But takoyaki does mean “octopus dumpling,” almost literally, except in Japanese. In practice, it is a loanword, that points to a specific form of octopus dumpling in a particular cultural context. This is a common problem in translation, even before computers. People who know the “languages” but not the broader context make this sort of mistake a lot in translation. The real question is whether Google algorithms can be taught to understand “nyu-an-su” in Japanese and how it is different from “nuance” in English (to be fair, I don’t exactly know the difference, other than that they are). I do think that this is actually the easier challenge that can, in due time, be addressed through technological means.
PS. If I know nothing of Japan, “octopus dumpling” makes heck of a lot more sense than takoyaki. In my past experience in translation, I’ve often found it easier to go for a simple, even if not quite so nuanced translation than try to explain the “nyu-an-su” of the context which gets too convoluted. We might find it desirable that we can better appreciate the context, but sometimes, it is better to just let a dumpling be a dumpling, rather than the dumpling and try to explain “the.” It’s a problem with the audiences you are dealing with: if you know the magic behind the words, the words are magical. If you don’t, then they are just words.