>>3028608Anyone have a better understanding of how the translation algorithm actually works? I was under the impression it's a neural net, and tries to find a "best fit" based on input. In fact, it's a known flaw of machine learning systems that nonsense input is still processed and results in a random answer based on how that input is classified within the "latent space." It's also possible this is an artifact of the translation system's role in steganography, i.e. encoding and decoding covert communication by hiding it in natural text and relying on the "imprecise" nature of translations as a cover for why certain words are interpreted differently.
