用贝叶斯方法估算词嵌入偏差的不确定性

IF 9.3 2区 计算机科学 Computational Linguistics Pub Date : 2024-01-08 DOI:10.1162/coli_a_00507
Alicja Dobrzeniecka, Rafal Urbaniak
{"title":"用贝叶斯方法估算词嵌入偏差的不确定性","authors":"Alicja Dobrzeniecka, Rafal Urbaniak","doi":"10.1162/coli_a_00507","DOIUrl":null,"url":null,"abstract":"Multiple measures, such as WEAT or MAC, attempt to quantify the magnitude of bias present in word embeddings in terms of a single-number metric. However, such metrics and the related statistical significance calculations rely on treating pre-averaged data as individual data points and employing bootstrapping techniques with low sample sizes. We show that similar results can be easily obtained using such methods even if the data are generated by a null model lacking the intended bias. Consequently, we argue that this approach generates false confidence. To address this issue, we propose a Bayesian alternative: hierarchical Bayesian modeling, which enables a more uncertainty-sensitive inspection of bias in word embeddings at different levels of granularity. To showcase our method, we apply it to Religion, Gender, and Race word lists from the original research, together with our control neutral word lists. We deploy the method using Google, Glove, and Reddit embeddings. Further, we utilize our approach to evaluate a debiasing technique applied to the Reddit word embedding. Our findings reveal a more complex landscape than suggested by the proponents of single-number metrics. The datasets and source code for the paper are publicly available.","PeriodicalId":49089,"journal":{"name":"Computational Linguistics","volume":"14 1","pages":""},"PeriodicalIF":9.3000,"publicationDate":"2024-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A Bayesian approach to uncertainty in word embedding bias estimation\",\"authors\":\"Alicja Dobrzeniecka, Rafal Urbaniak\",\"doi\":\"10.1162/coli_a_00507\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Multiple measures, such as WEAT or MAC, attempt to quantify the magnitude of bias present in word embeddings in terms of a single-number metric. However, such metrics and the related statistical significance calculations rely on treating pre-averaged data as individual data points and employing bootstrapping techniques with low sample sizes. We show that similar results can be easily obtained using such methods even if the data are generated by a null model lacking the intended bias. Consequently, we argue that this approach generates false confidence. To address this issue, we propose a Bayesian alternative: hierarchical Bayesian modeling, which enables a more uncertainty-sensitive inspection of bias in word embeddings at different levels of granularity. To showcase our method, we apply it to Religion, Gender, and Race word lists from the original research, together with our control neutral word lists. We deploy the method using Google, Glove, and Reddit embeddings. Further, we utilize our approach to evaluate a debiasing technique applied to the Reddit word embedding. Our findings reveal a more complex landscape than suggested by the proponents of single-number metrics. The datasets and source code for the paper are publicly available.\",\"PeriodicalId\":49089,\"journal\":{\"name\":\"Computational Linguistics\",\"volume\":\"14 1\",\"pages\":\"\"},\"PeriodicalIF\":9.3000,\"publicationDate\":\"2024-01-08\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Computational Linguistics\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://doi.org/10.1162/coli_a_00507\",\"RegionNum\":2,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computational Linguistics","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1162/coli_a_00507","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

WEAT 或 MAC 等多种度量方法都试图用单一数字度量词嵌入中存在的偏差大小。然而,这些指标和相关的统计显著性计算依赖于将预平均数据视为单个数据点,并采用低样本量的引导技术。我们的研究表明,即使数据是由缺乏预期偏差的空模型生成的,使用这种方法也能轻松获得类似的结果。因此,我们认为这种方法会产生错误的置信度。为了解决这个问题,我们提出了一种贝叶斯替代方法:分层贝叶斯建模,它可以在不同粒度水平上对词嵌入中的偏差进行对不确定性更加敏感的检验。为了展示我们的方法,我们将其应用于原始研究中的宗教、性别和种族词表,以及我们的对照中性词表。我们使用 Google、Glove 和 Reddit 嵌入来部署该方法。此外,我们还利用我们的方法对应用于 Reddit 词嵌入的去伪存真技术进行了评估。我们的研究结果揭示了比单一数字度量支持者所认为的更为复杂的情况。本文的数据集和源代码均可公开获取。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
A Bayesian approach to uncertainty in word embedding bias estimation
Multiple measures, such as WEAT or MAC, attempt to quantify the magnitude of bias present in word embeddings in terms of a single-number metric. However, such metrics and the related statistical significance calculations rely on treating pre-averaged data as individual data points and employing bootstrapping techniques with low sample sizes. We show that similar results can be easily obtained using such methods even if the data are generated by a null model lacking the intended bias. Consequently, we argue that this approach generates false confidence. To address this issue, we propose a Bayesian alternative: hierarchical Bayesian modeling, which enables a more uncertainty-sensitive inspection of bias in word embeddings at different levels of granularity. To showcase our method, we apply it to Religion, Gender, and Race word lists from the original research, together with our control neutral word lists. We deploy the method using Google, Glove, and Reddit embeddings. Further, we utilize our approach to evaluate a debiasing technique applied to the Reddit word embedding. Our findings reveal a more complex landscape than suggested by the proponents of single-number metrics. The datasets and source code for the paper are publicly available.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
Computational Linguistics
Computational Linguistics Computer Science-Artificial Intelligence
自引率
0.00%
发文量
45
期刊介绍: Computational Linguistics is the longest-running publication devoted exclusively to the computational and mathematical properties of language and the design and analysis of natural language processing systems. This highly regarded quarterly offers university and industry linguists, computational linguists, artificial intelligence and machine learning investigators, cognitive scientists, speech specialists, and philosophers the latest information about the computational aspects of all the facets of research on language.
期刊最新文献
Dotless Arabic text for Natural Language Processing Humans Learn Language from Situated Communicative Interactions. What about Machines? Exploring temporal sensitivity in the brain using multi-timescale language models: an EEG decoding study Meaning beyond lexicality: Capturing Pseudoword Definitions with Language Models Perception of Phonological Assimilation by Neural Speech Recognition Models
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1