So your data structure must store pairs of words (in 2 different languages), and you will use words from one language as lookup keys?
Do you need to do more than a million lookups a second? If not, don't worry about what structure is fastest. Just use whatever standard hash table structure is available in your language.
If you're not convinced: I just did a little experiment in Ruby, using a hash table with 1,000,000 entries. On my PC, I can do 1000 lookups in that table in about 1 millisecond.
IF YOU DO need more than 1,000,000 lookups a second: try a search trie.
And remember that since your structure is completely static and unchanging, this task is 100% parallelizable across multiple CPU cores.