Welcome to day four of The Nightmarathon Before Christmas. A couple of quick things before we get started: I just wanted to let you guys know that I created all of the videos this week completely out of order, and today’s upload was originally going to be live-action ghost-hunting. However, some things got in the way, and unfortunately, it fell through. Tomorrow’s upload was made incredibly early, and at the end of it, you’ll hear me mention how we “went ghost-hunting”. Obviously, this videos in its place, and we did not go ghost-hunting, so ignore that. Also, I’m sick, so if I sound weird, I apologize. But nonetheless, I’ve stumbled upon something more interesting than ghost-hunting would have ever been, and I’m excited to jump into today’s topic, which is the Translategate Mystery. Translategate is the name of a strange revelation found by users on 4chan, regarding hidden messages found within Google Translate. I had originally stumbled upon this strange phenomenon through a video created by a Youtuber named Elder’s Vault. I reached out to him, and… I’ll just let him fill you in with some background real fast. Elder’s Vault: Hey, what’s going on you crazy kitty cats? It’s Elder here, and let me just first say thanks so much for letting me come on and talk about some of the stuff that we found. This all started with the conspiracy on the paranormal board of 4chan, and really, it’s quite simple. You type broken letters, or words, in this Somali Google Translate, you get some very strange fragmented responses. Me and a good friend of mine, Deathly Logic, were interested in piecing together text that was coming out of the output of the Google Translate app, and we quickly realized this spread further than one isolated text. We started piecing together names, private email conversations, dating profiles, and fragmented Facebook links from all sorts of context, and sometimes completely different languages, such as Russian. When you add this with the fact that users were questioning Google Translate, while using fragmented input questions, we knew after receiving eerily correct responses. This is not only a potential breach of privacy, but something that might be a larger issue than we originally thought. Nightmare Expo: After playing with it, myself and various others who have tried this have gained some interesting results. As Elder said, there were strings of text resembling what appears to be private email conversations, dating profiles, and even Facebook links. However, there’s also what seems like a strange and ominous way to communicate with an artificial intelligence. It appears that if you put anything, ANYTHING in the translation field while it’s broken up into 1-3 letter chunks, you’ll get what appears to be a logical response. For instance, I asked the translator: “Is th is a ba ck do or to ta lk to you?”, and it responded with: “Is this a good idea to do you?” Now, the word “do” is thrown in there kind of randomly, but you get the overall message. One 4chan user simply typed “google” a bunch of times, broken up into parts, of course, to which it responded with: “Go and find a place like this. Find out and know what’s going on.” This phenomenon seems to be inconsistent as well, with multiple “responses” coming from the same translation. There were tons of other questions that were asked as well throughout this thread, and I’ll leave a link below for you to go and check some of them out. Now, with all of this, what in the hell could this mean? How would messages, emails, full names, and entire links come about from randomized input? After doing some digging on Google Translate, I came to find that they use “adaptive machine learning” for their translations, and the bot is constantly being improved through neural networks. This ties into this first theory that was made, in which I will call “The Data Theory”. It seems that possibly, possibly, the system was picking up text from various websites and Google services. One notable site that Elder pointed out in his video was a comment hosting service called DISQUS. He claimed that possibly, Google Translate was pulling strings of text from the comments that may have been made over time, since Google has such an enormous log of information on users, through decades of data collection. With this comes a sense of personal exposure, since millions of comments have been posted through the service, to date, that may also include private or sensitive information. It really doesn’t stop there, though. Google has its own version of nearly everything under the sun, from word processing, to cloud storage, to email services, to video hosting, you name it. This information could be coming from any of those, and it’s a serious problem. Another hypothesis that users have made is a bit of a stretch, but it’s what I’d like to call “The Predefined Theory”. The basis behind this assumption surrounds the idea that a programmer has physically hidden codes within the translations, themselves, since, as Elder said, it only works in Somali. Somali is only spoken by 0.0019% of the world population, and it’s believed by some that this was their motive for hiding codes within the language’s translations. Now, with this comes two important things that I want to point out. If you look at each photograph of a translation, you’ll notice one strikingly obvious detail. These translations aren’t in Somali at all. In fact, that’s only the setting that each user had manually set it to. Rather, Google automatically assumes the language of the text they are inputting, and makes a translation using its own preset, since what users are typing in is, well… it’s nonsense. So, for someone to theorize that somebody is secretly conspiring with Google to translate some hidden codes is unreasonable, and this actually ties into my next point. The bulk of the responses received by the translator appeared to be sensible strings of text, however, this seems extremely related to what ScareTheater had called “The Reverse Speech Theory.” Essentially, he claimed that when you listen to audio backwards, you make out what you want to hear. Someone else could easily listen to the same thing and get a different result, since our perceptions are relative. In this case, we are essentially fishing for creepy translations from the service, since that’s primarily what we want to see. There were plenty of things that I typed in that yielded no results at all, and a lot of the translations that I did get were nonsensical, at best. I believe that people are hunting for weirdness with this thing, and when they get it, they freak out, and share it. Now for the elephant in the room: The Facebook links, profiles, and emails. This largely remains a mystery to me. Like we said before, Google appears to be pulling texts from the comment service DISQUS. However, it honestly could be pulling this information from anywhere on its platform. Google is such a large company that has more data than you could ever imagine, and a bug like this could have some dire circumstances, since, for all we know, they actually could be pulling information from emails linked to the Gmail service. Links and private information have absolutely no place in a translation app, and while you could discount this as happenstance, these appear far too often in our tests to be a coincidence. It could be a huge privacy glitch, or even an error that Google has overlooked. I’m really not sure what to make of this, and to me, it honestly remains a mystery. I encourage you all to go over and let me know what creepy or weird translations you guys might find. Feel free to post your findings and screenshots on my subreddit. I’d love to see what you dig up. Also, if you find any more information about this translation bug, do let me know. I want to get to the bottom of this, and if it is a privacy issue, I want to have this resolved. Anyway, I want to thank Elder’s Vault for joining me in day four of The Nightmarathon Before Christmas. Tomorrow night, we’ll wrap up this awesome week at the same time, same place. You’ll know what to do. Lastly, thanks so much for watching. I love you all, and Good Night.