Languages Left Behind: Automation in Trust & Safety Across Border, Languages, and Contexts

The machine-learning based tools that online services use to moderate speech are integral for detecting and removing content that platforms don’t want to host at scale, such as spam, graphic violence, child sexual abuse material, and other abusive content. Trust and safety teams must consider how these tools can be used in a transparent and rights-respecting way, and how that use changes with the language and context of content. For text classifiers, in particular, dynamics around the availability of training data sets, resource allocation within companies, and access to language experts can heighten the challenges around moderating content in languages other than English. This panel will discuss the Center for Democracy & Technology's new paper "Lost in Translation: Large Language Models in Non-English Content Analysis," with a group of trust and safety and AI experts situating these tools within the broader trust and safety apparatus. The conversation will focus on how we can improve these tools' efficacy in non-English contexts, how to better incorporate input from local language experts, and how to mitigate the risks these tools might pose on users’ ability to speak and exchange information freely.

Location Name
Ballroom B
Thursday, July 13, 2023
2:50 PM - 3:40 PM
Session Type
Session Themes
Designing for Safety, Partnerships
All TrustCon attendees
Will this session be recorded?