Homework #4 - testing Machine Translation

1. Find a webpage in a foreign language, preferably about some topic you know something about.
   • Which language? The language has to be one of those supported by Google translation system (or some other translation system of your choice). If you know a foreign language and it is supported pick that one.
   • How to find such a web page? You can go to wikipedia and look through English articles - if there is a version of that article in another language, there is a link to it in the left bar. Or you can use google language tools to search through pages written in a particular language. If the language uses Roman alphabet, you can search for webpages about some person.
2. Go to language tools (or some other translation system of your choice) and ask it to translate the web page. It is better to give it the address (the URL) of the web page than the text directly. If your computer is not properly configured, the latter can copy letters with diacritics incorrectly.

1. Look at the result. Is it perfect? Do you understand it? If it is not perfect, can the system still be helpfull for some tasks?
2. Find 5 errors in the translation and explain what is going on. You have some knowledge about linguistics, so do not say "It is just wrong". Try to suggest whether the problem is in syntax (e.g. the system probably considered a wrong phrase to be the subject), morphology, etc. Try to select 5 errors that are not all of the same type.

   For example, German distinguishes grammatical gender (Tisch 'table' is he, Heft 'notebook' is it, Wohnung 'flat' is she, etc.), so when you see that English translation uses incorrect pronouns to refer to things, it is probably consequence of this. Maybe pronouns are correct if the thing they refer to is in the same sentence and often incorrect when the thing is in another sentence - If that happened you could guess the system translates each sentence separately without considering its context. etc.