Generating Confusion Sets for Context-Sensitive Error Correction

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 961–970, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Generating Confusion Sets for Context-Sensitive Error Correction Alla Rozovskaya and Dan Roth University of Illinois at Urbana-Champaign Urbana, IL 61801 {rozovska,danr}@illinois.edu Abstract In this paper, we consider the problem of gen- erating candidate corrections for the task of correcting errors in text. We focus on the task of correcting errors in preposition usage made by non-native English speakers, using discriminative classifiers. The standard ap- proach to the problem assumes that the set of candidate corrections for a preposition con- sists of all preposition choices participating in the task. We determine likely preposition confusions using an annotated corpus of non- native text and use this knowledge to produce smaller sets of candidates. We propose several methods of restricting candidate sets. These methods exclude candi- date prepositions that are not observed as valid corrections in the annotated corpus and take into account the likelihood of each preposi- tion confusion in the non-native text. We find that restricting candidates to those that are ob- served in the non-native data improves both the precision and the recall compared to the approach that views all prepositions as pos- sible candidates. Furthermore, the approach that takes into account the likelihood of each preposition confusion is shown to be the most effective. 1 Introduction We address the problem of generating candidate cor- rections for the task of correcting context-dependent mistakes in text, mistakes that involve confusing valid words in a language. A well-studied instance of this problem – context-sensitive spelling errors – has received a lot of attention in natural language research (Golding and Roth, 1999; Carlson et al., 2001; Carlson and Fette, 2007; Banko and Brill, 2001). The context-sensitive spelling correction task addresses the problem of correcting spelling mis- takes that result in legitimate words, such as confus- ing their and there or your and you’re. In this task, a candidate set or a confusion set is defined that spec- ifies a list of confusable words, e.g., {their, there} or {cite, site, sight}. Each occurrence of a confus- able word in text is represented as a vector of fea- tures derived from a small context window around the target. A classifier is trained on text assumed to be error-free, replacing each target word occur- rence (e.g. their) with a confusion set consisting of {their, there}, thus generating both positive and neg- ative examples, respectively, from the same context. Given a text to correct, for each word in text that be- longs to the confusion set the classifier predicts the most likely candidate in the confusion set. More recently, work in error correction has taken an interesting turn and focused on correcting errors made by English as a Second Language (ESL) learn- ers, with a special interest given to errors in article and preposition usage. These mistakes are some of the most common mistakes for non-native English speakers of all proficiency levels …

Content truncated to 500 words