Computational linguistics

Why does increasing the order of a count-based n-gram model often make reliable estimates harder to obtain from a fixed corpus?