Final Project: Filtering Parses with Lexical Semantics

Modern context-free grammars, especially featured ones, can achieve reasonably high accuracy rates in parsing. However, because any CFG that can handle any substantial chunk of a language must contain a very large number of rules, some of which will match in unintended ways, parsers suffer from an overload of parses: there are so many possible parses that the parser must prune away unlikely parses as they are generated. Generally, this is done simply by training the grammar on a large set of sentences. However, this approach can be impractical even in the presence of a large corpus, such as Penn TreeBank, because the sentences in the corpus may be too specialized, and hence is unsuitable for parsing the types of sentences that are not common in the corpus. An alternative or supplemental method for pruning parses is using semantic information to rule out or decrease the probability of parses that are nonsensical. For example, the sentence