natural-language parsing
The same grammar-and-parsing ideas built for programming languages also get aimed at human sentences. Natural-language parsing means taking a sentence of English, Chinese, or any human language and recovering its grammatical structure: which words form the subject, where the verb phrase is, how clauses nest. It is the school exercise of diagramming a sentence, done by machine, and it sits at the heart of how computers analyze and translate text.
Two practical differences from programming languages dominate. First, human grammar is wildly ambiguous: 'I saw the man with the telescope' has two perfectly good structures (who has the telescope?), and a single sentence can have dozens of valid parse trees. So natural-language parsers cannot use the strict LL or LR parsers that assume a unique structure; they reach for general context-free parsers like CYK and Earley, which return all parses, and they add probabilities to rank them, a probabilistic context-free grammar assigns a likelihood to each rule so the parser can prefer the most probable tree. Second, real language is messy and not fully context-free, so engineering uses richer grammar formalisms and, increasingly, statistical and neural models trained on large treebanks of hand-parsed sentences.
Natural-language parsing matters as the clearest sign that automata and grammar theory reach far beyond compilers. The same formal machinery, derivations, parse trees, the membership-versus-structure distinction, the cubic-time general parsers, underpins machine translation, question answering, grammar checking, and information extraction. The honest caveat is scope: classical context-free parsing captures much of sentence structure but not meaning, and human language stretches beyond pure context-free grammars, which is why modern systems blend formal grammars with heavy statistics rather than relying on grammar alone.
The sentence 'I saw the man with the telescope' has two valid parse trees: one where 'with the telescope' modifies 'saw' (you used the telescope) and one where it modifies 'the man' (he had the telescope). A probabilistic parser returns both but ranks the more likely one first.
Human sentences are deeply ambiguous, so NL parsing uses general parsers (CYK/Earley) plus probabilities.
Human language is not purely context-free, and parsing structure is not the same as understanding meaning. Classical CFG parsing is one ingredient; modern natural-language systems lean heavily on statistics and learned models on top of, or instead of, hand-written grammars.