Language profile training set size

[ashkarr]

Hi,
I am planing to change Turkish profile but i am not sure how many training set should i use ? (I do not want to exceed the average of the size.) But as you know the trainig set size is realy important in ML. And would you mind if I ask you to share your trainig set stats with us? For example how many keyword did you use to create profiles per Language ? I have more than 10 TB Turkish corpus if I use all of it I think every latin texts will be set as Turkish by detector :)