Parsing

• My doctoral thesis was on natural language parsing.
• With parsing, you take the words of a sentence and build a syntactic tree.
• What you're really interested in is the semantic output of a sentence, typically in a case frame.
• The dog chased the boy to the store. -> (S(NP the dog) (VP Chased (NP the boy) (PP to the store))) -> (chased (actor the dog) (object the boy) (location (to the store)))
• There are a lot of techniques to do this.
• People parse left to right with little or no backtracking, and my Plink parser worked that way.
• Collins parser is currently the best (with around 90% of branches correct a syntactic measurement). It is a probabalistic parser.
• How is this measurement done?