I feel it worth mentioning however that Google translte is a “statistical machine translation” and continues to learn.

A big super computer is fed thousands of pages of translated parallel text and told to go figure. This explains why initially it would have had more difficulties with colloquial terms, as in the beginning the text it was fed came from UN documents.

It’s still in Beta but is now open to wiki style alterations. people suggest their own translations, it looks at them, lays down some statistical analysis and then makes a decision.

Interestingly a number of the words in your list which it was getting wrong back in October it now seems to have learnt. :-)