Type the word "dog" into Google Translate 19 times, request that the nonsensical message be flipped from Maori into English, and out pops what appears to be a garbled religious prophecy.
"Doomsday Clock is three minutes at twelve," it reads. "We are experiencing characters and a dramatic developments in the world, which indicate that we are increasingly approaching the end times and Jesus' return."
That's just one of many bizarre and sometimes ominous translations that users on Reddit and elsewhere have dredged up from Google Translate, Google's decade-old service that can now interpret messages in over 100 languages. In Somali, for instance, strings of the word "ag" translate into missives about the "sons of Gershon," the "name of the LORD," and references to Biblical terminology like "cubits" and Deuteronomy.
Previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously.
Sounds like the primary training set for these obscure languages was...bibles.
Several Christian sects think it's really, really important for everybody to read their Book. This means you need to go find out how all the languages work, learn how to translate the Book into each language and print up copies for distribution.
As a result while the legal disclaimers for the Apple iPhone may be translated into dozens of languages, the Bible has been translated into thousands of languages.
This was definitely not the stupidest waste of resources by a religious sect, even before American Christians starting buying their fake prophets jet aeroplanes.
Parebibolia - a psychological phenomenon in which the neural net sees bibles where there are none.
Worth noting this work also sometimes involved developing a writing system for said language (usually crammed into the Latin alphabet for good or ill).
Bible printing may well have been one primary goal of an astounding amount of translation, but it's not been the only goal. It's important to evaluate scholarly work in the context of it's times, and without primary consideration of their preferred Magic Sky Weasel.
There’s probably a proper term but I call this phenomenon “AI overreach” where the neural net seems to be configured to always return an answer, any answer even if it is bad. Seems like it would be better for the AI to track confidence of potential matches and if he best result had low confidence then simply return an “I don’t know”. Kind of like how good interviewers never ask yes/no questions like “do you know how to configure and manage DB replication ?“ because with some interviewees the answer is always “yes” independent of actual experience. You ask “how do you implement replication ?“ instead.
That said J Peterson nailed it. I wonder whether the native speakers receiving the biblical tinged translations think that the sender sounds retro like someone out of King James court.
Dr. Bronner lives!
I like the last example. Shame they didn't make "Choose the form of the destructor" the fixed point.
Ph'nglui mglw'nafh Cthulhu etc.
Sing it for me, Google.
My absolute favourite is this one https://www.youtube.com/watch?v=3-rfBsWmo0M mainly because they guy reading it kills it.