Can we trust machine translations?
Machine translations (MT) can be relatively useful if the result is in your own language, and you can fix the mistakes, but otherwise the sentences can be unacceptable, or unintelligible. Do you have any examples of very bad MT? Let me give you a few:
Original (O): Tengo que ponerme a estudiar.
Machine Translation (MT): I have to get to school.
Human Translation (HT): I have to start studying.
O: Me molesta que grites.
MT: That bothers me scream.
HT: I hate it when you scream.
O: Escríbeselo a ver si se calla.
MT: To see if it is Escríbeselo silent.
HT: Write it down for him to see if he shuts up.
O: ¿A que no se lo has dicho?
MT: Did You are what you say?
HT: You haven't told her, right? / I bet you haven't told her.
6 Answers
In all fairness... It took 15-20 years to develop a chess-playing machine that could consistently beat the best human players and that 1) there are only 64 squares with 32 pieces on a chess board and the complete (and explicit) rules of chess can be written on a page or two. Typical human languages have vocabularies measuring in the hundreds of thousands of words. Grammar books (giving the "rules" of a language) run to hundreds of pages (and often do not agree with each other) and it is rare, indeed, to find the rule that does not have exceptions. If that were not enough, the vocabularies and grammar are not even permanent (but, rather, evolve over time).
(Aside for the computer science buffs): I suspect that the problem of a real "universal translator" is NP-complete, although, in the absence of an unambiguous definition of a language (any language) this would not be a provable assertion.
I'd like to believe that better artificial-intelligence programming could eventually solve some of the "choking" that occurs with MT's, but that's a discussion for a computer science forum. Even so, one of the intractable problems with MT's is that the slightest error / typo makes them go ka-punk, real fast. I put one of the infamous chat messages that are constantly being posted on this forum into the translator, and here is what it said.
Clearly, if you write but did not because you do not get my message're very special and clear that if I want to be with TIGO have December
How could a machine translation pick up on differences like
'a run on the banks'
'a run in her stocking'
'run for your life'
'run(') out of money'
Very tricky
One of the problems with computer translation is the fact that a language has an infinite number of possibilities in terms of sentence creation. Another is that words and clauses have subtle meaning changes depending on the context. That is very complex and difficult stuff for computer programs to solve. A chess game at least has a finite number of possible moves for each situation. That is easy for a computer to solve. The limiting factor was always the speed at which it could run through all of the possibilities to determine the best possible move.
The truth is, language translation has not evolved much since I studied computer science in college almost 30 years ago.
O: What in the world are you doing?
MT: ¿Qué en el mundo estás haciendo?
HT: ¿Qué demonios estás haciendo?
O: With fall come the delicious smells of pumpkin pie and apple cider.
MT: Con la caída llegado huele delicioso pastel de calabaza y de manzana de sidra.
HT: Con el otoño llegan los olores deliciosos de pastel de calabaza y de sidra de manzana.
I can't remember the exact translation, buy many years ago I saw a sentence that started with something like "Quito, capital de Ecuador," which was translated as "I take off the capital of Ecuador."
I'll try to think of some others.
And you should see what MT does to Japanese-English translation!
