Due: Monday April 14
The following problems pertain to interactions with deployed Machine Translation systems. Go to: http://translate.google.com/translate_t. Most, or all of these systems are at this point Google's own home-grown statistical machine translation systems.
You will see various language pairs such as English->Spanish, Spanish->English, English->German, German->French, etc.
The pairs allow you a few options to translate an English sentence through some other languages and then back into English. For example, you can translate a sentence from English to German, to French, then back to English.
Try the following experiment. Use the system to translate the sentence "I put the pig in the pen" into German. (If you know German, you might already see a problem.) Then take the German output and translate it into French. Then finally translate the resulting French into English. What do you get back? What went wrong? What if you try this in another order: English->French->German->English.
Try four additional sentences of your own choosing. You should pick things that have interesting ambiguities. How well does it work when you translate from English into another language and then back to English? Does it get worse if you get a third language in the loop, as in the above experiment?
Try to find as many problems as you can, either with a single translation step (if you know the other language well) or with a chain of translations such as English->B->C->English. Come up with some hypotheses as to what is causing the problems you note.