Why can regularization reduce overfitting?

It discourages the model to be complex by penalizing the weight coefficients to be zero.

Regularization is a broader term that encompasses several techniques including L1 and L2 regularization, noise injection, ensembling, early stopping, and data augmentation.