You need one stem dictionary for every language, but since the only language I actually tried is English, I'd recommend to start with just one. I also recommend stemming only databases containing documents in only one language - support for stemming in multiple languages is possible, but nontrivial and definitely unimplemented.