Smoothing: add-one and add-k for n-gram language models

Why one unseen word pair gives a sentence zero probability, and how add-one (Laplace) and add-k smoothing and the UNK token fix it.

Part of Predicting the next word.

Read next