Automated Whole Sentence Grammar Correction Using a Noisy Channel Model

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 934–944, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Automated Whole Sentence Grammar Correction Using a Noisy Channel Model Y. Albert Park Department of Computer Science and Engineering 9500 Gilman Drive La Jolla, CA 92037-404, USA yapark@ucsd.edu Roger Levy Department of Linguistics 9500 Gilman Drive La Jolla, CA 92037-108, USA rlevy@ucsd.edu Abstract Automated grammar correction techniques have seen improvement over the years, but there is still much room for increased perfor- mance. Current correction techniques mainly focus on identifying and correcting a specific type of error, such as verb form misuse or preposition misuse, which restricts the correc- tions to a limited scope. We introduce a novel technique, based on a noisy channel model, which can utilize the whole sentence context to determine proper corrections. We show how to use the EM algorithm to learn the pa- rameters of the noise model, using only a data set of erroneous sentences, given the proper language model. This frees us from the bur- den of acquiring a large corpora of corrected sentences. We also present a cheap and effi- cient way to provide automated evaluation re- sults for grammar corrections by using BLEU and METEOR, in contrast to the commonly used manual evaluations. 1 Introduction The process of editing written text is performed by humans on a daily basis. Humans work by first identifying the writer’s intent, and then transform- ing the text so that it is coherent and error free. They can read text with several spelling errors and gram- matical errors and still easily identify what the au- thor originally meant to write. Unfortunately, cur- rent computer systems are still far from such ca- pabilities when it comes to the task of recogniz- ing incorrect text input. Various approaches have been taken, but to date it seems that even many spell checkers such as Aspell do not take context into consideration, which prevents them from find- ing misspellings which have the same form as valid words. Also, current grammar correction systems are mostly rule-based, searching the text for de- fined types of rule violations in the English gram- mar. While this approach has had some success in finding various grammatical errors, it is confined to specifically defined errors. In this paper, we approach this problem by mod- eling various types of human errors using a noisy channel model (Shannon, 1948). Correct sentences are produced by a predefined generative proba- bilistic model, and lesioned by the noise model. We learn the noise model parameters using an expectation-maximization (EM) approach (Demp- ster et al., 1977; Wu, 1983). Our model allows us to deduce the original intended sentence by looking for the the highest probability parses over the entire sentence, which leads to automated whole sentence spelling and grammar correction based on contex- tual information. In Section 2, we discuss previous work, followed by an explanation of our model and its implementa- tion in Sections 3 and 4. …

Content truncated to 500 words