Fixes for model creation

[texttheater]

I got the first part (model creation) to work for training data (AUTO files) that contain words with non-ASCII characters. I only had to change a few little things, hereby humbly submitted for consideration. ^^

Now to try the actual training…