5/27/22, 8:18 PM Integrating autoconversion: Facebook’s path from Zawgyi to Unicode - Engineering at Meta Next, the server checks whether it is loading Burmese content. If the content encoding and device encoding don’t mat render properly. For instance, if a post was input with a Unicode content encoding, but it’s being read on a Zawgyi enc Zawgyi device. It’s important to train this model on Facebook content instead of on other publicly accessible content on the web. Peo paper: Facebook posts and messages are generally shorter and less formal, and they contain abbreviations, slang, an people share and read on our apps. Integrating autoconversion at Facebook scale The next challenge was to integrate this conversion across the different types of content that people can create on o names, comments, video subtitles, private messages, and more. Running our detection and conversion every time som resources required. There’s no single pipeline through which all possible Facebook content passes, which makes it dif Furthermore, not every web request is made from a person’s device. For example, when notifications and messages a comments are often very short, lowering detection accuracy. The font converter is now fully implemented on Facebook and Messenger. These tools will make a big difference for th friends and family. To continue supporting the people of Myanmar through this transition to Unicode, we are exploring as well as improving the quality of our automatic detection and conversion. We also intend to continue contributing to transition. Prev Networking @Scale 2019 recap Next MaRS: How Facebook keeps maps current and accurate Read More in Android MAY 9, 2022 MAR 8, 2022 Language packs: Meta’s mobile localization solution An open source compositional deadloc https://engineering.fb.com/2019/09/26/android/unicode-font-converter/ 4/7