This is a text-based language identification demo. It's very simple and not very robust because of the small training set and hackish handling of foreign characters, like the é in resumé. Also, more text will yield better (but slower) results. It's based on a hidden Markov model using character-level observations. The model has two states so I personally interpret the model as distinguishing statistical differences in vowel/consonant patterns across languages. The full version including training code can be found on my github.