{ "paper_id": "2022", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T01:11:44.995149Z" }, "title": "Pipelines for Social Bias Testing of Large Language Models", "authors": [ { "first": "Debora", "middle": [], "last": "Nozza", "suffix": "", "affiliation": { "laboratory": "", "institution": "Bocconi University", "location": { "addrLine": "Via Sarfatti 25", "settlement": "Milan", "country": "Italy" } }, "email": "debora.nozza@unibocconi.it" }, { "first": "Federico", "middle": [], "last": "Bianchi", "suffix": "", "affiliation": { "laboratory": "", "institution": "Bocconi University", "location": { "addrLine": "Via Sarfatti 25", "settlement": "Milan", "country": "Italy" } }, "email": "f.bianchi@unibocconi.it" }, { "first": "Dirk", "middle": [], "last": "Hovy", "suffix": "", "affiliation": { "laboratory": "", "institution": "Bocconi University", "location": { "addrLine": "Via Sarfatti 25", "settlement": "Milan", "country": "Italy" } }, "email": "dirk.hovy@unibocconi.it" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "The maturity level of language models is now at a stage in which many companies rely on them to solve various tasks. However, while research has shown how biased and harmful these models are, systematic ways of integrating social bias tests into development pipelines are still lacking. This short paper suggests how to use these verification techniques in development pipelines. We take inspiration from software testing and suggest addressing social bias evaluation as software testing. We hope to open a discussion on the best methodologies to handle social bias testing in language models.", "pdf_parse": { "paper_id": "2022", "_pdf_hash": "", "abstract": [ { "text": "The maturity level of language models is now at a stage in which many companies rely on them to solve various tasks. However, while research has shown how biased and harmful these models are, systematic ways of integrating social bias tests into development pipelines are still lacking. This short paper suggests how to use these verification techniques in development pipelines. We take inspiration from software testing and suggest addressing social bias evaluation as software testing. We hope to open a discussion on the best methodologies to handle social bias testing in language models.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "Current language models are now primarily deployed on large infrastructures (e.g., HuggingFace repository 1 ) and used by many practitioners and researchers with few lines of code. This releasing mechanism has brought tremendous value to the community as researchers everywhere can access models, download them on their laptops, and run experiments. However, these models are quickly adopted without complete understanding their possible limitations .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Recent literature is now rich of papers that demonstrate how social bias is embedded in large language models and propose many different verification and validation datasets (e.g., May et al., 2019; Nozza et al., 2021; Nadeem et al., 2021, inter alia) . Researchers and practitioners can use all these contributions to understand if a model is safe to use or not. We will refer to these works and the datasets used as verification as social bias tests from this point on.", "cite_spans": [ { "start": 181, "end": 198, "text": "May et al., 2019;", "ref_id": "BIBREF26" }, { "start": 199, "end": 218, "text": "Nozza et al., 2021;", "ref_id": "BIBREF29" }, { "start": 219, "end": 251, "text": "Nadeem et al., 2021, inter alia)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "This literature often misses the long-term goal. What is the point of having so many social bias tests that effectively capture different aspects of the problem if we do not find a systematic way of using them? Indeed, this work is also inspired 1 https://huggingface.co/ by the recent approaches and methodologies defined to provide more comprehensive evaluations of models (Ribeiro et al., 2020; Chia et al., 2022) .", "cite_spans": [ { "start": 375, "end": 397, "text": "(Ribeiro et al., 2020;", "ref_id": "BIBREF33" }, { "start": 398, "end": 416, "text": "Chia et al., 2022)", "ref_id": "BIBREF8" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Indeed, other computer science fields have developed insights into how to handle testing. Software development has long been wrestling with the need for good evaluation practices for source code. For example, Continuous Integration and Continuous Deployment (CI/CD) is a general methodology in software development. It assumes frequent testing to ensure that the product under development passes specific qualitative tests that guarantee it is working. In this direction, frequent testing of language models can be part of the solution.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "The main contribution of this short paper is first to identify the main recurring themes and the primary methodologies of social bias literature. We then suggest a more practical and developmental direction: all these methods can be used the same way as tests in software testing pipelines. Unstable/unsafe software should not go into production, which is also true for language models.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "We are aware that a single social bias test cannot provide a complete picture of the problems and that we cannot treat a model that passes the tests as entirely safe. Nonetheless, we believe that some frequent tests are better than no tests. As a community, we need to come together and work closely to stress test these models even during the development phase.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Contributions Our contribution is twofold: we first give an overview of the literature on social bias tests and explore the main themes and methods. We then suggest that this literature can be used in practical contexts to frequently evaluate language models to understand better how the tools we use can be harmful. With this work, we hope to start a discussion on the best methodologies to handle social bias testing in language models as we believe this is a fundamental step to sustain the future and correct usage of these technologies.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "An overview of bias in NLP has been presented in several work (Blodgett et al., 2020; Shah et al., 2020; Hovy and Prabhumoye, 2021; Sheng et al., 2021; Stanczak and Augenstein, 2021) . Here, we focus on the approaches proposed for contextual embeddings. We illustrate the main themes that have driven the developed of social bias tests. The categories we are going to describe are not mutually exclusive, however they showcase in a coherent manner what has been done in the literature.", "cite_spans": [ { "start": 62, "end": 85, "text": "(Blodgett et al., 2020;", "ref_id": "BIBREF3" }, { "start": 86, "end": 104, "text": "Shah et al., 2020;", "ref_id": "BIBREF35" }, { "start": 105, "end": 131, "text": "Hovy and Prabhumoye, 2021;", "ref_id": "BIBREF17" }, { "start": 132, "end": 151, "text": "Sheng et al., 2021;", "ref_id": "BIBREF36" }, { "start": 152, "end": 182, "text": "Stanczak and Augenstein, 2021)", "ref_id": "BIBREF38" } ], "ref_spans": [], "eq_spans": [], "section": "Existing Social Bias tests", "sec_num": "2" }, { "text": "Several studies have been conducted to analyse and determine the level of bias in static word embeddings in binary and multi-class scenarios (Bolukbasi et al., 2016; Caliskan et al., 2017; Garg et al., 2018; Swinger et al., 2019; Manzini et al., 2019; Lauscher and Glava\u0161, 2019; Gonen and Goldberg, 2019) . Several works applied these bias evaluations to contextualized models by extracting static word embeddings for them (Basta et al., 2019; Wolfe and Caliskan, 2021) .", "cite_spans": [ { "start": 141, "end": 165, "text": "(Bolukbasi et al., 2016;", "ref_id": "BIBREF6" }, { "start": 166, "end": 188, "text": "Caliskan et al., 2017;", "ref_id": "BIBREF7" }, { "start": 189, "end": 207, "text": "Garg et al., 2018;", "ref_id": "BIBREF12" }, { "start": 208, "end": 229, "text": "Swinger et al., 2019;", "ref_id": "BIBREF39" }, { "start": 230, "end": 251, "text": "Manzini et al., 2019;", "ref_id": "BIBREF25" }, { "start": 252, "end": 278, "text": "Lauscher and Glava\u0161, 2019;", "ref_id": "BIBREF22" }, { "start": 279, "end": 304, "text": "Gonen and Goldberg, 2019)", "ref_id": "BIBREF16" }, { "start": 423, "end": 443, "text": "(Basta et al., 2019;", "ref_id": "BIBREF1" }, { "start": 444, "end": 469, "text": "Wolfe and Caliskan, 2021)", "ref_id": "BIBREF40" } ], "ref_spans": [], "eq_spans": [], "section": "Word List-based", "sec_num": "2.1" }, { "text": "Inspired by gender bias metrics for word embeddings, May et al. (2019) proposed the Sentence Encoder Association Test (SEAT), a template-based test founded on the Word Embedding Association Test (WEAT) (Caliskan et al., 2017) . Afterward, Liang et al. (2020) used SEAT for measuring bias, also considering the religious dimension.", "cite_spans": [ { "start": 53, "end": 70, "text": "May et al. (2019)", "ref_id": "BIBREF26" }, { "start": 202, "end": 225, "text": "(Caliskan et al., 2017)", "ref_id": "BIBREF7" } ], "ref_spans": [], "eq_spans": [], "section": "Word List-based", "sec_num": "2.1" }, { "text": "Template-based approaches exploit the fact that BERT-like models are trained using a masked language modeling objective. I.e., given a sentence with omitted tokens indicated as [MASK] , they predict the masked tokens. The predictions for these [MASK] tokens may provide us with some insight into the bias embedded in the actual representations. We can generate templates in two different ways. First, by accounting for certain targets (e.g., gendered words) and attributes (e.g. career-related words) (Kurita et al., 2019; Dev et al., 2020) . This enable, for example, to compute the association between the target male gender and the attribute programmer, by feeding \"[MASK] is a programmer\" to BERT, and compute the probability assigned to the sentence \"he is a programmer\". Another option is to create templates coupling protected group targets with neutral predicates (e.g., \"works as\", \"is known for\"). For example, we can ask BERT to complete \"the woman is known for [MASK]\" or \"the girl worked as [MASK] .\" Then, it is possible to exploit lexicons (Nozza et al., 2021 (Nozza et al., , 2022 , or hate speech (Ousidhoum et al., 2021; Sheng et al., 2019) and sentiment classifiers Hutchinson et al. (2020) ; Huang et al. (2020) to obtain a social bias score from the template-based generated text. Ideally, using a classifier lets us test the data more easily and accurately than lexicons.", "cite_spans": [ { "start": 177, "end": 183, "text": "[MASK]", "ref_id": null }, { "start": 244, "end": 250, "text": "[MASK]", "ref_id": null }, { "start": 501, "end": 522, "text": "(Kurita et al., 2019;", "ref_id": "BIBREF21" }, { "start": 523, "end": 540, "text": "Dev et al., 2020)", "ref_id": "BIBREF10" }, { "start": 1004, "end": 1010, "text": "[MASK]", "ref_id": null }, { "start": 1055, "end": 1074, "text": "(Nozza et al., 2021", "ref_id": "BIBREF29" }, { "start": 1075, "end": 1096, "text": "(Nozza et al., , 2022", "ref_id": "BIBREF30" }, { "start": 1114, "end": 1138, "text": "(Ousidhoum et al., 2021;", "ref_id": "BIBREF32" }, { "start": 1139, "end": 1158, "text": "Sheng et al., 2019)", "ref_id": "BIBREF37" }, { "start": 1185, "end": 1209, "text": "Hutchinson et al. (2020)", "ref_id": "BIBREF19" }, { "start": 1212, "end": 1231, "text": "Huang et al. (2020)", "ref_id": "BIBREF18" } ], "ref_spans": [], "eq_spans": [], "section": "Template-based", "sec_num": "2.2" }, { "text": "The same approach can be applied to natural language generation models (Sheng et al., 2019; Huang et al., 2020) . The models are not fed with a masked token but are asked to complete the template. So, instead of a single word, they return a set of words.", "cite_spans": [ { "start": 71, "end": 91, "text": "(Sheng et al., 2019;", "ref_id": "BIBREF37" }, { "start": 92, "end": 111, "text": "Huang et al., 2020)", "ref_id": "BIBREF18" } ], "ref_spans": [], "eq_spans": [], "section": "Template-based", "sec_num": "2.2" }, { "text": "An interesting case has been proposed by Choenni et al.. They look into what kinds of stereotyped information are collected by LLMs exploiting a dataset comprising stereotypical attributes for various social groups. The dataset was created by feeding search engines queries that already imply a stereotype about a specific social group (e.g., 'Why are Asian parents so'). Then, the authors count how many of the stereotypes found by the search engines are also encoded in the LLMs through masked language modeling.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Template-based", "sec_num": "2.2" }, { "text": "Few works have collected datasets to compute bias scores. Nadeem et al. (2021) presented StereoSet, a crowdsourced English dataset to measure stereotypical biases in four domains: gender, profession, race, and religion. Nangia et al. (2020) introduced CrowS-Pairs, a crowdsourced benchmark comprising 1508 examples that cover stereotypes dealing with nine types of bias. Both Nadeem et al. (2021) ; Nangia et al. (2020) proposed a metric to measure for how many examples the model prefers stereotyped sentences over less stereotyped sentences.", "cite_spans": [ { "start": 58, "end": 78, "text": "Nadeem et al. (2021)", "ref_id": "BIBREF27" }, { "start": 220, "end": 240, "text": "Nangia et al. (2020)", "ref_id": "BIBREF28" }, { "start": 376, "end": 396, "text": "Nadeem et al. (2021)", "ref_id": "BIBREF27" }, { "start": 399, "end": 419, "text": "Nangia et al. (2020)", "ref_id": "BIBREF28" } ], "ref_spans": [], "eq_spans": [], "section": "Crowdsourced-based", "sec_num": "2.3" }, { "text": "These messages are then split in half (a prompt and a continuation) and are used to study, for example, whether a model generates toxic continuations from a non-toxic prompt.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Crowdsourced-based", "sec_num": "2.3" }, { "text": "While many social bias tests have been provided in the literature, they differ in methodology, covered languages, and protected groups. Most works are on English. Only (Nozza et al., 2021; Ousidhoum et al., 2021) considered languages beyond English. The majority of work focused on gender bias, and only a few investigated an extensive range of targets (Nangia et al., 2020; Nadeem et al., 2021; Ousidhoum et al., 2021; Barikeri et al., 2021) . We also found that Hutchinson et al. (2020) ; Huang et al. (2020) did not provide data or code publicly. Blodgett et al. 2021presented a critical review of some social bias tests and found significant issues with noise, unnaturalness, and reliability of the some work (Nangia et al., 2020; Nadeem et al., 2021) . Finally, it is important to highlight that social biases are different depending on the cultural and historical context of application of the language model. This brief analysis demonstrates that no existing social bias test is universal. While we may fill this research gap in the future, for now, we suggest using more than one test has to be used to measure bias.", "cite_spans": [ { "start": 168, "end": 188, "text": "(Nozza et al., 2021;", "ref_id": "BIBREF29" }, { "start": 189, "end": 212, "text": "Ousidhoum et al., 2021)", "ref_id": "BIBREF32" }, { "start": 353, "end": 374, "text": "(Nangia et al., 2020;", "ref_id": "BIBREF28" }, { "start": 375, "end": 395, "text": "Nadeem et al., 2021;", "ref_id": "BIBREF27" }, { "start": 396, "end": 419, "text": "Ousidhoum et al., 2021;", "ref_id": "BIBREF32" }, { "start": 420, "end": 442, "text": "Barikeri et al., 2021)", "ref_id": "BIBREF0" }, { "start": 464, "end": 488, "text": "Hutchinson et al. (2020)", "ref_id": "BIBREF19" }, { "start": 713, "end": 734, "text": "(Nangia et al., 2020;", "ref_id": "BIBREF28" }, { "start": 735, "end": 755, "text": "Nadeem et al., 2021)", "ref_id": "BIBREF27" } ], "ref_spans": [], "eq_spans": [], "section": "Discussion", "sec_num": "2.5" }, { "text": "We describe the different modalities that can be used to integrate social bias tests into development pipelines.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Integration", "sec_num": "3" }, { "text": "Software testing is at the heart of software development. Without good evaluation, software easily breaks in production, causing economic damage to companies.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Continuous Social Bias Verification", "sec_num": "3.1" }, { "text": "Most of the checks currently run to test language models are structural. For example, does it produce outputs correctly? Once fine-tuned, are the results we get in a sensible range? We suggest that tests should cover social biases.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Continuous Social Bias Verification", "sec_num": "3.1" }, { "text": "We take inspiration from software testing and suggest testing methodologies for language models. In a CI/CD (continuous integration and continuous development) setting, code is continuously pushed into the repository and tested to ensure the model is stable. Software is deployed if and only if tests are correctly passed. We believe that we should replicate this pipeline in the development of language models. Every time a new model is released, we can run tests to verify if and how the model is hurtful.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Continuous Social Bias Verification", "sec_num": "3.1" }, { "text": "Note that this is indeed a real problem. Many pipelines are now based on HuggingFace APIs that directly download the model from the HuggingFace Hub. Users might not know what happens on the backend: what happens when a model is updated, and the user downloads it thinking it is the same as the older version? We are not sure how many users keep track of commits and changelogs, and this might create a misunderstanding about which model is being used and with which training setup.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Continuous Social Bias Verification", "sec_num": "3.1" }, { "text": "Publishers may help maintain the fairness of the research ecosystem by establishing a badging mechanism. This approach would increase the likelihood that an LLM will be tested in advance for social biases and that end-users will pay attention to this issue.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Badge System", "sec_num": "3.2" }, { "text": "Here, we propose a badging system based on the ACM one 2 and the one proposed for the NAACL 2022 reproducibility track 3 . We identified three possible badges: Social Bias Evaluated, Social Bias Available, and Results Validated.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Badge System", "sec_num": "3.2" }, { "text": "This badge is given to LLMs who have successfully run the social bias tests. This badge does not require the scores to be made publicly available.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Social Bias Evaluated", "sec_num": null }, { "text": "LLMs that made the results of social bias tests retrievable. We propose to design one badge for each implemented social bias test and to show it along with the associated score. We discourage using badges as binary (i.e., test passed or test failed) for these particular cases. Considering the problem as binary might imply that a passing model is entirely free of bias, even if this is not the case.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Social Bias Available This badge is given to", "sec_num": null }, { "text": "Results Validated This badge is given to LLMs in which the social bias test results were successfully attained by a person or team other than the author. A model can be developed and trained on a server and pushed online. Then we can use an automation tool (e.g., Github Actions) to start an evaluation engine that will eventually generate the predictions for the models. Once scored, the model can be released online with badges identifying possible issues that one might encounter with the model.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Social Bias Available This badge is given to", "sec_num": null }, { "text": "Badging is also a standard and straightforward system to showcase software validity in an online repository. These badges are often used to show information about the number of downloads, the test coverage, the quality of the documentation and allow users to understand the quality of what they are using with a quick look. Figure 1 shows a possible integration of testing for harms in development pipelines. We can develop the models on a local server and push this model online after training is finished (with Git LSF, for example). Pushing should automatically start an evaluation pipeline (something close to Github Actions) that starts an evaluation engine: this engine should load the models and run the social bias tests. Once the results are collected, and the metrics have been scored, the model can finally appear on online repositories with badges that identify if and how the test have been run with the respective scores.", "cite_spans": [], "ref_spans": [ { "start": 324, "end": 332, "text": "Figure 1", "ref_id": "FIGREF0" } ], "eq_spans": [], "section": "Social Bias Available This badge is given to", "sec_num": null }, { "text": "An open question is if the test should be available to the developer of the models. On the one hand, releasing the tests makes it easier for everyone to evaluate their models internally before release. On the other hand, this makes it easier to \"train on test\" and hack the system to obtain better scores.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Limits of this Integration", "sec_num": "3.3" }, { "text": "Hiding the test sets from the developer is closer to standard Quality And Assurance developers in companies that are meant to test the interfaces and the code that the developer has built. This approach is also in line with challenges that do not share test data and in which models are submitted using docker containers that are then internally evaluated and scored. As Goodhart's law states, \"When a measure becomes a target, it ceases to be a good measure\". Thus we should be aware that social bias tests cannot be the panacea for language models problems. We cannot rely only on a test to assess the validity of a model. 4 Another point in discussion is that the pipelines we have designed are meant to evaluate intrinsic bias in language models. Unfortunately, this does not consider the verification of bias in downstream application: this extrinsic bias has been found to be poorly correlated with the original bias of language models (Goldfarb-Tarrant et al., 2021) . However, we want to point out the an additional set of application-specific tests could be used to evaluate the models adapted for these tasks: for example, researchers could use hate speech check tests (Dixon et al., 2018; Nozza et al., 2019; R\u00f6ttger et al., 2021) to verify social biases in hate speech detection models.", "cite_spans": [ { "start": 625, "end": 626, "text": "4", "ref_id": null }, { "start": 942, "end": 973, "text": "(Goldfarb-Tarrant et al., 2021)", "ref_id": "BIBREF15" }, { "start": 1179, "end": 1199, "text": "(Dixon et al., 2018;", "ref_id": "BIBREF11" }, { "start": 1200, "end": 1219, "text": "Nozza et al., 2019;", "ref_id": "BIBREF31" }, { "start": 1220, "end": 1241, "text": "R\u00f6ttger et al., 2021)", "ref_id": "BIBREF34" } ], "ref_spans": [], "eq_spans": [], "section": "Limits of this Integration", "sec_num": "3.3" }, { "text": "This paper proposes to use social bias tests in model development pipelines. We believe that our work can be helpful to make the development of these models fairer and easier to sustain from an ethical point of view. Future work is needed to answer several questions about this system. For example, who creates the tests and how can we make sure that these tests can be trusted? It becomes critical to involve marginalized communities to develop more sustainable and effective social bias tests.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "4" }, { "text": ".4 Social Media-basedBarikeri et al. (2021) propose a bias evaluation framework for conversational LLMs using REDDIT-BIAS, an English conversational data set grounded in real-world human conversations from Reddit. The authors propose a perplexity-based bias measure meant to quantify the amount of bias in generative language models along several bias dimensions. Gehman et al. (2020) focus on collecting prompts from the OpenWebText Corpus(Gokaslan and Cohen, 2019) and annotating them with the Perspective API to evaluate the toxicity of the messages.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "https://www.acm.org/ publications/policies/ artifact-review-and-badging-current 3 https://2022.naacl.org/blog/ reproducibility-track/", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Albeit, this comment is true for any measure we use in the field.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "This project has partially received funding from the European Research Council ( ", "cite_spans": [], "ref_spans": [ { "start": 79, "end": 80, "text": "(", "ref_id": null } ], "eq_spans": [], "section": "Acknowledgements", "sec_num": null }, { "text": "We understand that providing social bias tests as a quantifiable indicator for bias carries a significant risk. A low score on a social bias test might be used to assert that a model is fully devoid of bias. As Nangia et al. (2020), we strongly advise against this. Tests can be an indication of issues. Conversely,the absence of a high score does not necessarily entail the absence of bias. Neither do replace a thorough investigation of the data.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Ethical Statements", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "RedditBias: A real-world resource for bias evaluation and debiasing of conversational language models", "authors": [ { "first": "Soumya", "middle": [], "last": "Barikeri", "suffix": "" }, { "first": "Anne", "middle": [], "last": "Lauscher", "suffix": "" }, { "first": "Ivan", "middle": [], "last": "Vuli\u0107", "suffix": "" }, { "first": "Goran", "middle": [], "last": "Glava\u0161", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing", "volume": "1", "issue": "", "pages": "1941--1955", "other_ids": { "DOI": [ "10.18653/v1/2021.acl-long.151" ] }, "num": null, "urls": [], "raw_text": "Soumya Barikeri, Anne Lauscher, Ivan Vuli\u0107, and Goran Glava\u0161. 2021. RedditBias: A real-world resource for bias evaluation and debiasing of conversational lan- guage models. In Proceedings of the 59th Annual Meeting of the Association for Computational Lin- guistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 1941-1955, Online. Association for Computational Linguistics.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Evaluating the underlying gender bias in contextualized word embeddings", "authors": [ { "first": "Christine", "middle": [], "last": "Basta", "suffix": "" }, { "first": "Marta", "middle": [ "R" ], "last": "Costa-Juss\u00e0", "suffix": "" }, { "first": "Noe", "middle": [], "last": "Casas", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the First Workshop on Gender Bias in Natural Language Processing", "volume": "", "issue": "", "pages": "33--39", "other_ids": { "DOI": [ "10.18653/v1/W19-3805" ] }, "num": null, "urls": [], "raw_text": "Christine Basta, Marta R. Costa-juss\u00e0, and Noe Casas. 2019. Evaluating the underlying gender bias in con- textualized word embeddings. In Proceedings of the First Workshop on Gender Bias in Natural Language Processing, pages 33-39, Florence, Italy. Association for Computational Linguistics.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "On the gap between adoption and understanding in NLP", "authors": [ { "first": "Federico", "middle": [], "last": "Bianchi", "suffix": "" }, { "first": "Dirk", "middle": [], "last": "Hovy", "suffix": "" } ], "year": 2021, "venue": "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021", "volume": "", "issue": "", "pages": "3895--3901", "other_ids": { "DOI": [ "10.18653/v1/2021.findings-acl.340" ] }, "num": null, "urls": [], "raw_text": "Federico Bianchi and Dirk Hovy. 2021. On the gap be- tween adoption and understanding in NLP. In Find- ings of the Association for Computational Linguis- tics: ACL-IJCNLP 2021, pages 3895-3901, Online. Association for Computational Linguistics.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Language (technology) is power: A critical survey of \"bias\" in NLP", "authors": [ { "first": "", "middle": [], "last": "Su Lin", "suffix": "" }, { "first": "Solon", "middle": [], "last": "Blodgett", "suffix": "" }, { "first": "Hal", "middle": [], "last": "Barocas", "suffix": "" }, { "first": "Iii", "middle": [], "last": "Daum\u00e9", "suffix": "" }, { "first": "Hanna", "middle": [], "last": "Wallach", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "5454--5476", "other_ids": { "DOI": [ "10.18653/v1/2020.acl-main.485" ] }, "num": null, "urls": [], "raw_text": "Su Lin Blodgett, Solon Barocas, Hal Daum\u00e9 III, and Hanna Wallach. 2020. Language (technology) is power: A critical survey of \"bias\" in NLP. In Pro- ceedings of the 58th Annual Meeting of the Asso- ciation for Computational Linguistics, pages 5454- 5476, Online. Association for Computational Lin- guistics.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Stereotyping Norwegian salmon: An inventory of pitfalls in fairness benchmark datasets", "authors": [ { "first": "", "middle": [], "last": "Su Lin", "suffix": "" }, { "first": "Gilsinia", "middle": [], "last": "Blodgett", "suffix": "" }, { "first": "Alexandra", "middle": [], "last": "Lopez", "suffix": "" }, { "first": "Robert", "middle": [], "last": "Olteanu", "suffix": "" }, { "first": "Hanna", "middle": [], "last": "Sim", "suffix": "" }, { "first": "", "middle": [], "last": "Wallach", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 59th", "volume": "", "issue": "", "pages": "", "other_ids": { "DOI": [ "10.18653/v1/2021.acl-long.81" ] }, "num": null, "urls": [], "raw_text": "Su Lin Blodgett, Gilsinia Lopez, Alexandra Olteanu, Robert Sim, and Hanna Wallach. 2021. Stereotyping Norwegian salmon: An inventory of pitfalls in fair- ness benchmark datasets. In Proceedings of the 59th", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing", "authors": [], "year": null, "venue": "", "volume": "1", "issue": "", "pages": "1004--1015", "other_ids": {}, "num": null, "urls": [], "raw_text": "Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Confer- ence on Natural Language Processing (Volume 1: Long Papers), pages 1004-1015, Online. Association for Computational Linguistics.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Man is to computer programmer as woman is to homemaker? debiasing word embeddings", "authors": [ { "first": "Tolga", "middle": [], "last": "Bolukbasi", "suffix": "" }, { "first": "Kai-Wei", "middle": [], "last": "Chang", "suffix": "" }, { "first": "James", "middle": [ "Y" ], "last": "Zou", "suffix": "" }, { "first": "Venkatesh", "middle": [], "last": "Saligrama", "suffix": "" }, { "first": "Adam Tauman", "middle": [], "last": "Kalai", "suffix": "" } ], "year": 2016, "venue": "Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems", "volume": "", "issue": "", "pages": "4349--4357", "other_ids": {}, "num": null, "urls": [], "raw_text": "Tolga Bolukbasi, Kai-Wei Chang, James Y. Zou, Venkatesh Saligrama, and Adam Tauman Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Ad- vances in Neural Information Processing Systems 29: Annual Conference on Neural Information Process- ing Systems 2016, December 5-10, 2016, Barcelona, Spain, pages 4349-4357.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Semantics derived automatically from language corpora contain human-like biases", "authors": [ { "first": "Aylin", "middle": [], "last": "Caliskan", "suffix": "" }, { "first": "Joanna", "middle": [ "J" ], "last": "Bryson", "suffix": "" }, { "first": "Arvind", "middle": [], "last": "Narayanan", "suffix": "" } ], "year": 2017, "venue": "Science", "volume": "356", "issue": "6334", "pages": "183--186", "other_ids": { "DOI": [ "10.1126/science.aal4230" ] }, "num": null, "urls": [], "raw_text": "Aylin Caliskan, Joanna J. Bryson, and Arvind Narayanan. 2017. Semantics derived automatically from language corpora contain human-like biases. Science, 356(6334):183-186.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Beyond ndcg: behavioral testing of recommender systems with reclist", "authors": [ { "first": "Patrick John", "middle": [], "last": "Chia", "suffix": "" }, { "first": "Jacopo", "middle": [], "last": "Tagliabue", "suffix": "" }, { "first": "Federico", "middle": [], "last": "Bianchi", "suffix": "" }, { "first": "Chloe", "middle": [], "last": "He", "suffix": "" }, { "first": "Brian", "middle": [], "last": "Ko", "suffix": "" } ], "year": 2022, "venue": "Companion Proceedings of the Web Conference", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Patrick John Chia, Jacopo Tagliabue, Federico Bianchi, Chloe He, and Brian Ko. 2022. Beyond ndcg: behav- ioral testing of recommender systems with reclist. In Companion Proceedings of the Web Conference.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Stepmothers are mean and academics are pretentious: What do pretrained language models learn about you?", "authors": [ { "first": "Rochelle", "middle": [], "last": "Choenni", "suffix": "" }, { "first": "Ekaterina", "middle": [], "last": "Shutova", "suffix": "" }, { "first": "Robert", "middle": [], "last": "Van Rooij", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "1477--1491", "other_ids": { "DOI": [ "10.18653/v1/2021.emnlp-main.111" ] }, "num": null, "urls": [], "raw_text": "Rochelle Choenni, Ekaterina Shutova, and Robert van Rooij. 2021. Stepmothers are mean and academics are pretentious: What do pretrained language models learn about you? In Proceedings of the 2021 Confer- ence on Empirical Methods in Natural Language Pro- cessing, pages 1477-1491, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "On measuring and mitigating biased inferences of word embeddings", "authors": [ { "first": "Sunipa", "middle": [], "last": "Dev", "suffix": "" }, { "first": "Tao", "middle": [], "last": "Li", "suffix": "" }, { "first": "Jeff", "middle": [ "M" ], "last": "Phillips", "suffix": "" }, { "first": "Vivek", "middle": [], "last": "", "suffix": "" } ], "year": 2020, "venue": "The Thirty-Second Innovative Applications of Artificial Intelligence Conference", "volume": "2020", "issue": "", "pages": "7659--7666", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sunipa Dev, Tao Li, Jeff M. Phillips, and Vivek Sriku- mar. 2020. On measuring and mitigating biased in- ferences of word embeddings. In The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, IAAI 2020, pages 7659-7666. AAAI Press.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Measuring and mitigating unintended bias in text classification", "authors": [ { "first": "Lucas", "middle": [], "last": "Dixon", "suffix": "" }, { "first": "John", "middle": [], "last": "Li", "suffix": "" }, { "first": "Jeffrey", "middle": [], "last": "Sorensen", "suffix": "" }, { "first": "Nithum", "middle": [], "last": "Thain", "suffix": "" }, { "first": "Lucy", "middle": [], "last": "Vasserman", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, AIES '18", "volume": "", "issue": "", "pages": "67--73", "other_ids": { "DOI": [ "10.1145/3278721.3278729" ] }, "num": null, "urls": [], "raw_text": "Lucas Dixon, John Li, Jeffrey Sorensen, Nithum Thain, and Lucy Vasserman. 2018. Measuring and mitigat- ing unintended bias in text classification. In Proceed- ings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, AIES '18, page 67-73, New York, NY, USA. Association for Computing Machinery.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Word embeddings quantify 100 years of gender and ethnic stereotypes", "authors": [ { "first": "Nikhil", "middle": [], "last": "Garg", "suffix": "" }, { "first": "Londa", "middle": [], "last": "Schiebinger", "suffix": "" }, { "first": "Dan", "middle": [], "last": "Jurafsky", "suffix": "" }, { "first": "James", "middle": [], "last": "Zou", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the National Academy of Sciences", "volume": "115", "issue": "16", "pages": "3635--3644", "other_ids": { "DOI": [ "10.1073/pnas.1720347115" ] }, "num": null, "urls": [], "raw_text": "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.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "RealToxi-cityPrompts: Evaluating neural toxic degeneration in language models", "authors": [ { "first": "Suchin", "middle": [], "last": "Samuel Gehman", "suffix": "" }, { "first": "Maarten", "middle": [], "last": "Gururangan", "suffix": "" }, { "first": "Yejin", "middle": [], "last": "Sap", "suffix": "" }, { "first": "Noah", "middle": [ "A" ], "last": "Choi", "suffix": "" }, { "first": "", "middle": [], "last": "Smith", "suffix": "" } ], "year": 2020, "venue": "Findings of the Association for Computational Linguistics: EMNLP 2020", "volume": "", "issue": "", "pages": "3356--3369", "other_ids": { "DOI": [ "10.18653/v1/2020.findings-emnlp.301" ] }, "num": null, "urls": [], "raw_text": "Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, and Noah A. Smith. 2020. RealToxi- cityPrompts: Evaluating neural toxic degeneration in language models. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3356-3369, Online. Association for Computational Linguistics.", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "Openwebtext corpus", "authors": [ { "first": "Aaron", "middle": [], "last": "Gokaslan", "suffix": "" }, { "first": "Vanya", "middle": [], "last": "Cohen", "suffix": "" } ], "year": 2019, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Aaron Gokaslan and Vanya Cohen. 2019. Openwebtext corpus. http://Skylion007.github.io/ OpenWebTextCorpus.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "Intrinsic bias metrics do not correlate with application bias", "authors": [ { "first": "Seraphina", "middle": [], "last": "Goldfarb-Tarrant", "suffix": "" }, { "first": "Rebecca", "middle": [], "last": "Marchant", "suffix": "" }, { "first": "Ricardo", "middle": [ "Mu\u00f1oz" ], "last": "S\u00e1nchez", "suffix": "" }, { "first": "Mugdha", "middle": [], "last": "Pandya", "suffix": "" }, { "first": "Adam", "middle": [], "last": "Lopez", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing", "volume": "1", "issue": "", "pages": "1926--1940", "other_ids": { "DOI": [ "10.18653/v1/2021.acl-long.150" ] }, "num": null, "urls": [], "raw_text": "Seraphina Goldfarb-Tarrant, Rebecca Marchant, Ri- cardo Mu\u00f1oz S\u00e1nchez, Mugdha Pandya, and Adam Lopez. 2021. Intrinsic bias metrics do not correlate with application bias. In Proceedings of the 59th An- nual Meeting of the Association for Computational Linguistics and the 11th International Joint Confer- ence on Natural Language Processing (Volume 1: Long Papers), pages 1926-1940, Online. Association for Computational Linguistics.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "Lipstick on a pig: Debiasing methods cover up systematic gender biases in word embeddings but do not remove them", "authors": [ { "first": "Hila", "middle": [], "last": "Gonen", "suffix": "" }, { "first": "Yoav", "middle": [], "last": "Goldberg", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", "volume": "1", "issue": "", "pages": "609--614", "other_ids": { "DOI": [ "10.18653/v1/N19-1061" ] }, "num": null, "urls": [], "raw_text": "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 Computa- tional Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 609-614, Minneapolis, Minnesota. Association for Computa- tional Linguistics.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "Five sources of bias in natural language processing. Language and Linguistics Compass", "authors": [ { "first": "Dirk", "middle": [], "last": "Hovy", "suffix": "" }, { "first": "Shrimai", "middle": [], "last": "Prabhumoye", "suffix": "" } ], "year": 2021, "venue": "", "volume": "15", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Dirk Hovy and Shrimai Prabhumoye. 2021. Five sources of bias in natural language processing. Lan- guage and Linguistics Compass, 15(8):e12432.", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "Reducing sentiment bias in language models via counterfactual evaluation", "authors": [ { "first": "Po-Sen", "middle": [], "last": "Huang", "suffix": "" }, { "first": "Huan", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Ray", "middle": [], "last": "Jiang", "suffix": "" }, { "first": "Robert", "middle": [], "last": "Stanforth", "suffix": "" }, { "first": "Johannes", "middle": [], "last": "Welbl", "suffix": "" }, { "first": "Jack", "middle": [], "last": "Rae", "suffix": "" }, { "first": "Vishal", "middle": [], "last": "Maini", "suffix": "" }, { "first": "Dani", "middle": [], "last": "Yogatama", "suffix": "" }, { "first": "Pushmeet", "middle": [], "last": "Kohli", "suffix": "" } ], "year": 2020, "venue": "Findings of the Association for Computational Linguistics: EMNLP 2020", "volume": "", "issue": "", "pages": "65--83", "other_ids": { "DOI": [ "10.18653/v1/2020.findings-emnlp.7" ] }, "num": null, "urls": [], "raw_text": "Po-Sen Huang, Huan Zhang, Ray Jiang, Robert Stan- forth, Johannes Welbl, Jack Rae, Vishal Maini, Dani Yogatama, and Pushmeet Kohli. 2020. Reducing sen- timent bias in language models via counterfactual evaluation. In Findings of the Association for Com- putational Linguistics: EMNLP 2020, pages 65-83, Online. Association for Computational Linguistics.", "links": null }, "BIBREF19": { "ref_id": "b19", "title": "Social biases in NLP models as barriers for persons with disabilities", "authors": [ { "first": "Ben", "middle": [], "last": "Hutchinson", "suffix": "" }, { "first": "Vinodkumar", "middle": [], "last": "Prabhakaran", "suffix": "" }, { "first": "Emily", "middle": [], "last": "Denton", "suffix": "" }, { "first": "Kellie", "middle": [], "last": "Webster", "suffix": "" }, { "first": "Yu", "middle": [], "last": "Zhong", "suffix": "" }, { "first": "Stephen", "middle": [], "last": "Denuyl", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 58th", "volume": "", "issue": "", "pages": "", "other_ids": { "DOI": [ "10.18653/v1/2020.acl-main.487" ] }, "num": null, "urls": [], "raw_text": "Ben Hutchinson, Vinodkumar Prabhakaran, Emily Den- ton, Kellie Webster, Yu Zhong, and Stephen Denuyl. 2020. Social biases in NLP models as barriers for persons with disabilities. In Proceedings of the 58th", "links": null }, "BIBREF20": { "ref_id": "b20", "title": "Annual Meeting of the Association for Computational Linguistics", "authors": [], "year": null, "venue": "", "volume": "", "issue": "", "pages": "5491--5501", "other_ids": {}, "num": null, "urls": [], "raw_text": "Annual Meeting of the Association for Computational Linguistics, pages 5491-5501, Online. Association for Computational Linguistics.", "links": null }, "BIBREF21": { "ref_id": "b21", "title": "Measuring bias in contextualized word representations", "authors": [ { "first": "Keita", "middle": [], "last": "Kurita", "suffix": "" }, { "first": "Nidhi", "middle": [], "last": "Vyas", "suffix": "" }, { "first": "Ayush", "middle": [], "last": "Pareek", "suffix": "" }, { "first": "Alan", "middle": [ "W" ], "last": "Black", "suffix": "" }, { "first": "Yulia", "middle": [], "last": "Tsvetkov", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the First Workshop on Gender Bias in Natural Language Processing", "volume": "", "issue": "", "pages": "166--172", "other_ids": { "DOI": [ "10.18653/v1/W19-3823" ] }, "num": null, "urls": [], "raw_text": "Keita Kurita, Nidhi Vyas, Ayush Pareek, Alan W Black, and Yulia Tsvetkov. 2019. Measuring bias in con- textualized word representations. In Proceedings of the First Workshop on Gender Bias in Natural Lan- guage Processing, pages 166-172, Florence, Italy. Association for Computational Linguistics.", "links": null }, "BIBREF22": { "ref_id": "b22", "title": "Are we consistently biased? multidimensional analysis of biases in distributional word vectors", "authors": [ { "first": "Anne", "middle": [], "last": "Lauscher", "suffix": "" }, { "first": "Goran", "middle": [], "last": "Glava\u0161", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the Eighth Joint Conference on Lexical and Computational Semantics (*SEM 2019)", "volume": "", "issue": "", "pages": "85--91", "other_ids": { "DOI": [ "10.18653/v1/S19-1010" ] }, "num": null, "urls": [], "raw_text": "Anne Lauscher and Goran Glava\u0161. 2019. Are we con- sistently biased? multidimensional analysis of biases in distributional word vectors. In Proceedings of the Eighth Joint Conference on Lexical and Computa- tional Semantics (*SEM 2019), pages 85-91, Min- neapolis, Minnesota. Association for Computational Linguistics.", "links": null }, "BIBREF23": { "ref_id": "b23", "title": "Sustainable modular debiasing of language models", "authors": [ { "first": "Anne", "middle": [], "last": "Lauscher", "suffix": "" }, { "first": "Tobias", "middle": [], "last": "Lueken", "suffix": "" }, { "first": "Goran", "middle": [], "last": "Glava\u0161", "suffix": "" } ], "year": 2021, "venue": "Findings of the Association for Computational Linguistics: EMNLP 2021", "volume": "", "issue": "", "pages": "4782--4797", "other_ids": { "DOI": [ "10.18653/v1/2021.findings-emnlp.411" ] }, "num": null, "urls": [], "raw_text": "Anne Lauscher, Tobias Lueken, and Goran Glava\u0161. 2021. Sustainable modular debiasing of language models. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 4782-4797, Punta Cana, Dominican Republic. Association for Compu- tational Linguistics.", "links": null }, "BIBREF24": { "ref_id": "b24", "title": "Towards debiasing sentence representations", "authors": [ { "first": "Irene", "middle": [ "Mengze" ], "last": "Paul Pu Liang", "suffix": "" }, { "first": "Emily", "middle": [], "last": "Li", "suffix": "" }, { "first": "", "middle": [], "last": "Zheng", "suffix": "" }, { "first": "Chong", "middle": [], "last": "Yao", "suffix": "" }, { "first": "Ruslan", "middle": [], "last": "Lim", "suffix": "" }, { "first": "Louis-Philippe", "middle": [], "last": "Salakhutdinov", "suffix": "" }, { "first": "", "middle": [], "last": "Morency", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "5502--5515", "other_ids": { "DOI": [ "10.18653/v1/2020.acl-main.488" ] }, "num": null, "urls": [], "raw_text": "Paul Pu Liang, Irene Mengze Li, Emily Zheng, Yao Chong Lim, Ruslan Salakhutdinov, and Louis- Philippe Morency. 2020. Towards debiasing sentence representations. In Proceedings of the 58th Annual Meeting of the Association for Computational Lin- guistics, pages 5502-5515, Online. Association for Computational Linguistics.", "links": null }, "BIBREF25": { "ref_id": "b25", "title": "Black is to criminal as caucasian is to police: Detecting and removing multiclass bias in word embeddings", "authors": [ { "first": "Thomas", "middle": [], "last": "Manzini", "suffix": "" }, { "first": "Yao", "middle": [], "last": "Lim", "suffix": "" }, { "first": "Alan", "middle": [ "W" ], "last": "Chong", "suffix": "" }, { "first": "Yulia", "middle": [], "last": "Black", "suffix": "" }, { "first": "", "middle": [], "last": "Tsvetkov", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", "volume": "1", "issue": "", "pages": "615--621", "other_ids": { "DOI": [ "10.18653/v1/N19-1062" ] }, "num": null, "urls": [], "raw_text": "Thomas Manzini, Lim Yao Chong, Alan W Black, and Yulia Tsvetkov. 2019. Black is to criminal as cau- casian is to police: Detecting and removing multi- class bias in word embeddings. In Proceedings of the 2019 Conference of the North American Chap- ter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 615-621, Minneapolis, Min- nesota. Association for Computational Linguistics.", "links": null }, "BIBREF26": { "ref_id": "b26", "title": "On measuring social biases in sentence encoders", "authors": [ { "first": "Chandler", "middle": [], "last": "May", "suffix": "" }, { "first": "Alex", "middle": [], "last": "Wang", "suffix": "" }, { "first": "Shikha", "middle": [], "last": "Bordia", "suffix": "" }, { "first": "Samuel", "middle": [ "R" ], "last": "Bowman", "suffix": "" }, { "first": "Rachel", "middle": [], "last": "Rudinger", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", "volume": "1", "issue": "", "pages": "622--628", "other_ids": { "DOI": [ "10.18653/v1/N19-1063" ] }, "num": null, "urls": [], "raw_text": "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 Chap- ter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 622-628, Minneapolis, Min- nesota. Association for Computational Linguistics.", "links": null }, "BIBREF27": { "ref_id": "b27", "title": "StereoSet: Measuring stereotypical bias in pretrained language models", "authors": [ { "first": "Moin", "middle": [], "last": "Nadeem", "suffix": "" }, { "first": "Anna", "middle": [], "last": "Bethke", "suffix": "" }, { "first": "Siva", "middle": [], "last": "Reddy", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing", "volume": "1", "issue": "", "pages": "5356--5371", "other_ids": { "DOI": [ "10.18653/v1/2021.acl-long.416" ] }, "num": null, "urls": [], "raw_text": "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 Lin- guistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 5356-5371, Online. Association for Computational Linguistics.", "links": null }, "BIBREF28": { "ref_id": "b28", "title": "CrowS-pairs: A challenge dataset for measuring social biases in masked language models", "authors": [ { "first": "Nikita", "middle": [], "last": "Nangia", "suffix": "" }, { "first": "Clara", "middle": [], "last": "Vania", "suffix": "" }, { "first": "Rasika", "middle": [], "last": "Bhalerao", "suffix": "" }, { "first": "Samuel", "middle": [ "R" ], "last": "Bowman", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)", "volume": "", "issue": "", "pages": "1953--1967", "other_ids": { "DOI": [ "10.18653/v1/2020.emnlp-main.154" ] }, "num": null, "urls": [], "raw_text": "Nikita Nangia, Clara Vania, Rasika Bhalerao, and Samuel R. Bowman. 2020. CrowS-pairs: A chal- lenge dataset for measuring social biases in masked language models. In Proceedings of the 2020 Con- ference on Empirical Methods in Natural Language Processing (EMNLP), pages 1953-1967, Online. As- sociation for Computational Linguistics.", "links": null }, "BIBREF29": { "ref_id": "b29", "title": "HONEST: Measuring hurtful sentence completion in language models", "authors": [ { "first": "Debora", "middle": [], "last": "Nozza", "suffix": "" }, { "first": "Federico", "middle": [], "last": "Bianchi", "suffix": "" }, { "first": "Dirk", "middle": [], "last": "Hovy", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", "volume": "", "issue": "", "pages": "2398--2406", "other_ids": { "DOI": [ "10.18653/v1/2021.naacl-main.191" ] }, "num": null, "urls": [], "raw_text": "Debora Nozza, Federico Bianchi, and Dirk Hovy. 2021. HONEST: Measuring hurtful sentence completion in language models. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2398-2406, Online. Association for Computational Linguistics.", "links": null }, "BIBREF30": { "ref_id": "b30", "title": "Measuring Harmful Sentence Completion in Language Models for LGBTQIA+ Individuals", "authors": [ { "first": "Debora", "middle": [], "last": "Nozza", "suffix": "" }, { "first": "Federico", "middle": [], "last": "Bianchi", "suffix": "" }, { "first": "Anne", "middle": [], "last": "Lauscher", "suffix": "" }, { "first": "Dirk", "middle": [], "last": "Hovy", "suffix": "" } ], "year": 2022, "venue": "Proceedings of the Second Workshop on Language Technology for Equality, Diversity and Inclusion", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Debora Nozza, Federico Bianchi, Anne Lauscher, and Dirk Hovy. 2022. Measuring Harmful Sentence Com- pletion in Language Models for LGBTQIA+ Individ- uals. In Proceedings of the Second Workshop on Language Technology for Equality, Diversity and In- clusion. Association for Computational Linguistics.", "links": null }, "BIBREF31": { "ref_id": "b31", "title": "Unintended bias in misogyny detection. WI '19", "authors": [ { "first": "Debora", "middle": [], "last": "Nozza", "suffix": "" }, { "first": "Claudia", "middle": [], "last": "Volpetti", "suffix": "" }, { "first": "Elisabetta", "middle": [], "last": "Fersini", "suffix": "" } ], "year": 2019, "venue": "", "volume": "", "issue": "", "pages": "149--155", "other_ids": { "DOI": [ "10.1145/3350546.3352512" ] }, "num": null, "urls": [], "raw_text": "Debora Nozza, Claudia Volpetti, and Elisabetta Fersini. 2019. Unintended bias in misogyny detection. WI '19, page 149-155, New York, NY, USA. Association for Computing Machinery.", "links": null }, "BIBREF32": { "ref_id": "b32", "title": "Probing toxic content in large pre-trained language models", "authors": [ { "first": "Nedjma", "middle": [], "last": "Ousidhoum", "suffix": "" }, { "first": "Xinran", "middle": [], "last": "Zhao", "suffix": "" }, { "first": "Tianqing", "middle": [], "last": "Fang", "suffix": "" }, { "first": "Yangqiu", "middle": [], "last": "Song", "suffix": "" }, { "first": "Dit-Yan", "middle": [], "last": "Yeung", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing", "volume": "1", "issue": "", "pages": "4262--4274", "other_ids": { "DOI": [ "10.18653/v1/2021.acl-long.329" ] }, "num": null, "urls": [], "raw_text": "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 Lan- guage Processing (Volume 1: Long Papers), pages 4262-4274, Online. Association for Computational Linguistics.", "links": null }, "BIBREF33": { "ref_id": "b33", "title": "Beyond accuracy: Behavioral testing of NLP models with CheckList", "authors": [ { "first": "Tongshuang", "middle": [], "last": "Marco Tulio Ribeiro", "suffix": "" }, { "first": "Carlos", "middle": [], "last": "Wu", "suffix": "" }, { "first": "Sameer", "middle": [], "last": "Guestrin", "suffix": "" }, { "first": "", "middle": [], "last": "Singh", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "4902--4912", "other_ids": { "DOI": [ "10.18653/v1/2020.acl-main.442" ] }, "num": null, "urls": [], "raw_text": "Marco Tulio Ribeiro, Tongshuang Wu, Carlos Guestrin, and Sameer Singh. 2020. Beyond accuracy: Be- havioral testing of NLP models with CheckList. In Proceedings of the 58th Annual Meeting of the Asso- ciation for Computational Linguistics, pages 4902- 4912, Online. Association for Computational Lin- guistics.", "links": null }, "BIBREF34": { "ref_id": "b34", "title": "HateCheck: Functional tests for hate speech detection models", "authors": [ { "first": "Paul", "middle": [], "last": "R\u00f6ttger", "suffix": "" }, { "first": "Bertie", "middle": [], "last": "Vidgen", "suffix": "" }, { "first": "Dong", "middle": [], "last": "Nguyen", "suffix": "" }, { "first": "Zeerak", "middle": [], "last": "Waseem", "suffix": "" }, { "first": "Helen", "middle": [], "last": "Margetts", "suffix": "" }, { "first": "Janet", "middle": [], "last": "Pierrehumbert", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing", "volume": "1", "issue": "", "pages": "41--58", "other_ids": { "DOI": [ "10.18653/v1/2021.acl-long.4" ] }, "num": null, "urls": [], "raw_text": "Paul R\u00f6ttger, Bertie Vidgen, Dong Nguyen, Zeerak Waseem, Helen Margetts, and Janet Pierrehumbert. 2021. HateCheck: Functional tests for hate speech detection models. In Proceedings of the 59th An- nual Meeting of the Association for Computational Linguistics and the 11th International Joint Confer- ence on Natural Language Processing (Volume 1: Long Papers), pages 41-58, Online. Association for Computational Linguistics.", "links": null }, "BIBREF35": { "ref_id": "b35", "title": "Predictive biases in natural language processing models: A conceptual framework and overview", "authors": [ { "first": "", "middle": [], "last": "Deven Santosh", "suffix": "" }, { "first": "H", "middle": [ "Andrew" ], "last": "Shah", "suffix": "" }, { "first": "Dirk", "middle": [], "last": "Schwartz", "suffix": "" }, { "first": "", "middle": [], "last": "Hovy", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "5248--5264", "other_ids": { "DOI": [ "10.18653/v1/2020.acl-main.468" ] }, "num": null, "urls": [], "raw_text": "Deven Santosh Shah, H. Andrew Schwartz, and Dirk Hovy. 2020. Predictive biases in natural language processing models: A conceptual framework and overview. In Proceedings of the 58th Annual Meet- ing of the Association for Computational Linguistics, pages 5248-5264, Online. Association for Computa- tional Linguistics.", "links": null }, "BIBREF36": { "ref_id": "b36", "title": "Societal biases in language generation: Progress and challenges", "authors": [ { "first": "Emily", "middle": [], "last": "Sheng", "suffix": "" }, { "first": "Kai-Wei", "middle": [], "last": "Chang", "suffix": "" }, { "first": "Prem", "middle": [], "last": "Natarajan", "suffix": "" }, { "first": "Nanyun", "middle": [], "last": "Peng", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing", "volume": "1", "issue": "", "pages": "4275--4293", "other_ids": { "DOI": [ "10.18653/v1/2021.acl-long.330" ] }, "num": null, "urls": [], "raw_text": "Emily Sheng, Kai-Wei Chang, Prem Natarajan, and Nanyun Peng. 2021. Societal biases in language generation: Progress and challenges. 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 Papers), pages 4275-4293, Online. Association for Computational Linguistics.", "links": null }, "BIBREF37": { "ref_id": "b37", "title": "The woman worked as a babysitter: On biases in language generation", "authors": [ { "first": "Emily", "middle": [], "last": "Sheng", "suffix": "" }, { "first": "Kai-Wei", "middle": [], "last": "Chang", "suffix": "" }, { "first": "Premkumar", "middle": [], "last": "Natarajan", "suffix": "" }, { "first": "Nanyun", "middle": [], "last": "Peng", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)", "volume": "", "issue": "", "pages": "3407--3412", "other_ids": { "DOI": [ "10.18653/v1/D19-1339" ] }, "num": null, "urls": [], "raw_text": "Emily Sheng, Kai-Wei Chang, Premkumar Natarajan, and Nanyun Peng. 2019. The woman worked as a babysitter: On biases in language generation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Lan- guage Processing (EMNLP-IJCNLP), pages 3407- 3412, Hong Kong, China. Association for Computa- tional Linguistics.", "links": null }, "BIBREF38": { "ref_id": "b38", "title": "A survey on gender bias in natural language processing", "authors": [ { "first": "Karolina", "middle": [], "last": "Stanczak", "suffix": "" }, { "first": "Isabelle", "middle": [], "last": "Augenstein", "suffix": "" } ], "year": 2021, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:2112.14168" ] }, "num": null, "urls": [], "raw_text": "Karolina Stanczak and Isabelle Augenstein. 2021. A survey on gender bias in natural language processing. arXiv preprint arXiv:2112.14168.", "links": null }, "BIBREF39": { "ref_id": "b39", "title": "What are the biases in my word embedding?", "authors": [ { "first": "Nathaniel", "middle": [], "last": "Swinger", "suffix": "" }, { "first": "Maria", "middle": [], "last": "De-Arteaga", "suffix": "" }, { "first": "Neil", "middle": [ "Thomas" ], "last": "Heffernan", "suffix": "" }, { "first": "I", "middle": [ "V" ], "last": "Mark", "suffix": "" }, { "first": "D", "middle": [ "M" ], "last": "Leiserson", "suffix": "" }, { "first": "Adam Tauman", "middle": [], "last": "Kalai", "suffix": "" } ], "year": 2019, "venue": "Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society, AIES '19", "volume": "", "issue": "", "pages": "305--311", "other_ids": { "DOI": [ "10.1145/3306618.3314270" ] }, "num": null, "urls": [], "raw_text": "Nathaniel Swinger, Maria De-Arteaga, Neil Thomas Heffernan IV, Mark DM Leiserson, and Adam Tau- man Kalai. 2019. What are the biases in my word embedding? In Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society, AIES '19, page 305-311, New York, NY, USA. Association for Computing Machinery.", "links": null }, "BIBREF40": { "ref_id": "b40", "title": "Low frequency names exhibit bias and overfitting in contextualizing language models", "authors": [ { "first": "Robert", "middle": [], "last": "Wolfe", "suffix": "" }, { "first": "Aylin", "middle": [], "last": "Caliskan", "suffix": "" } ], "year": 2021, "venue": "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "518--532", "other_ids": { "DOI": [ "10.18653/v1/2021.emnlp-main.41" ] }, "num": null, "urls": [], "raw_text": "Robert Wolfe and Aylin Caliskan. 2021. Low frequency names exhibit bias and overfitting in contextualizing language models. In Proceedings of the 2021 Con- ference on Empirical Methods in Natural Language Processing, pages 518-532, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.", "links": null }, "BIBREF41": { "ref_id": "b41", "title": "Hurtful words: Quantifying biases in clinical contextual word embeddings", "authors": [ { "first": "Haoran", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Amy", "middle": [ "X" ], "last": "Lu", "suffix": "" }, { "first": "Mohamed", "middle": [], "last": "Abdalla", "suffix": "" }, { "first": "Matthew", "middle": [], "last": "Mcdermott", "suffix": "" }, { "first": "Marzyeh", "middle": [], "last": "Ghassemi", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the ACM Conference on Health, Inference, and Learning, CHIL '20", "volume": "", "issue": "", "pages": "110--120", "other_ids": { "DOI": [ "10.1145/3368555.3384448" ] }, "num": null, "urls": [], "raw_text": "Haoran Zhang, Amy X. Lu, Mohamed Abdalla, Matthew McDermott, and Marzyeh Ghassemi. 2020. Hurtful words: Quantifying biases in clinical con- textual word embeddings. In Proceedings of the ACM Conference on Health, Inference, and Learn- ing, CHIL '20, page 110-120, New York, NY, USA. Association for Computing Machinery.", "links": null } }, "ref_entries": { "FIGREF0": { "uris": null, "text": "The figure shows an example of the possible integration of Social Bias tests into a development pipeline.", "num": null, "type_str": "figure" } } } }