[ppod]

A regularized prior would help.

[wooorm]

I’m also really interested in trying something like this: http://www.slideshare.net/shuyo/short-text-language-detectio... (slide 6). But I’d need a lot of training data, more than UDHR.