DeVan Hankerson Madrigal, Research Manager at @CenDemTech. Previously @GTownTechLaw. All tweets are my views only; retweets and likes may not be. she/her
Most content moderation systems weren’t built with the Global South in mind. CDT’s new research shows how they fail in Tamil, Kiswahili, Quechua & more.
Join our Aug 7 webinar for findings, recommendations, and global voices on language equity in tech.
🚨 تقرير جديد 🚨
١/يسعدنا ان نعلن عن نشر الترجمة العربية للتقرير " إدارة المحتوى العربي المغاربي على وسائل التواصل الاجتماعي"
https://t.co/adhAvGRGq1 #إشراف_المحتوي
🚨NEW REPORT:🚨 We’re excited to announce the release of the Arabic language version of our report, “Moderating Maghrebi Arabic Content on Social Media.”https://t.co/adhAvGRGq1 #ContentModeration
https://t.co/x8uSBvnC4E
.@MonaElswah in @Politico’s MorningTech talking ab CDT’s new research on #ContentModeration in Maghrebi Arabic. “Content moderation for a dialect of Arabic spoken in Northern Africa is lacking, according to a new report by @CenDemTech.”
* Direct link: https://t.co/adhAvGRGq1
Excited to share my summer research @CenDemTech😃
I interviewed NLP researchers building automated mod tools in low-resource languages🎙️
Takeaway: Mod is hard but more so for low-resource languages due to colonial biases+systemic inequities affecting mod research pipeline ⬇️
Maghrebi Arabic users deploy "algospeak" to evade moderation, citing political censorship concerns. They also rely on mass reporting to combat harassing content.
There is a real issue of distrust towards social media companies growing in the Maghreb region.
The lack of representation of Maghrebi Arabic speakers in developing automated moderation systems significantly reduces the accuracy and fairness of moderation decisions.
We need to hire native annotators and to hire more diverse NLP teams to get things right. 👨🏾💻👩🏾💻
@DigiCitizensOrg@ISOC_Foundation@monaelswah The lack of representation of Maghrebi Arabic speakers in developing automated moderation systems significantly reduces the accuracy and fairness of moderation decisions. Check out the report for more insights: https://t.co/adhAvGRGq1
@DigiCitizensOrg@ISOC_Foundation@monaelswah The report reveals that content moderators reviewing Arabic content are often assigned content from any country in the region regardless of their native dialect, leading to errors and inconsistencies in moderation. https://t.co/adhAvGRGq1
NEW RESEARCH: CDT’s report (in partnership with @DigiCitizensOrg & funded by @ISOC_Foundation), “Moderating Maghrebi Arabic Content on Social Media,” analyzes the challenges of moderating Global South languages. https://t.co/adhAvGRGq1
👩🏾💻👩🏾💻New research from me and @CenDemTech
We know too little about content moderation of Maghrebi Arabic dialects, so I did some research on this.
Most interestingly, I was able to speak to content moderators in the Maghreb region. I found some interesting things (a thread) 🧵
Join @CenDemTech as we kick off FOSO, our annual Future of Speech Online conference, ft. the incredible @noupside and a panel with experts from @factchequeado, @dangerousspeech, @civilrightsorg and more. https://t.co/vYvcpfAZPY for the program & RSVP here: https://t.co/N8bM5ZGOTU
LISTEN NOW: Join experts Vandana Chaudhry & Lydia X. Z. Brown in a new episode of CDT’s Tech Talk, as they explore the intersection of disability rights, data, and algorithmic decision-making in the U.S. & India. https://t.co/2rVqiue68h
🚨NEW REPORT: To understand how to effectively regulate AI, we need to understand how people use it. How can that be done while preserving users’ privacy? Our new report, “Grounding AI Policy: Towards Researcher Access to AI Usage Data” provides insight: https://t.co/QCiPBaRebG
.@CenDemTech’s research ft. in @Forbes: “@VP Kamala Harris faces unique challenges as a woman of color in politics… A study from CDT found that [women of color] candidates face higher rates of violent threats online.” https://t.co/8s7tCfxb8x cc: @iamAliciaGtv
Our @cendemtech research from 2022 is more relevant than ever—showing that women of color political candidates are at least twice as likely to be targeted with disinformation and abuse online. @dvhankerson@thakurdhanaraj
https://t.co/7aiFo7sqJx
Thank you @_51percent for the conversation. As @CenDemTech@PlanGlobal and many others research has shown, online #GBV directly effects democratic participation, media pluralism & the future of gender equity, especially for women & non-binary POC #EndGBV#FreeExpression#DSA