Computational linguistics

A word sequence never occurs in the training data for an unsmoothed n-gram model. What is the usual consequence for that sequence's estimated probability?