[feminintendo]

Google "natural language processing."

Natural languages are highly ambiguous. Using a naive approach like Early's algorithm can lead to thousands of parse trees for a single sentence. Probabilistic grammar parsing algorithms attempt to discover the "most likely" parse tree with respect to a particular statistical model of the language grammar. As with parsing algorithms for formal languages, different algorithms make different trade-offs. "Best" only makes sense with respect to a particular metric: "Best" in what sense?

I personally don't know much about this area beyond what I just said.