Automatic Grammar Correction: Using PCFGs and Whole Sentence Context.

摘要

We explore the problem of automatic grammar correction and extend the work of [Park and Levy, 2011]. We use a noisy channel model that uses whole sentence context to generate a grammatically correct sentence with the highest probability. Our major contribution is to explore the idea of using a better language model than n-gram to represent the rules of the English language. We use Probabilistic Context Free Grammar (PCFG) and explain how we can combine it with noise models that are represented with Weighted Finite State Transducers (wFST) to build our noisy channel model. We also extend V-expectation semirings [Eisner, 2002] to CKY parsing, a popular parsing algorithm for parsing a sentence of a language.