Detecting Erroneous Sentences using Automatically Mined Sequential Patterns

Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, pages 81–88, Prague, Czech Republic, June 2007. c⃝2007 Association for Computational Linguistics Detecting Erroneous Sentences using Automatically Mined Sequential Patterns Guihua Sun ∗ Xiaohua Liu Gao Cong Ming Zhou Chongqing University Microsoft Research Asia sunguihua5018@163.com {xiaoliu, gaocong, mingzhou}@microsoft.com Zhongyang Xiong John Lee † Chin-Yew Lin Chongqing University MIT Microsoft Research Asia zyxiong@cqu.edu.cn jsylee@mit.edu cyl@microsoft.com Abstract This paper studies the problem of identify- ing erroneous/correct sentences. The prob- lem has important applications, e.g., pro- viding feedback for writers of English as a Second Language, controlling the quality of parallel bilingual sentences mined from the Web, and evaluating machine translation results. In this paper, we propose a new approach to detecting erroneous sentences by integrating pattern discovery with super- vised learning models. Experimental results show that our techniques are promising. 1 Introduction Detecting erroneous/correct sentences has the fol- lowing applications. First, it can provide feedback for writers of English as a Second Language (ESL) as to whether a sentence contains errors. Second, it can be applied to control the quality of parallel bilin- gual sentences mined from the Web, which are criti- cal sources for a wide range of applications, such as statistical machine translation (Brown et al., 1993) and cross-lingual information retrieval (Nie et al., 1999). Third, it can be used to evaluate machine translation results. As demonstrated in (Corston- Oliver et al., 2001; Gamon et al., 2005), the better human reference translations can be distinguished from machine translations by a classification model, the worse the machine translation system is. ∗Work done while the author was a visiting student at MSRA †Work done while the author was a visiting student at MSRA The previous work on identifying erroneous sen- tences mainly aims to find errors from the writing of ESL learners. The common mistakes (Yukio et al., 2001; Gui and Yang, 2003) made by ESL learners include spelling, lexical collocation, sentence struc- ture, tense, agreement, verb formation, wrong Part- Of-Speech (POS), article usage, etc. The previous work focuses on grammar errors, including tense, agreement, verb formation, article usage, etc. How- ever, little work has been done to detect sentence structure and lexical collocation errors. Some methods of detecting erroneous sentences are based on manual rules. These methods (Hei- dorn, 2000; Michaud et al., 2000; Bender et al., 2004) have been shown to be effective in detect- ing certain kinds of grammatical errors in the writ- ing of English learners. However, it could be ex- pensive to write rules manually. Linguistic experts are needed to write rules of high quality; Also, it is difficult to produce and maintain a large num- ber of non-conflicting rules to cover a wide range of grammatical errors. Moreover, ESL writers of differ- ent first-language backgrounds and skill levels may make different errors, and thus different sets of rules may be required. Worse still, it is hard to write rules for some grammatical errors, for example, detecting errors concerning the articles and singular plural us- …

Content truncated to 500 words