In 2115 we might be able to technically do a very good approximation of it, but it might not have much use except for very casual contexts.
I think it really comes down to controlling one's communication. For instance, right now businessmen learning Mandarin are not cheaping on translator services, they want a better understanding of someone's culture, the concepts they think in, and want to speak in their own words.
Communication is and always will be extremely important, and as soon as there will be enough stakes, the time we 'lose' in learning will always be worth the gain in control.
Of course a universal translator would help for tourism and casual matters. But I think for translating a menu, street signs or simple instructions we are already having pretty decent applications and won't need to wait for another century.
http://www.foresightexchange.com/fx-bin/Claim?claim=Tran
Even up through 2008, most people thought it was going to happen by 2015! Of course the people who were more actively following AI knew that it wouldn't. I probably doubled my fake money on FX, if only I could remember my login.
That isn't to say I agree with the sentiment that efficient and accurate translation will make English obsolete. Pretty much every major programming language, along with things like URLs and markup, requires using ISO Latin characters found on a QWERTY keyboard, which greatly limits the power of a major class of contenders. English just happened to be the lingua franca at the time that cultural globalization allowed for a language to keep that status long past the nation that made it such stopped being a super power (which, of course, may or may not happen to the US in the next 100 years).
For any mildly complex concept, heuristics, context analysis, pattern matching and word databases won't be enough to match two exact sentences in different languages. Even a sentence like 'I felt sick watching him treating his partner this way' will need an extreme lot of intelligence to correctly parse and convey in a different language.
If in 100 years AI really reaches the holy grail and we have actual artifical intelligence, translation will be a solved problem. Now, will we be able to reach that goal in 100 years ? I hope, but that's not a given I think.
And here it is in 2014:
http://arstechnica.com/information-technology/2014/12/skype-...
And the comments include "Google has been doing this for ages".
Siri, Cortana, Google Now, all have speech recognition. Google and Microsoft have translation engines. Microsoft and WordLens (and Google?) have live image OCR and translation.
It's no Star Trek Universal Translator, but it's a thing and it exists.
Those are some lofty claims about Skype Translate, but it's not putting human translators out of business.
For example, the German 'ich mag sie' could mean 'I like her' or 'I like them' or 'I like you' in formal settings, while the English 'at least' could be translated to German as 'zumindest', or 'mindestens', or 'wenigstens', with different emotional significance.
I'm not sure you could really do adhoc translation at the language level with any real fluidity: you would probably need your phone to monitor your brain before you spoke to get at the pre-language meaning of what you are saying, if that is even possible.
No doubt it's possible to construct examples that support your agument in just about any language, though. I'd recommend East Asian ones, since they're less heavy on the pronouns than Indo-European languages, and you can generally often omit subject or object if it's clear from context, or are even expected to for good style. Korean poetry even relies on intentional ambiguity a lot of the time, forcing the reader to consider the options, for their enjoyment or to challenge them.
Machine translation is of course aware of this problem, and doing the (rough) equivalent of "whole program analysis" is an active concern. I.e. you can try to source info elsewhere in a longer input to decide between multiple possible translations for another bit. The problem is really these short inputs where no context is easily available to the program - it's an input problem in some sense; it's just much easier for a human translator to guess from cues the program simply has no input channels for. One of those cues is who exactly they've been hired by, though, and that can be replicated with user profiles, too.
Especially since this is something where it's not possible for machines to truly just eclipse humans -- if you're fluent, you're fluent. Of course machine translation easily surpasses human speakers on vocab, but you're more or less bounded by what the recipient/listener can understand, anyway.
But demand for translation services will likely dwindle.
That said, one of the best reasons to learn languages is simply the mental exercise, and unless we change the way humans learn and maintain mental performance, that will remain true.
Why learn Chinese or Hindi if your smartphone or perhaps your glasses will seamlessly translate back and forth between your native language? The pressure to learn a second language will be lessened greatly.