Will Google Translate Make Language Learning Obsolete?
Google Translate’s massive improvements recently and its foray into machine learning has become a major topic. Automatic translations are becoming more accurate, and getting closer to what a human can do. For a language learner, reading an article like this can leave you feeling down. You’ve been spending months or years mastering a language, and in the next several years, that may have all been for nothing, right?
The first reaction people have to this type of story of translations nearing perfection are the counter-examples. Shortly after Google switched to its neural network for translation, a fully Google-translated version of Final Fantasy 4 came out from the Legends of Localization site. I grew up playing this game, and this is a thing of beauty.
A few of my favorite Google translation highlights:
The creator of the site points out that he had started the Google translation project before Google switched its system to the machine learning one. So he switched to the new system as well. The results were not much better and were actually worse in quite a few ways.
Why the contradiction?
It’s easy to brush off machine translation, and find many counter-examples to show the opposite. But I remember the original Google translate. This is an entirely different world. For example, take one of the Japanese articles written on Jalup.
Sure, there is still plenty of strange stuff going on here. But it’s becoming more understandable.
Fear the AI Google Translation revolution?
Are we in the land of the universal translator yet? I wrote an article about this 3 years ago, and thought it would be wise to revisit the subject, since things change so quickly.
Even with how far Google Translate has come, it’s still not there just yet. One day it probably will be. There are a number of remaining hurdles, but one of the most major ones is localization.
Localization is the “adaptation of a product or service to meet the needs of a particular language, culture or desired population’s look-and-feel.”
Localization is not easy, and can often be as important as translation itself. It especially becomes problematic when you take different cultures like Japan and Western countries. I haven’t seen much discussed on how Google Translate will learn to localize, but it is an AI skill that must be developed.
Two extremes
Discussions on the subject take 2 sides. Websites like the NY Times magazine show how it is already so perfect that it can translate great works of literature with minimal mistakes. The opposite end of the spectrum are sites that show how laughably bad it still is. Because of these two discrepancies, it can be hard to figure out what is actually going on.
For example, is the translation industry doomed?
The fate of the translator
What happens to translation jobs?
The same thing that will happen to most jobs in the next decade or so as AI and technology improves. Things are going to change. It’s what happens in an advanced technologically growing society in the 21st century. You can’t just single out translators.
However, things are still early. Google Translate has and will continue to affect the translation industry in waves. The first, and current change is casual translation. You need an e-mail translated. Or a general website translation. Or some general research in a foreign language. Perfection is not required. Where perfection isn’t yet required, Google Translate shines. It will take some time to go beyond this, but don’t be surprised if it happens sooner than you expect.
Will Google Translate render language learning obsolete?
This is a completely different subject, and the firm answer is no. Translation and language ability are two entirely different beasts.
When I started learning Japanese, finding translations of a lot of Japanese popular culture (games, anime, manga, etc.) was hard. Now it’s become significantly easier. Yet people learning the Japanese language have increased from a decade ago. If the only reason/value of learning the language is to understand material that hasn’t been translated, this wouldn’t be the trend.
The experience of learning a language goes beyond just a direct exchange of meaning. It goes deep into understanding the culture and people of the language, and the country it comes from. Google Translate, or even a magical universal translator does not change this. The power that comes from truly understanding Japan and its culture isn’t to be underestimated, and is now more important than ever. People are still more important than machines.
The Evolution of Google Translate
If you continue to follow Google Translate’s path from nonsense, to some sense, to perfect sense, don’t turn it into a question of whether you studying Japanese, or any language, is a waste of time. Even if you are a translator, or have aspirations to be a translator, you are safe for now (but not forever). The value of language is still unparalleled. So get back to studying, and enjoy the fruits just like everyone else.
Founder of Jalup. iOS Software Engineer. Former attorney, translator, and interpreter. Still watching 月曜から夜ふかし weekly since 2013.
Honestly I think people drastically overestimate what AI will be capable of in the near future. It’s kind of like how we were all supposed to have flying cards and hoverboards 20 years ago. There are some very practical reasons why machine translation will not make it as far as people think within our lifetimes, let alone the next few decades.
1: Being able to determine implied meaning from unwritten context is an incredibly difficult challenge. All of the world’s greatest AI up to this point have only been able to work their magic in a “perfect information” scenario. Anything that requires judgment calls to be made based on imperfect information will be orders of magnitude more difficult for AI to handle in a reliable way.
2: Translation isn’t just about understanding the source and re-wording it in a target language. Many creative and stylistic decisions need to be made as part of the process. Jokes and wordplay in the source material need to be handled on a case-by-case basis – is a play on “saboru” and “sabotage” something that can be preserved in the English text? Is there a different joke you could make there? Do you just cut it and accept the loss? If the text relies heavily on cultural references your target audience is unlikely to understand, do you change those references? Or do you add footnotes somewhere to explain? The list goes on and on. At the very least when it comes to creative projects, professional translators won’t be made obsolete by machines any sooner than writers, directors, artists, architects, engineers, etc.
3: Even the best translation (human or machine) is no substitute for time devoted to cultural understanding. Language and culture are inseparable, and truly understanding one requires that you understand the other. This sort of understanding will still play a central role in intercultural human relationships no matter how good machines become at bridging the gap between languages.
As I’ve talked about these exact points in the Universal Translator post, I agree. There is a long way to go (and what is ultimately possible is still questionable). Any current influence it has on the translation industry now are things that people probably wouldn’t have even used a translator in the first place.
Your points are pretty much right, but I doubt we’re overestimating AI (or maybe the public is, much like the hoverboard stuff), any reasonable AI scientist would agree with you honestly. Natural Language Processing is one of the apexes of AI and it’s going to take much more than AI to make the perfect translator (we don’t understand the brain well enough yet to process language naturally for instance, and I doubt we will even in 80 years, and that’s only one of the foundations of translation).
The way I see it is that in 10 years we’ll have something that can reasonably get the main points across in every subject but just barring the nuances and localization.
In any instances, even in a thousand years, I predict learning languages will be useful anyways, even for “basic” skills like writing or reading for instance (it’s no coincidence that most of the literacy geniuses all around the world like Soseki, Hugo, Beaudelaire and others are extremely proeficient in one or more foreign languages).
It is possible that an AI translation gets close to perfect translation, but only if the writter writes in perfect language, or speaks in perfect language.
There are many slangs, abbreviations, ways of saying things; and if it is spoken language, there is also the accent, which can be tricky for AI to identify.
I believe we are very close to perfect translation on a perfect environment, but not on every aspect of the language.
This especially becomes problematic when a writer purposely uses incorrect language, or language he made up, or he just has typos. Can you/should automatically correctly translate what is incorrect.
You make a good point about the perfect environment translation coming sooner. But that perfect environment is not that common.
Translation jobs aside, theoretical “perfect” computer translations don’t affect my own motivation for studying Japanese — the whole point for me is to not need translations! Most of the material I want to read/watch/&c. is already translated anyway, and in the case of novels, is translated really damn well by some really talented people. It’s still not the same as reading the author’s own words though, no matter how eloquent or accurate it may be.
As far as human translation goes, I think Adam and Matt are dead-on: there is so much nuance and context that goes into a piece of writing, especially if it’s a work of art, that it will be tough for AI to match any time soon.
I guess in the end it just depends on what the goal of the text is. If its goal is to teach me how to operate my radiator, nuance isn’t as important as if it’s attempting to convey some subtle emotion or element of characterization.
Yes, it shouldn’t affect your desire to learn the language at all. That’s the important thing to remember!
It really depends on how far computers can go. I get the feeling we are reaching the limits, but people felt the same way 100 years ago so who knows. If computers could ever be like human brains with all the connections and emotions then yes.
Context also plays a huge role in translation. Unless the computer could pick up all the context itself (another huge challenge) it would have to be manually put in along with the text. i.e. “This person is a 20-year-old male bunny living in a fantasy world who is normally grumpy but is currently trying to seduce this girl for his nefarious schemes.” Any one of those details could change the translation.