Chahuneau et al 2013 - Knowledge-Rich Morphological Priors for Bayesian Language Models Combines a finite-state guesser (that was constructed in 3 hours) with Bayesian non-parametrics to learn the correct morphological analysis. Limitation: assumes each word has one best morphological analysis, taken out of context. This could be corrected with contextualized model, like a sequence model for the part-of-speech tags