Oversight Board overturns Meta's decisions in the Violence Against Women cases
The Oversight Board has overturned Meta’s decisions to remove two Instagram posts which condemned gender-based violence. The Board recommends that Meta include the exception for allowing content that condemns or raises awareness of gender-based violence in the public language of the Hate Speech policy, as well as update its internal guidance to reviewers to ensure such posts are not mistakenly removed.
About the cases
In this decision, the Board considers two posts from an Instagram user in Sweden together. Meta removed both posts for violating its Hate Speech Community Standard. After the Board identified the cases, Meta decided that the first post had been removed in error but maintained its decision on the second post.
The first post contains a video with an audio recording and its transcription, both in Swedish, of a woman describing her experience in a violent intimate relationship, including how she felt unable to discuss the situation with her family. The caption notes that the woman in the audio recording consented to its publication, and that the voice has been modified. It says that there is a culture of blaming victims of gender-based violence, and little understanding of how difficult it is for women to leave a violent partner. The caption says, “men murder, rape and abuse women mentally and physically – all the time, every day.” It also shares information about support organizations for victims of intimate partner violence, mentions the International Day for the Elimination of Violence against Women, and says it hopes women reading the post will realize they are not alone.
After one of Meta’s classifiers identified the content as potentially violating Meta’s rules on hate speech, two reviewers examined the post and removed it. This decision was then upheld by the same two reviewers on different levels of review. As a result of the Board selecting this case, Meta determined that it had removed the content in error, restoring the post.
As the Board began to assess the first post, it received another appeal from the same user. The second post, also shared on Instagram, contains a video of a woman speaking in Swedish and pointing at words written in Swedish on a notepad. In the video, the speaker says that although she is a man-hater, she does not hate all men. She also states that she is a man-hater for condemning misogyny and that hating men is rooted in fear of violence. Meta removed the content for violating its rules on hate speech. The user appealed the removal to Meta, but the company upheld its original decision after human review. After being informed that the Board had selected this case, Meta did not change its position.
Since at least 2017, digital campaigns have highlighted that Facebook’s hate speech policies result in the removal of phrases associated with calling attention to gender-based violence and harassment. For example, women and activists have coordinated posting phrases such as “men are trash” and “ men are scum” and protested their subsequent removal on the grounds of being anti-men hate speech.
Key findings
The Board finds that neither of the two posts violates Meta’s rules on hate speech.
On the first post, the Board finds that the statement “Men murder, rape and abuse women mentally and physically – all the time, every day” is a qualified statement which does not violate Meta’s Hate Speech policy. Given that the post refers to international campaigns against violence against women and provides local resources for organizations that work to help women victims, it is clear the language describes men who commit violence against women.
In addition, the Board finds that the second post is not an expression of contempt towards men but condemns violence against women and discusses the roots of gender-based hate. While Meta argues that the user’s statement that she does not hate all men does not impact the assessment of other parts of the post, the Board disagrees and assesses the post as a whole. The Board finds that the other aspects of the post that Meta cited as potentially violating are not violating when read within the context of the post. Some Board Members disagreed that the posts in question did not violate Meta’s hate speech rules.
The Board is concerned that Meta’s approach to enforcing gender-based hate speech may result in the disproportionate removal of content raising awareness of and condemning gender-based violence. Meta states, for example, that the first post should be allowed on its platforms and that the Hate Speech policy is “designed to allow room for raising awareness of gender-based violence.” However, neither the public-facing policy nor its internal guideline documents to moderators contain clear guidance to ensure that posts like these would not be mistakenly removed. The company’s confusing guidance makes it virtually impossible for moderators to reach the right conclusion. While Meta relied on contextual cues to determine the first post was not violating once it was identified by the Board, the company’s guidance for moderators limits the possibility of contextual analysis significantly.
The Board finds that within this context, it is critical that statements that condemn and raise awareness of gender-based violence not be mistakenly removed. The Board’s concern that this may be happening is particularly pronounced given that an allowance for this type of content, while highlighted by Meta, is not communicated clearly to the public and the guidance provided to moderators is confusing. To address this, Meta should clarify its public rules and provide appropriate guidance to moderators that better reflects this allowance.
The Oversight Board’s decision
The Oversight Board overturns Meta’s decisions to remove both posts. The Board recommends that Meta:
- Include the exception for allowing content that condemns or raises awareness of gender-based violence in the public language of the Hate Speech policy.
- Update guidance to its at-scale moderators with specific attention to rules around qualification to ensure that content condemning and raising awareness of gender-based violence is not removed in error.
- Update its Transparency Center with information on what penalties are associated with the accumulation of strikes on Instagram. The Board appreciates that Meta has provided additional information about strikes for Facebook users in response to Board recommendations. It believes this should be done for Instagram users as well.
- Assess how its current review routing protocol impacts accuracy. The Board believes Meta would improve content moderation accuracy by adjusting this protocol to prioritize sending secondary review jobs to different reviewers than those who previously assessed the content.
For further information
To read the full decision, click here.
To read a synopsis of public comments for these cases, please click the attachment below.