Machine translation is awful and barring some fairly world changing things happening, things we may live to see but nobody has a timeline for, always will be
Still... for various disabled or hard of sight people we have had text engines output their text, most games using encoded text rather than graphical means that is quite doable for a lot of games, by various means, typically lua on any games that are emulated or on the PC which you can do that for anyway. I have spoken to a few let's play types that wanted to do auto captions/subs too and were also looking to do that. The 3ds does have the means to debug games on the fly and thus you could get the text out, and the 3ds does also have the ability to soft replace files (or indeed the debug stuff twiddle data on the fly). The timing could be an issue but probably solved with a forced constant decode and some bounds checking to make it stay on the screen, not necessarily trivial but done before and well within the realm of reason.
So yes the elements are all there and have been used for years to do things like that. The result however would not be great so nobody has put it all together that way. Looking at the video above it also throws in some optical character recognition (OCR) and I guess you could take that route rather than straight text output if you wanted another layer of complexity.