I have already done this, but it is difficult to squeeze the running algorithm memory. The current POS solution uses only 400MB memory. Also, 2grams are not that good as collocations.
Microsoft have Language Portal with a lot of technical terms translations in a multiple languages. It gives you the translations in different context too.
1) Missing square element in real footage.
2) What bothers me more is that shadow properties are not consistent. In CGI it is more diffuse and changes properties like if the Sun is positioned 10m above.
I have implemented TST (ternary search tree) in C to create fast dictionary (histogram) for counting words in Wikipedia. It is fast and memory efficient.
Another one www.google.com/ncr - sets default Google search domain from google.cc to google.com, where CC is country tld.
Useful if you prefer to search in .com version and not the localized one.