009 .

[54]  Peter Trudgill,et al.  Sociolinguistic Variation and Change , 2001 .

[55]  B. Norton.  Language, Identity, and the Ownership of English , 1997 .

[56]  G. Hofstede,et al.  Culture′s Consequences: International Differences in Work-Related Values , 1980 .

[57]  Wilson L. Taylor,et al.  “Cloze Procedure”: A New Tool for Measuring Readability , 1953 .

[58]  Johannes Kiesel,et al.  Identifying the Human Values behind Arguments , 2022, ACL.

[59]  Dit-Yan Yeung,et al.  Probing Toxic Content in Large Pre-Trained Language Models , 2021, ACL.

[60]  Ming-Wei Chang,et al.  BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , 2019, NAACL.

[61]  M. L. Jones.  Hofstede - Culturally questionable? , 2007 .

[62]  Geert Hofstede,et al.  Culture's Recent Consequences , 2005, IWIPS.

[63]  Sydney Gregory,et al.  Culture's consequences: international differences in work-related values , 1982 .

[64]  W. Labov.  The social motivation of a sound change , 1963 .

[65]  Dragomir R. Radev,et al.  of the Association for Computational Linguistics , 2022 .