Correcting ESL Errors Using Phrasal SMT Techniques

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 249–256, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Correcting ESL Errors Using Phrasal SMT Techniques Chris Brockett, William B. Dolan, and Michael Gamon Natural Language Processing Group Microsoft Research One Microsoft Way, Redmond, WA 98005, USA {chrisbkt,billdol,mgamon}@microsoft.com Abstract This paper presents a pilot study of the use of phrasal Statistical Machine Trans- lation (SMT) techniques to identify and correct writing errors made by learners of English as a Second Language (ESL). Using examples of mass noun errors found in the Chinese Learner Error Cor- pus (CLEC) to guide creation of an engi- neered training set, we show that applica- tion of the SMT paradigm can capture er- rors not well addressed by widely-used proofing tools designed for native speak- ers. Our system was able to correct 61.81% of mistakes in a set of naturally- occurring examples of mass noun errors found on the World Wide Web, suggest- ing that efforts to collect alignable cor- pora of pre- and post-editing ESL writing samples offer can enable the develop- ment of SMT-based writing assistance tools capable of repairing many of the complex syntactic and lexical problems found in the writing of ESL learners. 1 Introduction Every day, in schools, universities and busi- nesses around the world, in email and on blogs and websites, people create texts in languages that are not their own, most notably English. Yet, for writers of English as a Second Language (ESL), useful editorial assistance geared to their needs is surprisingly hard to come by. Grammar checkers such as that provided in Microsoft Word have been designed primarily with native speakers in mind. Moreover, despite growing demand for ESL proofing tools, there has been remarkably little progress in this area over the last decade. Research into computer feedback for ESL writers remains largely focused on small- scale pedagogical systems implemented within the framework of CALL (Computer Aided Lan- guage Learning) (Reuer 2003; Vanderventer Faltin, 2003), while commercial ESL grammar checkers remain brittle and difficult to customize to meet the needs of ESL writers of different first-language (L1) backgrounds and skill levels. Some researchers have begun to apply statis- tical techniques to identify learner errors in the context of essay evaluation (Chodorow & Lea- cock, 2000; Lonsdale & Strong-Krause, 2003), to detect non-native text (Tomokiyo & Jones, 2001), and to support lexical selection by ESL learners through first-language translation (Liu et al., 2000). However, none of this work appears to directly address the more general problem of how to robustly provide feedback to ESL writ- ers—and for that matter non-native writers in any second language—in a way that is easily tai- lored to different L1 backgrounds and second- language (L2) skill levels. In this paper, we show that a noisy channel model instantiated within the paradigm of Statis- tical Machine Translation (SMT) (Brown et al., 1993) can successfully provide editorial assis- tance for non-native writers. In particular, the SMT approach provides …

Content truncated to 500 words