Building "Letra Chica"

The use of social media is governed by the rules that each platform establishes regarding user behavior within the community. Known as community guidelines, these set out guidelines for creating accounts, using third-party material, and sharing content, among other things. For most people, community guidelines tend to go unnoticed or only come into focus when they interfere in some way with the user experience. In other words, they become part of the fine print that no one ever reads. This informal regulatory and adjudicatory environment on social media—generally understood as content moderation—often creates ongoing tension with national laws and international human rights instruments. In particular, there are many questions regarding the exercise of freedom of expression, due process, and transparency.

The use of social media is governed by the rules that each platform establishes regarding user behavior within the community. Known as community guidelines, these set forth guidelines for creating accounts, using third-party material, and sharing content, among other things. For most people, community guidelines tend to go unnoticed or only come into focus when they interfere in some way with the user experience. In other words, they become part of the fine print that no one ever reads. This informal environment of regulation and adjudication on social media—generally understood as content moderation—often creates ongoing tension with national laws and international human rights instruments. In particular, there are many questions regarding the exercise of freedom of expression, due process, and transparency. This year, both the COVID-19 pandemic and the U.S. presidential election illustrated the magnitude of this issue. On the one hand—as CELE addresses in a forthcoming publication—the pandemic led social media platforms to create rules on the fly to curb misinformation and give greater visibility to public health sources. On the other hand, the false allegations of election fraud that Donald Trump publicly aired during the campaign forced these intermediaries to prepare for a scenario that actually came to pass: the U.S. president used Twitter and Facebook to undermine his country’s democratic process. The social media platforms’ unprecedented—and not without controversy—response was to consistently flag and label the disinformation spread by Trump and the Republican Party. It is in this context that LetraChica emerged: an initiative of the CELE, with the support of Linterna Verde, to shed more light on the process of developing community guidelines for Facebook, Twitter, and YouTube. We’re interested in understanding how these changes unfold over time, who is included in them, how they’re translated into Spanish, and how they respond to specific political and social contexts. What do we do, and how? Letra Chica monitors changes to the community guidelines of Facebook, Twitter, and YouTube using the following sources:

• URL Changes. Using automated software, Letra Chica has been monitoring approximately 300 links or URLs since May 2020 where the community guidelines for Facebook, Twitter, and YouTube are posted in English and Spanish. Every 24 hours, the tool checks the content of these sites and sends a notification of any changes. Literally any change: it even records the addition of a comma.
• Official announcements. We constantly monitor the platforms’ blogs and news sites. Measures are announced there that sometimes do not appear in the community standards, and the rationale behind certain changes is even explained.
• Recent updates, product team minutes, and transparency reports. Facebook, in particular, publishes the meeting minutes of the company’s product team (Product Policy Forum), where the rules are analyzed and potential changes are discussed. Facebook also has a Recent Updates (Recent Updates), where it highlights changes it has made to its Community Standards since May 2018 and where additional information on the rationale behind the changes is sometimes provided. Finally, Facebook and YouTube publish periodic reports on the enforcement of their policies. We review these documents to identify relevant updates. The Letra Chica website compiles and explains the results of this process. It also offers some analysis and provides links to other relevant sources of information. Given the scope of this task, in this initial phase we are focusing Letra Chica’s content on the following policies: (i) hate speech; (ii) nudity and sexual content; (iii) elections and other civic engagement processes; and (iv) public interest exceptions. Other categories will be added as the project progresses. The articles are accompanied by tables that highlight and describe each of the changes. Why in English and Spanish? We track changes in both languages because we want to understand how long it takes platforms to adopt changes for Spanish-speaking countries after introducing them in the original English version. In addition, we want to identify relevant differences between the original text and its Spanish translation. We have noticed that some changes take time to reach the Spanish version. For example, modifications made by Facebook to its policy on hate speech regarding insults (February 2020) or the ban on Holocaust denial (October 2020) have not been implemented in Spanish. It is unclear whether the failure to incorporate the change means that the new rule does not yet apply in Spanish-speaking countries, or whether it does apply despite the lack of translation. Furthermore, the translations are not always clear. For example, as part of its policy on hate speech, Facebook prohibits calls for exclusion. However, it allows the use of gender-exclusive language to control access to health-related support groups, such as a breastfeeding group exclusively for women. However, the term “gender-exclusive language” was inappropriately translated into Spanish as “sexist language.” What did we find? An early finding in the process was that the task proved much more complex than common sense would have suggested, given the way platforms create and present their policies. The policies that impact user activity are scattered across multiple documents, sites, and formats. Ad hoc adjustments, such as those made due to the pandemic or the 2020 U.S. elections, are not always incorporated into the body of the community guidelines but are announced in a less organized manner on the companies’ blogs or, in the case of Facebook, on the personal account of its CEO, Mark Zuckerberg. Some documents provide additional guidance on the application of the rules—details not necessarily found in the Community Standards. For example, in its hate speech policy, Facebook prohibits generalizations (“immigrants are terrorists”), but allows statements related to specific incidents (“a 19-year-old immigrant stabbed a woman yesterday”), provided that such statements are “qualified.” However, the policy does not explain what constitutes a qualified statement. The answer lies in the aforementioned minutes from Facebook’s product team meetings: permitted statements are those that include: (i) qualifiers that limit the scope of protected characteristics (such as race, nationality, gender, or religion) (ii) indicators that the statements refer to a specific event, pattern of events, or discussion of general behavior, or (iii) language describing firsthand experiences or reporting (for example, “I saw” or “this happened to me”). In this same vein, it is difficult to know when a platform is actually changing its policies or when it is simply making them more transparent. Following the previous example, if Facebook were to include an explanation of “qualified” statements in its policies, it would not be changing the policy, but rather making the way it already applies it more transparent. Finally, we have seen that Facebook’s “Recent Updates” page is a valuable resource, but not entirely reliable. For example, in August 2019, Facebook made a change to its policy on the non-consensual sharing of intimate images, but its “Recent Updates” page lists the change as having taken place in December 2019. What does the project offer, and what are its limitations? Letra Chica is a pioneering project in the region. However, it is not without limitations that must be taken into account. First, we depend on the automated monitoring tool functioning properly. Second, we have faced—and will continue to face—technical dilemmas, such as the fact that in some cases the software only allows us to capture the mobile version of certain URLs—and, as a result, there may be discrepancies with the desktop version. Finally, it is impossible to rule out human error when comparing and explaining the changes. Faced with a monumental task like the one we set out to accomplish, we have chosen to carry out the monitoring in stages, constantly refine and adapt the methodology, and clearly explain our progress and the final product. You can follow this process on this blog. Ultimately, the goal is clear: through this site, CELE and Linterna Verde aim to make an informed and technical contribution to the discussion on online content moderation—perhaps the most important debate regarding the exercise of free speech in the digital environment. And we want to do so from the perspective of knowledge and the context of our region.