disambiguate doesn't handle words not in vocabulary

[glicerico]

To disambiguate(), we currently need a context that is composed of words which exist in the AdaGram model. When this context contains a word that's not in the model vocabulary, disambiguate() returns a KeyError.
Would it be convenient to discard the unknown words and disambiguate with the rest of the context? In the extreme case that no words in the context exist in the model, then we'd be using an empty list, which still returns the prior probabilities of the word...
This may be done with a warning to announce which word(s) from the context are being ignored