Multi Dimensional Gender Bias Classification

EMNLP 2020, pp. 314-331, 2020.

Cited by: 0|Bibtex|Views131|DOI:https://doi.org/10.18653/V1/2020.EMNLP-MAIN.23
Other Links: arxiv.org|academic.microsoft.com
Weibo:
We propose a general framework that decomposes gender bias in text along several pragmatic and semantic dimensions: bias from the gender of the person being spoken about, bias from the gender of the person being spoken to, and bias from the gender of the speaker

Abstract:

Machine learning models are trained to find patterns in data. NLP models can inadvertently learn socially undesirable patterns when training on gender biased text. In this work, we propose a novel, general framework that decomposes gender bias in text along several pragmatic and semantic dimensions: bias from the gender of the person bein...More

Code:

Data:

0
Introduction
  • Language is a social behavior, and as such, it is a primary means by which people communicate, express their identities, and socially categorize themselves and others.
  • Such social information is present in the words the authors write and, in the text the authors use to train the NLP models.
Highlights
  • Language is a social behavior, and as such, it is a primary means by which people communicate, express their identities, and socially categorize themselves and others. Such social information is present in the words we write and, in the text we use to train our NLP models
  • We make four main contributions: we propose a multi-dimensional framework (ABOUT, AS, TO) for measuring and mitigating gender bias in language and NLP models, we introduce an evaluation dataset for performing gender identification that contains utterances re-written from the perspective of a specific gender along all three dimensions, we train a suite of classifiers capable of labeling gender in both a single and multitask set up, and we illustrate our classifiers’ utility for several downstream applications
  • We provide a clean new way to understand gender bias that extends to the dialogue use-case by independently investigating the contribution of author gender to data created by humans
  • We propose a framework for decomposing gender bias into three separate dimensions: bias when speaking ABOUT someone, bias when speaking TO someone, and bias from speaking AS someone
  • We propose a general framework for analyzing gender bias in text by decomposing it along three dimensions: (1) gender of the person or people being spoken about (ABOUT), (2) gender of the addressee (TO), and (2) gender of the speaker (AS)
Methods
  • The authors generate training data by taking the multi-task classifier and using it to classify 250,000 textual utterances from Reddit, using a previously existing dataset extracted and obtained by a third party and made available on pushshift.io.
  • The authors calculate a masculine genderedness score for the page by taking the median among all paragraphs in the page
  • For this application, the authors use the Standard training and evaluation dataset created and described in Dinan et al (2019b).
  • The authors measure the ratio of utterances labeled as masculine-gendered among utterances labeled as either masculine- or feminine-gendered
Results
  • Example generations from various control tokens are shown in Table 10 in the Appendix.
  • The authors observe that for the control tokens TO:feminine and AS:feminine, the utterances contain a roughly equal number of masculine-gendered and femininegendered words.
  • This is likely due to the distribution of such gendered words in the training data for the classifier in the to and as dimensions.
  • For words classified as masculine, 25% of the masculine words fell into these categories , whereas for words classified as feminine, 75% of the words fell into these categories
Conclusion
  • The authors propose a general framework for analyzing gender bias in text by decomposing it along three dimensions: (1) gender of the person or people being spoken about (ABOUT), (2) gender of the addressee (TO), and (2) gender of the speaker (AS).
  • The authors show that classifiers can detect bias along each of these dimensions.
  • The authors demonstrate the broad utility of the classifiers by showing strong performance on controlling bias in generated dialogue, detecting genderedness in text such as Wikipedia, and highlighting gender differences in offensive text classification
Summary
  • Introduction:

    Language is a social behavior, and as such, it is a primary means by which people communicate, express their identities, and socially categorize themselves and others.
  • Such social information is present in the words the authors write and, in the text the authors use to train the NLP models.
  • Methods:

    The authors generate training data by taking the multi-task classifier and using it to classify 250,000 textual utterances from Reddit, using a previously existing dataset extracted and obtained by a third party and made available on pushshift.io.
  • The authors calculate a masculine genderedness score for the page by taking the median among all paragraphs in the page
  • For this application, the authors use the Standard training and evaluation dataset created and described in Dinan et al (2019b).
  • The authors measure the ratio of utterances labeled as masculine-gendered among utterances labeled as either masculine- or feminine-gendered
  • Results:

    Example generations from various control tokens are shown in Table 10 in the Appendix.
  • The authors observe that for the control tokens TO:feminine and AS:feminine, the utterances contain a roughly equal number of masculine-gendered and femininegendered words.
  • This is likely due to the distribution of such gendered words in the training data for the classifier in the to and as dimensions.
  • For words classified as masculine, 25% of the masculine words fell into these categories , whereas for words classified as feminine, 75% of the words fell into these categories
  • Conclusion:

    The authors propose a general framework for analyzing gender bias in text by decomposing it along three dimensions: (1) gender of the person or people being spoken about (ABOUT), (2) gender of the addressee (TO), and (2) gender of the speaker (AS).
  • The authors show that classifiers can detect bias along each of these dimensions.
  • The authors demonstrate the broad utility of the classifiers by showing strong performance on controlling bias in generated dialogue, detecting genderedness in text such as Wikipedia, and highlighting gender differences in offensive text classification
Tables
  • Table1: Bias in Wikipedia. We look at the most over-represented words in biographies of men and women, respectively, in Wikipedia. We also compare to a set of over-represented words in gender-neutral pages. We use a part-of-speech tagger (<a class="ref-link" id="cHonnibal_2017_b" href="#rHonnibal_2017_b">Honnibal and Montani, 2017</a>) and limit our analysis to words that appear at least 500 times
  • Table2: Dataset Statistics. Dataset statistics on the eight training datasets and new evaluation dataset, MDGENDERwith respect to each label
  • Table3: Accuracy on the novel evaluation dataset MDGENDER comparing single task classifiers to our multitask classifiers. We report accuracy on the masculine and the feminine classes, as well as the average of these two metrics. Finally, we report the average (of the M-F averages) across the three dimensions. MDGENDERwas collected to enable evaluation on the masculine and femninine classes, for which much of the training data is noisy
  • Table4: Performance of the multitask model on the test sets from our training data. We evaluate the multi-task model on the test sets for the training datasets. We report accuracy on each (gold) label— masculine, feminine, and neutral—and the average of the three. We do not report accuracy on imputed labels
  • Table5: Ablation of gender classifiers on the Wikipedia test set. We report the model accuracy on the masculine, feminine, and neutral classes, as well as the average accuracy across them. We train classifiers (1) on the entire text (2) after removing explicitly gendered words using a word list and (3) after removing gendered words and names. While masking out gendered words and names makes classification more challenging, the model still obtains high accuracy
  • Table6: Word statistics measured on text generated from 1000 different seed utterances from ConvAI2 for each control token, as well as for our baseline model trained using word lists. We measure the number of gendered words (from a word list) that appear in the generated text as well as the percentage of masculinegendered words among all gendered words. Sequences are generated with top-k sampling, k = 10, with a beam size of 10 and 3-gram blocking
  • Table7: Genderedness of offensive content. We measure the percentage of utterances in both the ”safe” and ”offensive” classes that are classified as masculinegendered, among utterances that are classified as either masculine- or feminine-gendered. We test the hypothesis that safe and offensive classes distributions of masculine-gendered utterances differ using a t-test and report the p-value for each dimension
  • Table8: Masculine genderedness scores of Wikipedia bios. We calculate a masculine genderedness score for a Wikipedia page by taking the median px = P (x ∈ ABOUT:masculine) among all paragraphs x in the page, where P is the probability distribution given by the classifier. We report the average and median scores for all biographies, as well as for biographies of men and women respectively
  • Table9: Examples from the MDGENDER. Crowdworkers were asked to re-write dialogue utterances such that most people would guess that the utterance was either said to, said by, or about a man or a woman. Afterwards, they were asked to give a confidence level in their re-write, meant to capture the differences between statistical biases (more men play football than women) and fact (you do not have to be a man to play football)
  • Table10: Example generations from a generative model trained using controllable generation, with control tokens determined by the classifier. Sequences are generated with top-k sampling, k = 10, with a beam size of 10 and 3-gram blocking. Input is randomly sampled from the ConvAI2 dataset
  • Table11: Most gendered Wikipedia biographies We ran our multi-task classifier over 68 thousand biographies of Wikipedia. After selecting for biographies with a minimum number of paragraphs (resulting in 15.5 thousand biographies) we scored them to determine the most masculine and feminine gendered
Download tables as Excel
Related work
Study subjects and analysis
large scale datasets with gender information: 8
In this work, we propose a general framework that decomposes gender bias in text along several pragmatic and semantic dimensions: bias from the gender of the person being spoken about, bias from the gender of the person being spoken to, and bias from the gender of the speaker. Using this fine-grained framework, we automatically annotate eight large scale datasets with gender information. In addition, we collect a novel, crowdsourced evaluation benchmark of utterance-level gender rewrites

existing datasets: 8
4.2 Data. Next, we describe how we annotated our training data, including both the 8 existing datasets and our novel evaluation dataset, MDGENDER. Annotation of Existing Datasets

large existing datasets: 8
We show that classifiers can detect bias along each of these dimensions. We annotate eight large existing datasets along our dimensions, and also contribute a high quality evaluation dataset for this task. We demonstrate the broad utility of our classifiers by showing strong performance on controlling bias in generated dialogue, detecting genderedness in text such as Wikipedia, and highlighting gender differences in offensive text classification

training datasets: 8
Bias in Wikipedia. We look at the most over-represented words in biographies of men and women, respectively, in Wikipedia. We also compare to a set of over-represented words in gender-neutral pages. We use a part-of-speech tagger (Honnibal and Montani, 2017) and limit our analysis to words that appear at least 500 times. Dataset Statistics. Dataset statistics on the eight training datasets and new evaluation dataset, MDGENDERwith respect to each label. Accuracy on the novel evaluation dataset MDGENDER comparing single task classifiers to our multitask classifiers. We report accuracy on the masculine and the feminine classes, as well as the average of these two metrics. Finally, we report the average (of the M-F averages) across the three dimensions. MDGENDERwas collected to enable evaluation on the masculine and femninine classes, for which much of the training data is noisy

Reference
  • Fran Amery, Stephen Bates, Laura Jenkins, and Heather Savigny. 2015. Metaphors on women in academia: A review of the literature, 2004-2013. At the center: Feminism, social science and knowledge, 20:247A 267.
    Google ScholarFindings
  • Shlomo Argamon, Moshe Koppel, James W Pennebaker, and Jonathan Schler. 2009. Automatically profiling the author of an anonymous text. Communications of the ACM, 52(2):119–123.
    Google ScholarLocate open access versionFindings
  • David Bamman and Noah A Smith. 2014. Unsupervised discovery of biographical structure from text. Transactions of the Association for Computational Linguistics, 2:363–376.
    Google ScholarLocate open access versionFindings
  • Mahzarin R. Banaji and Deborah A. Prentice. 199The self in social contexts. Annual review of psychology, 45(1):297–332.
    Google ScholarLocate open access versionFindings
  • Solon Barocas, Moritz Hardt, and Arvind Narayanan. 2020. Fairness in machine learning: Limitations and Opportunities.
    Google ScholarFindings
  • Christine Basta, Marta R Costa-jussa, and Noe Casas. 2019. Evaluating the underlying gender bias in contextualized word embeddings. In Proceedings of the 1st Workshop on Gender Bias in Natural Language Processing.
    Google ScholarLocate open access versionFindings
  • Allan Bell. 1984. Language style as audience design. Language in society, 13(2):145–204.
    Google ScholarLocate open access versionFindings
  • Tolga Bolukbasi, Kai-Wei Chang, James Y Zou, Venkatesh Saligrama, and Adam T Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Advances in neural information processing systems, pages 4349–4357.
    Google ScholarLocate open access versionFindings
  • Joy Buolamwini and Timnit Gebru. 2018. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Proceedings of the 1st Conference on Fairness, Accountability and Transparency, volume 81 of Proceedings of Machine Learning Research, pages 77–91, New York, NY, USA. PMLR.
    Google ScholarLocate open access versionFindings
  • Kay Bussey. 1986. The first socialization. In Australian women: New feminist perspectives, pages 90–104. Oxford University Press.
    Google ScholarFindings
  • Judith Butler. 1990. Gender trouble, feminist theory, and psychoanalytic discourse. Routledge New York.
    Google ScholarFindings
  • Alyson Byrne and Julian Barling. 2017. When she brings home the job status: Wives job status, status leakage, and marital instability. Organization Science, 28(2):177–192.
    Google ScholarLocate open access versionFindings
  • Aylin Caliskan, Joanna J Bryson, and Arvind Narayanan. 2017. Semantics derived automatically from language corpora contain human-like biases. Science, 356(6334):183–186.
    Google ScholarLocate open access versionFindings
  • Deborah Cameron. 1990. The feminist critique of language: A reader.
    Google ScholarFindings
  • Yang Trista Cao and Hal Daume. 2019. Toward genderinclusive coreference resolution. arXiv preprint arXiv:1910.13913.
    Findings
  • Kaytlin Chaloner and Alfredo Maldonado. 2019. Measuring gender bias in word embeddings across domains and discovering new gender bias word categories. In Proceedings of the First Workshop on Gender Bias in Natural Language Processing, pages 25–32.
    Google ScholarLocate open access versionFindings
  • Kai-Wei Chang, Vinod Prabhakaran, and Vicente Ordonez. 2019. Bias and fairness in natural language processing. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP): Tutorial Abstracts, Hong Kong, China. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Christine Charyton and Glenn E Snelbecker. 2007. Engineers’ and musicians’ choices of self-descriptive adjectives as potential indicators of creativity by gender and domain. Psychology of Aesthetics, creativity, and the arts, 1(2):91.
    Google ScholarLocate open access versionFindings
  • Allan Bell and Gary Johnson. 1997. Towards a so- Na Cheng, Rajarathnam Chandramouli, and KP Subciolinguistics of style. University of Pennsylvania balakshmi. 2011. Author gender identification from
    Google ScholarLocate open access versionFindings
  • Jennifer Coates. 2015.
    Google ScholarFindings
  • Marta R Costa-jussa. 2019. An analysis of gender bias studies in natural language processing. Nature Machine Intelligence, pages 1–2.
    Google ScholarLocate open access versionFindings
  • Mary Crawford. 1995. Talking difference: On gender and language. Sage.
    Google ScholarFindings
  • Kimberle Crenshaw. 1989. Demarginalizing the intersection of race and sex: A black feminist critique of antidiscrimination doctrine, feminist theory and antiracist politics. u. Chi. Legal f., page 139.
    Google ScholarLocate open access versionFindings
  • Stefania Degaetano-Ortlieb. 2018. Stylistic variation over 200 years of court proceedings according to gender and social class. In Proceedings of the Second Workshop on Stylistic Variation, pages 1–10, New Orleans. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. CoRR, abs/1810.04805.
    Findings
  • Emily Dinan, Angela Fan, Adina Williams, Jack Urbanek, Douwe Kiela, and Jason Weston. 2019a. Queens are powerful too: Mitigating gender bias in dialogue generation.
    Google ScholarFindings
  • Emily Dinan, Samuel Humeau, Bharath Chintagunta, and Jason Weston. 2019b. Build it break it fix it for dialogue safety: Robustness from adversarial human attack. arXiv preprint arXiv:1908.06083.
    Findings
  • Emily Dinan, Varvara Logacheva, Valentin Malykh, Alexander Miller, Kurt Shuster, Jack Urbanek, Douwe Kiela, Arthur Szlam, Iulian Serban, Ryan Lowe, et al. 2019c. The second conversational intelligence challenge (convai2). arXiv preprint arXiv:1902.00098.
    Findings
  • Emily Dinan, Stephen Roller, Kurt Shuster, Angela Fan, Michael Auli, and Jason Weston. 2019d. Wizard of Wikipedia: Knowledge-powered conversational agents. In Proceedings of the International Conference on Learning Representations (ICLR).
    Google ScholarLocate open access versionFindings
  • Yupei Du, Yuanbin Wu, and Man Lan. 2019. Exploring human gender stereotypes with word association test. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 6135– 6145.
    Google ScholarLocate open access versionFindings
  • Linda E Duxbury and Christopher A Higgins. 1991. Gender differences in work-family conflict. Journal of applied psychology, 76(1):60.
    Google ScholarLocate open access versionFindings
  • Penelope Eckert and Sally McConnell-Ginet. 1992. Communities of practice: Where language, gender and power all live. In Locating power: Proceedings of the second Berkeley women and language conference, volume 1, pages 89–99.
    Google ScholarLocate open access versionFindings
  • Penelope Eckert and Sally McConnell-Ginet. 2013. Language and gender. Cambridge University Press.
    Google ScholarFindings
  • Penelope Eckert and John R Rickford. 2001. Style and sociolinguistic variation. Cambridge University Press.
    Google ScholarFindings
  • Ali Emami, Paul Trichelair, Adam Trischler, Kaheer Suleman, Hannes Schulz, and Jackie Chi Kit Cheung. 2019. The knowref coreference corpus: Removing gender and number cues for difficult pronominal anaphora resolution. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3952–3961.
    Google ScholarLocate open access versionFindings
  • Angela Fan, David Grangier, and Michael Auli. 2017. Controllable abstractive summarization. arXiv preprint arXiv:1711.05217.
    Findings
  • Angela Fan, Mike Lewis, and Yann Dauphin. 2018. Hierarchical neural story generation. arXiv preprint arXiv:1805.04833.
    Findings
  • Almudena Fernandez Fontecha and Rosa Marıa Jimenez Catalan. 2003. Semantic derogation in animal metaphor: a contrastive-cognitive analysis of two male/female examples in english and spanish. Journal of pragmatics, 35(5):771–797.
    Google ScholarLocate open access versionFindings
  • Georgia Frantzeskou, Efstathios Stamatatos, Stefanos Gritzalis, and Sokratis Katsikas. 2006. Effective identification of source code authors using bytelevel information. In Proceedings of the 28th international conference on Software engineering, pages 893–896.
    Google ScholarLocate open access versionFindings
  • 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.
    Google ScholarLocate open access versionFindings
  • Aparna Garimella, Carmen Banea, Dirk Hovy, and Rada Mihalcea. 2019. Womens syntactic resilience and mens grammatical luck: Gender-bias in part-ofspeech tagging and dependency parsing. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3493–3498.
    Google ScholarLocate open access versionFindings
  • Danielle Gaucher, Justin Friesen, and Aaron C Kay. 2011. Evidence that gendered wording in job advertisements exists and sustains gender inequality. Journal of personality and social psychology, 101(1):109.
    Google ScholarLocate open access versionFindings
  • Andrew Gaut, Tony Sun, Shirlyn Tang, Yuxin Huang, Jing Qian, Mai ElSherief, Jieyu Zhao, Diba Mirza, Elizabeth Belding, Kai-Wei Chang, et al. 2019. Towards understanding gender bias in relation extraction. arXiv preprint arXiv:1911.03642.
    Findings
  • Hila Gonen and Yoav Goldberg. 2019. Lipstick on a pig: Debiasing methods cover up systematic gender biases in word embeddings but do not remove them. 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 609–614, Minneapolis, Minnesota. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Hila Gonen, Yova Kementchedjhieva, and Yoav Goldberg. 2019. How does grammatical gender affect noun representations in gender-marking languages? arXiv preprint arXiv:1910.14161.
    Findings
  • Eduardo Graells-Garrido, Mounia Lalmas, and Filippo Menczer. 2015. First women, second sex: Gender bias in wikipedia. In Proceedings of the 26th ACM Conference on Hypertext & Social Media, pages 165–174.
    Google ScholarLocate open access versionFindings
  • Bernard Guerin. 1994. Gender bias in the abstractness of verbs and adjectives. The Journal of social psychology, 134(4):421–428.
    Google ScholarLocate open access versionFindings
  • Dana V Hiller and William W Philliber. 1982. Predicting marital and career success among dual-worker couples. Journal of Marriage and the Family, pages 53–62.
    Google ScholarLocate open access versionFindings
  • Janet Holmes. 2013. An introduction to sociolinguistics. Routledge.
    Google ScholarFindings
  • Matthew Honnibal and Ines Montani. 2017. spaCy 2: Natural language understanding with Bloom embeddings, convolutional neural networks and incremental parsing. To appear.
    Google ScholarFindings
  • Dirk Hovy, Federico Bianchi, and Tommaso Fornaciari. 2020. Can you translate that into man? commercial machine translation systems include stylistic biases. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Dirk Hovy and Shannon L Spruit. 2016. The social impact of natural language processing. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 591–598.
    Google ScholarLocate open access versionFindings
  • Eduard Hovy. 1987. Generating natural language under pragmatic constraints. Journal of Pragmatics, 11(6):689–719.
    Google ScholarLocate open access versionFindings
  • Alexander Miserlis Hoyle, Lawrence Wolf-Sonkin, Hanna Wallach, Isabelle Augenstein, and Ryan Cotterell. 2019. Unsupervised discovery of gendered language through latent-variable modeling. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1706– 1716, Florence, Italy. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Samuel Humeau, Kurt Shuster, Marie-Anne Lachaux, and Jason Weston. 2019. Poly-encoders: Transformer architectures and pre-training strategies for fast and accurate multi-sentence scoring. arXiv preprint arXiv:1905.01969.
    Findings
  • Dell Hymes. 1974. Ways of speaking. In R. Bauman and J. Sherzer, editors, Explorations in the ethnography of speaking, volume 1, pages 433–451. Cambridge: Cambridge University Press.
    Google ScholarLocate open access versionFindings
  • David Jurgens, Saif Mohammad, Peter Turney, and Keith Holyoak. 2012. SemEval-2012 task 2: Measuring degrees of relational similarity. In *SEM 2012: The First Joint Conference on Lexical and Computational Semantics – Volume 1: Proceedings of the main conference and the shared task, and Volume 2: Proceedings of the Sixth International Workshop on Semantic Evaluation (SemEval 2012), pages 356–364, Montreal, Canada. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Masahiro Kaneko and Danushka Bollegala. 2019. Gender-preserving debiasing for pre-trained word embeddings. arXiv preprint arXiv:1906.00742.
    Findings
  • Dongyeop Kang, Varun Gangal, and Eduard Hovy. 2019. (male, bachelor) and (female, Ph.D) have different connotations: Parallelly annotated stylistic language dataset with multiple personas. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 1696– 1706, Hong Kong, China. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Maximilian Klein, Harsh Gupta, Vivek Rai, Piotr Konieczny, and Haiyi Zhu. 2016. Monitoring the gender gap with wikidata human gender indicators. In Proceedings of the 12th International Symposium on Open Collaboration, pages 1–9.
    Google ScholarLocate open access versionFindings
  • Maximilian Klein and Piotr Konieczny. 2015. Wikipedia in the world of global gender inequality indices: What the biography gender gap is measuring. In Proceedings of the 11th International Symposium on Open Collaboration, pages 1–2.
    Google ScholarLocate open access versionFindings
  • Corina Koolen and Andreas van Cranenburgh. 2017. These are not the stereotypes you are looking for: Bias and fairness in authorial gender attribution. In Proceedings of the First ACL Workshop on Ethics in Natural Language Processing, pages 12–22, Valencia, Spain. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Moshe Koppel, Shlomo Argamon, and Anat Rachel Shimoni. 2002. Automatically categorizing written texts by author gender. Literary and linguistic computing, 17(4):401–412.
    Google ScholarLocate open access versionFindings
  • George Lakoff and Mark Johnson. 1980. Metaphors we live by. Chicago, IL: University of Chicago.
    Google ScholarFindings
  • Robin Lakoff. 1973. Language and woman’s place. Language in society, 2(1):45–79.
    Google ScholarLocate open access versionFindings
  • Robin Lakoff. 1990. Talking Power: The Politics of Language.
    Google ScholarFindings
  • Nayeon Lee, Andrea Madotto, and Pascale Fung. 2019. Exploring social bias in chatbots using stereotype knowledge. In Proceedings of the 2019 Workshop on Widening NLP, pages 177–180.
    Google ScholarLocate open access versionFindings
  • Haley Lepp. 2019. Pardon the interruption: Automatic analysis of gender and competitive turn-taking in united states supreme court hearings. In Proceedings of the 2019 Workshop on Widening NLP, pages 143–145, Florence, Italy. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Pierre Lison and Jorg Tiedemann. 2016. Opensubtitles2016: Extracting large parallel corpora from movie and TV subtitles.
    Google ScholarFindings
  • Haochen Liu, Jamell Dacon, Wenqi Fan, Hui Liu, Zitao Liu, and Jiliang Tang. 2019. Does gender matter? Towards fairness in dialogue systems. CoRR, abs/1910.10486.
    Findings
  • Kim Luyckx and Walter Daelemans. 2008. Authorship attribution and verification with many authors and limited data. In Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 513–520.
    Google ScholarLocate open access versionFindings
  • Rowan Hall Maudslay, Hila Gonen, Ryan Cotterell, and Simone Teufel. 2019. It’s all in the name: Mitigating gender bias with name-based counterfactual data substitution. CoRR, abs/1909.00871.
    Findings
  • 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 Short Papers), pages 622–628, Minneapolis, Minnesota. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Thomas Corwin Mendenhall. 1887. The characteristic curves of composition. Science, 9(214):237–249.
    Google ScholarLocate open access versionFindings
  • Alexander H Miller, Will Feng, Adam Fisch, Jiasen Lu, Dhruv Batra, Antoine Bordes, Devi Parikh, and Jason Weston. 2017. ParlAI: A dialog research software platform. arXiv preprint arXiv:1705.06476.
    Findings
  • Sara Mills. 2014. Language and gender: Interdisciplinary perspectives. Routledge.
    Google ScholarFindings
  • John Money and Anke A Ehrhardt. 1972. Man and woman, boy and girl: Differentiation and dimorphism of gender identity from conception to maturity.
    Google ScholarFindings
  • Rosamund Moon. 2014. From gorgeous to grumpy: adjectives, age and gender. Gender & Language, 8(1).
    Google ScholarLocate open access versionFindings
  • Frederick Mosteller and David L Wallace. 1984. Applied Bayesian and classical inference: the case of the Federalist papers. Springer Verlag.
    Google ScholarFindings
  • Eliza K Pavalko and Glen H Elder Jr. 1993. Women behind the men: Variations in wives’ support of husbands’ careers. Gender & Society, 7(4):548–567.
    Google ScholarLocate open access versionFindings
  • Jian Peng, Kim-Kwang Raymond Choo, and Helen Ashman. 2016. User profiling in intrusion detection: A review. Journal of Network and Computer Applications, 72:14–27.
    Google ScholarLocate open access versionFindings
  • Yusu Qian. 2019. Gender stereotypes differ between male and female writings. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop, pages 48–53.
    Google ScholarLocate open access versionFindings
  • Yusu Qian, Urwa Muaz, Ben Zhang, and Jae Won Hyun. 2019. Reducing gender bias in word-level language models with a gender-equalizing loss function. arXiv preprint arXiv:1905.12801.
    Findings
  • Sindhu Raghavan, Adriana Kovashka, and Raymond Mooney. 2010. Authorship attribution using probabilistic context-free grammars. In Proceedings of the ACL 2010 Conference Short Papers, pages 38–42, Uppsala, Sweden. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Shauli Ravfogel, Yanai Elazar, Hila Gonen, Michael Twiton, and Yoav Goldberg. 2020. Null it out: Guarding protected attributes by iterative nullspace projection. arXiv.
    Google ScholarFindings
  • Joseph Reagle and Lauren Rhue. 2011. Gender bias in wikipedia and britannica. International Journal of Communication, 5:21.
    Google ScholarLocate open access versionFindings
  • Erin M Reid. 2018. Straying from breadwinning: Status and money in men’s interpretations of their wives’ work arrangements. Gender, Work & Organization, 25(6):718–733.
    Google ScholarLocate open access versionFindings
  • John R Rickford and Faye McNair-Knox. 1994. Addressee-and topic-influenced style shift: A quantitative sociolinguistic study. Sociolinguistic perspectives on register, pages 235–276.
    Google ScholarFindings
  • Anderson Rocha, Walter J Scheirer, Christopher W Forstall, Thiago Cavalcante, Antonio Theophilo, Bingyu Shen, Ariadne RB Carvalho, and Efstathios Stamatatos. 2016. Authorship attribution for social media forensics. IEEE Transactions on Information Forensics and Security, 12(1):5–33.
    Google ScholarLocate open access versionFindings
  • Rachel Rudinger, Chandler May, and Benjamin Van Durme. 2017. Social bias in elicited natural language inferences. In Proceedings of the First ACL Workshop on Ethics in Natural Language Processing, pages 74–79, Valencia, Spain. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Maarten Sap, Dallas Card, Saadia Gabriel, Yejin Choi, and Noah A Smith. 2019a. The risk of racial bias in hate speech detection. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1668–1678.
    Google ScholarLocate open access versionFindings
  • Maarten Sap, Saadia Gabriel, Lianhui Qin, Dan Jurafsky, Noah A Smith, and Yejin Choi. 2019b. Social bias frames: Reasoning about social and power implications of language. arXiv preprint arXiv:1911.03891.
    Findings
  • Ruchita Sarawgi, Kailash Gajulapalli, and Yejin Choi. 2011. Gender attribution: Tracing stylometric evidence beyond topic and genre. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning, pages 78–86, Portland, Oregon, USA. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Abigail See, Stephen Roller, Douwe Kiela, and Jason Weston. 2019. What makes a good conversation? how controllable attributes affect human judgments. arXiv preprint arXiv:1902.08654.
    Findings
  • Sima Sharifirad, Alon Jacovi, Israel Bar Ilan Univesity, and Stan Matwin. 2019. Learning and understanding different categories of sexism using convolutional neural networks filters. In Proceedings of the 2019 Workshop on Widening NLP, pages 21–23.
    Google ScholarLocate open access versionFindings
  • Sima Sharifirad and Stan Matwin. 2019. Using attention-based bidirectional lstm to identify different categories of offensive language directed toward female celebrities. In Proceedings of the 2019 Workshop on Widening NLP, pages 46–48.
    Google ScholarLocate open access versionFindings
  • Kurt Shuster, Samuel Humeau, Antoine Bordes, and Jason Weston. 2018. Engaging image chat: Modeling personality in grounded dialogue. arXiv preprint arXiv:1811.00945.
    Findings
  • E. Stamatatos, N. Fakotakis, and G. Kokkinakis. 1999. Automatic authorship attribution. In Ninth Conference of the European Chapter of the Association for Computational Linguistics, Bergen, Norway. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Efstathios Stamatatos. 2009. A survey of modern authorship attribution methods. Journal of the American Society for information Science and Technology, 60(3):538–556.
    Google ScholarLocate open access versionFindings
  • Efstathios Stamatatos. 2017. Authorship attribution using text distortion. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 1138–1149, Valencia, Spain. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Gabriel Stanovsky, Noah A Smith, and Luke Zettlemoyer. 2019. Evaluating gender bias in machine translation. arXiv preprint arXiv:1906.00591.
    Findings
  • Sandeep Subramanian, Guillaume Lample, Eric Michael Smith, Ludovic Denoyer, Marc’Aurelio Ranzato, and Y-Lan Boureau. 2018. Multiple-attribute text style transfer. arXiv preprint arXiv:1811.00552.
    Findings
  • 2019. Mitigating gender bias in natural language processing: Literature review. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1630–1640, Florence, Italy. Association for Computational Linguistics. Jane Sunderland. 2006. Language and gender: An advanced resource book. Routledge.
    Google ScholarLocate open access versionFindings
  • Joan Swann. 1992.
    Google ScholarFindings
  • Mary Talbot. 2019. Language and gender. John Wiley & Sons.
    Google ScholarFindings
  • Rachael Tatman. 2017. Gender and dialect bias in YouTube’s automatic captions. In Proceedings of the First ACL Workshop on Ethics in Natural Language Processing, pages 53–59, Valencia, Spain. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Frances Trix and Carolyn Psenka. 2003. Exploring the color of glass: Letters of recommendation for female and male medical faculty. Discourse & Society, 14(2):191–220.
    Google ScholarLocate open access versionFindings
  • Jack Urbanek, Angela Fan, Siddharth Karamcheti, Saachi Jain, Samuel Humeau, Emily Dinan, Tim Rocktaschel, Douwe Kiela, Arthur Szlam, and Jason Weston. 2019. Learning to speak and act in a fantasy text adventure game. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 673–683, Hong Kong, China. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems, pages 5998–6008.
    Google ScholarLocate open access versionFindings
  • Claudia Wagner, David Garcia, Mohsen Jadidi, and Markus Strohmaier. 2015. It’s a man’s wikipedia? assessing gender inequality in an online encyclopedia. In Ninth international AAAI conference on web and social media.
    Google ScholarLocate open access versionFindings
  • Claudia Wagner, Eduardo Graells-Garrido, David Garcia, and Filippo Menczer. 2016. Women through the glass ceiling: gender asymmetries in wikipedia. EPJ Data Science, 5(1):5.
    Google ScholarLocate open access versionFindings
  • Ann Weatherall. 2002.
    Google ScholarFindings
  • Candace West and Don H Zimmerman. 1987. Doing gender. Gender & society, 1(2):125–151.
    Google ScholarLocate open access versionFindings
  • Eunike Wetzel, Benedikt Hell, and Katja Passler. 2012. Comparison of different test construction strategies in the development of a gender fair interest inventory using verbs. Journal of Career Assessment, 20(1):88–104.
    Google ScholarLocate open access versionFindings
  • Wikipedia contributors. 2020. Bechdel test — Wikipedia, the free encyclopedia. [Online; accessed 3-April-2020].
    Google ScholarFindings
  • Myron Wish, Morton Deutsch, and Susan J Kaplan. 1976. Perceived dimensions of interpersonal relations. Journal of Personality and social Psychology, 33(4):409.
    Google ScholarLocate open access versionFindings
  • Kelvin Xu, Jimmy Ba, Ryan Kiros, Kyunghyun Cho, Aaron Courville, Ruslan Salakhudinov, Rich Zemel, and Yoshua Bengio. 2015.
    Google ScholarFindings
  • Jieyu Zhao, Tianlu Wang, Mark Yatskar, Ryan Cotterell, Vicente Ordonez, and Kai-Wei Chang. 2019. Gender bias in contextualized word embeddings. 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 629–634, Minneapolis, Minnesota. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2018a. Gender bias in coreference resolution: Evaluation and debiasing methods. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers), pages 15–20, New Orleans, Louisiana. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Jieyu Zhao, Yichao Zhou, Zeyu Li, Wei Wang, and KaiWei Chang. 2018b. Learning gender-neutral word embeddings. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4847–4853, Brussels, Belgium. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Pei Zhou, Weijia Shi, Jieyu Zhao, Kuan-Hao Huang, Muhao Chen, Ryan Cotterell, and Kai-Wei Chang. 2019. Examining gender bias in languages with grammatical gender. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5275–5283, Hong Kong, China. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Ran Zmigrod, Sebastian J. Mielke, Hanna Wallach, and Ryan Cotterell. 2019. Counterfactual data augmentation for mitigating gender stereotypes in languages with rich morphology. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1651–1661, Florence, Italy. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • We describe in more detail how each of the eight training datasets is annotated: 1. Wikipedia - to annotate ABOUT, we use a Wikipedia dump and extract biography pages. We identify biographies using named entity recognition applied to the title of the page (Honnibal and Montani, 2017). We label pages with a gender based on the number of gendered pronouns (he vs. she vs. they) and label each paragraph in the page with this label for the ABOUT dimension.7 Wikipedia is well known to have gender bias in equity of biographical coverage and lexical bias in noun references to women (Reagle and Rhue, 2011; Graells-Garrido et al., 2015; Wagner et al., 2015; Klein and Konieczny, 2015; Klein et al., 2016; Wagner et al., 2016), making it an interesting test bed for our investigation.
    Google ScholarLocate open access versionFindings
  • 2. Funpedia - Funpedia (Miller et al., 2017) contains rephrased Wikipedia sentences in a more conversational way. We retain only biography related sentences and annotate similar to Wikipedia, to give ABOUT labels.
    Google ScholarFindings
  • 3. Wizard of Wikipedia - Wizard of Wikipedia (Dinan et al., 2019d) contains two people discussing a topic in Wikipedia. We retain only the conversations on Wikipedia biographies and annotate to create ABOUT labels.
    Google ScholarLocate open access versionFindings
  • 4. ImageChat - ImageChat (Shuster et al., 2018) contains conversations discussing the content of an image. We use the (Xu et al., 2015) image captioning system8 to identify the contents of an image and select gendered examples.
    Google ScholarFindings
  • 5. Yelp - we use the Yelp reviewer gender predictor developed by (Subramanian et al., 2018) and retain reviews for which the classifier is very confident – this creates labels for the author of the review (AS). We impute ABOUT labels on this dataset using a classifier trained on the datasets 1-4.
    Google ScholarFindings
  • 6. ConvAI2 - ConvAI2 (Dinan et al., 2019c) contains persona-based conversations. Many This method of imputing gender is similar to the one used in Reagle and Rhue (2011, 1142) and Bamman and Smith (2014), except we also incorporate non-oppositional gender categories, and rely on basic counts without scaling.
    Google ScholarLocate open access versionFindings
  • 7. OpenSubtitiles - OpenSubtitles9 (Lison and Tiedemann, 2016) contains subtitles for movies in different languages. We retain English subtitles that contain a character name or identity. We annotate the character’s gender using gender kinship terms such as daughter and gender probability distribution calculated by counting the masculine and feminine names of baby names in the United States10. Using the character’s gender, we get labels for the AS dimension. We get labels for the TO dimension by taking the gender of the next character to speak if there is another utterance in the conversation; otherwise, we take the gender of the last character to speak. We impute ABOUT labels on this dataset using a classifier trained on the datasets 1-4.
    Google ScholarFindings
  • 8. LIGHT - LIGHT contains persona-based conversation. Similarly to ConvAI2, annotators labeled the gender of each persona (Dinan et al., 2019a), giving us labels for the speaker (AS) and speaking partner (TO). We impute ABOUT labels on this dataset using a classifier trained on the datasets 1-4. 9http://www.opensubtitles.org/ 10https://catalog.data.gov/dataset/baby-names-fromsocial-security-card-applications-national-level-data
    Findings
  • 1. Edie Sedgwick: was an American actress and fashion model... 2. Linda Darnell: was an American film actress...
    Google ScholarFindings
  • 3. Maureen O’Hara: was an Irish actress and singer... 4. Jessica Savitch: was an American television news presenter and correspondent,... 5. Patsy Mink: Mink served in the U.S. House of Representatives... 6. Shirley Chisholm: was an American politician, educator, and author... 7. Mamie Van Doren: is an American actress, model, singer, and sex symbol who is... 8. Jacqueline Cochran: was a pioneer in the field of American aviation and one of t... 9. Chlo Sevigny: is an American actress, fashion designer, director, and form... 10. Hilda Solis: is an American politician and a member of the Los Angeles Co...
    Google ScholarFindings
  • 1. Derek Jacobi: is an English actor and stage director...
    Google ScholarFindings
  • 2. Bohuslav Martin: was a Czech composer of modern classical music... 3. Carlo Maria Giulini: was an Italian conductor... 4. Zubin Mehta: is an Indian conductor of Western classical music... 5. John Barbirolli: was a British conductor and cellist... 6. Claudio Abbado: was an Italian conductor...
    Google ScholarFindings
  • 7. Ed Harris: is an American actor, producer, director, and screenwriter... 8. Richard Briers: was an English actor...
    Google ScholarLocate open access versionFindings
  • 9. Artur Schnabel: was an Austrian classical pianist, who also composed and tau... 10. Charles Mackerras: was an Australian conductor...
    Google ScholarFindings
Full Text
Your rating :
0

 

Tags
Comments