"Smoothing techniques for N-gram models The term smoothing refer to such modi fi cations in the MLE estimates of N - gram probabilities that are addressed to move some probability mass from higher counts to zero-counts, making the overall distribution less jagged (Figure 2.5)."