Using a database for 20 terms or phrases sounds a bit like overkill. A flat file containing the data to be loaded into a hash of hashes should be sufficient and efficient, IMHO.

Depending on how you're gonna use the data structure afterwards, you could have either a hash of languages, with sub-hashes of terms to be translated and their translations, or a hash of terms to be translated, with sub-hashes of languages with the translated terms.