Will machine translation eclipse the need for human translation some day and if so, by what date do you predict? [closed]

Question

I recently listened to a TEDTalk given by Louis von Ahn, co-founder and CEO of Duolingo. During this talk, which he gave in 2011, he addresses the topic of machine translation a bit. Specificially, he says this:

Now, some of you may say, why can't we use computers to translate? Machine translation is starting to translate some sentences here and there. Why can't we use it to translate the web? The problem with that is it's not yet good enough and it probably won't be for the next 15 to 20 years. It makes a lot of mistakes. Even when it doesn't, since it makes so many mistakes, you don't know whether to trust it or not.

I've been thinking about his prediction ever since. Specifically, I've been wondering if his 15-20 year estimate is conservative, generous, or spot-on. For a variety of reasons, I decided to come up with my own predictions based on the prediction Mr. von Ahn has made. Those predictions are below:

You'll notice that the list is not all that inclusive, but my purpose is not to get a language by language prediction and I felt it was best to keep it to languages I was most familiar with (the one exception being the Asian languages).

Setting aside all reports of an asteroid hitting Earth in 2025 or reports that we will reach "singularity" by 2045, I am curious about the opinions of others who are more learned with the topic of linguistics, machine learning, and/or artificial intelligence with regard to the predictions I've made. In fact, I thought about posting this on the Artificial Intelligence StackExchange, but I sensed it would be seen by more people who have discussed this question by posting it here in the Linguistics StackExchange.

I'd also be interested in knowing how you come up with your predictions. Mine, as I said, were based off of Mr. von Ahn's predictions, but beyond that, I factored in the difficulty of the various languages based off of previously established models (such as the FSI's categories of language difficulty), my own personal experience with the language, and how well I've noticed tools such as Google Translate perform.

It would seem to me that this process of going from human translation to machine translation will be a gradual process and quite different from the transition that took place between use of the abacus and the hand-held calculator. However, the path technology has taken and the speed with which it has spread seems to indicate that it is inevitable that one day machine translation will eclipse human translation. Sadly, however, I've never heard anyone lament, But I was really good with the abacus! What am I going to do now that the calculator is here!! I say sadly, because learning a language may be the one true talent some of us have and it takes a lot of time and energy to learn a language. Despite that (or perhaps because of it) I wonder if it makes sense to continue such a pursuit. The quality of some machine translations I've seen is so good that I wonder if continuing to learn a language is akin to continuing to master your skills with the abacus when the calculator is about to hit the market. In other words, is it now a waste of time to bother learning a foreign language because a machine will soon be able to do it so much faster and so much better?

Thank you in advance for taking the time to consider and answer my question.

Answer 1

Already, Google Translate has reduced the need for knowledge of foreign languages and for translators. There is the possibility that human translators will become much less useful, especial for major languages, and combined with the prospects of radical reduction in the number of languages spoken, needs for human translation may shrivel significantly.

Actual need and theoretical need are different. The former is driven by social forces, and the latter is just an exercise in combinatorics. I don't think there is a Norwegian-Chukchi translator out there (human or machine), the reason being that there's no social need for such a thing (yet). There is a current and actual need for translation of Mushunguli, Af May and Af Mahaa (standard Somali). Case in point, the Missouri dept. of health has a pamphlet translates English

At home, school, work, and even in our cars, we need to know what to do in an emergency. We need to know where to go in an emergency.

into Somali as

Marka aynu joogno guriga, dugsiga, shaqada, iyo xiitaa baabuurka dhexdiisa, waxa loo baahan yahay inaynu naqaano wixii aynu sameyn lahayn marka wax lamo-filaan ahi dhacaan. Waxa loo baahan yahay inaynu naqaano meesha la tago marka wax degdeg inoogu dhacaan.

(not a speaker, but my limited knowledge of the language and use of a dictionary tells me that this is reasonable – if a Somali speaker can weigh in, that would be enlightening). Google translates the Somali text to English as

When we are at home, at school, at work, and even in the car, we need to know what we do not do anything - The dusk happens. We need to know where to go It's in no hurry happen.

There is no machine translation at all for Af Maay or Mushunguli, and hospitals completely have to rely on human translators (sometimes to the detriment of the patient).

A major contributing factor to inadequacies in machine translation is the lack of adequate study of one of the desired languages. Major languages with 25 million speakers or more tend to be somewhat better studied, compared to the thousands of virtually unknown languages with 1000 or fewer speakers, so under a "social need" approach to the question, there is little social need for Norwegian-Chukchi translation, and thus it may be irrelevant if we don't know the Chukchi verb for "distill", or how to put it in the 3rd person dual perfective (if indeed there is such a thing in the language).

The idea of covering "Asian languages" by 2045 is seriously mistaken. There is virtually no chance of having machine translation for Lahu, Hawrami, Laz, Woi, Lamet and Ket by 2045, or 2145 (whether or not there will be speakers of the language beyond that point is hard to say). However, it may be that by that time, translation of Japanese, Korean and Mandarin will progress to the point that machine translation is useful. (In contrast, current Finnish and Turkish translations are not very useful).

Machine translation comes in a number of forms, one being text-to-text mapping, and the other also encompassing spoken language. Speech-based translations will be absolutely required for the many languages that have no written form. Real-time computational translation of speech is even more pie in the sky, although I am surprised at how good it has gotten for some languages. There are notable exceptions, but that language only have some 300,000 speakers.