With the advent of the Watson jeopardy system, public awareness of computer comprehension of natural language is at it's highest. It is common knowledge that computers of old have been comically bad at translating language, the meat is spoiled but the vodka is good comes to mind, but how do they fare now?


It is hard to believe that computers are already better than humans at translating than the interpreters at the UN, but that isn't a level very many people get to anyway. I propose a study/challenge to the language departments of our campus. Pit a range of our students against machine translation software and see how long does it take to exceed the accuracy of current machine translators.


The language departments could select or ask students over a range of skill levels, people aiming for the major at 1st, 2nd, 3rd 4th and perhaps even graduate students. We could then compare their ability with machine translations of a variety of free software options. Google comes to mind.


Judges should be 3, 5 or 7 in number to facilitate of split agreements. people capable of evaluating both english and the translated language would be optimal choice for judging.


Perhaps several sentence inputs would be offered, english being composed by perhaps rhetoric or english professors, and foreign languages produced by their respective departments.


judging could be as simple as taking input A seeing whether output B is equivalent or not. Or it could be more complicated, involving a gradient of errors or inadequacies in translation. Inputs should range in grammatical complexity so that presumably later year students will have more successfully translated sentences than the lower year students, presumably the computers would fall somewhere on this gradient.


It would be important for parties outside of the language departments to participate to verify that the grading criterion are impartial enough to get a legitimate evaluation and that they are carried out.