- Some code optimization, trying to analyze words first with the assumption that no unknown morphemes are involved, if the PoS is already determined
- Adding unseen possibilities to cached observed values for words that are seen less than 10 times in training data
- Slight change in format of training data
- Corrections in German training data.