Dependency Grammar Based English Subject-Verb Agreement Evaluation

Dependency Grammar Based English Subject-Verb Agreement Evaluation1 Dongfeng Cai a, Yonghua Hu a, Xuelei Miao a, and Yan Song b aKnowledge Engineering Research Center Shenyang Institute of Aeronautical Engineering No.37 Daoyi South Avenue, Daoyi Development District, Shenyang China 110136 caidf@vip.163.com, sam830829@hotmail.com, miaoxl@ge-soft.com bDepartment of Chinese, Translation and Linguistics City University of Hong Kong 83 Tat Chee Ave, Kowloon, Hong Kong yansong@student.cityu.edu.hk Abstract. As a key factor in English grammar checking, subject-verb agreement evaluation plays an important part in assessing translated English texts. In this paper, we propose a hybrid method for subject-verb agreement evaluation on dependency grammars with the processing of phrase syntactic parsing and sentence simplification for subject-verb discovery. Experimental results on patent text show that we achieve an F-score of 91.98% for subject-verb pair recognition, and a precision rate of 97.93% for subject-verb agreement evaluation on correctly recognized pairs in the previous stage. Keywords: Subject-verb agreement, Sentence simplification, Dependency grammar, phrase syntactic parsing 1 Introduction Subject-verb agreement error is the most common type of mistakes made in translating other languages to English text, and affects the quality of the generated text considerably. By making a detailed analysis on 300,000 error-noted English patent texts, we found that the subject-verb agreement errors comprise 21.7% of all the translation errors. It is obviously indicated that subject-verb agreement is one of the common problems translators would encounter. Due to the complicate grammar and flexible usage of sentence types, especially the complicated relationship between subjects and predicate verbs, the subject-verb agreement evaluation is a difficult mission to tackle. Currently, manual proofreading is still the main approach widely applied in detecting subject-verb agreement errors made by translators. However, it costs too much while in low efficiency, and manual work is not capable of reuse. To solve this problem, a computational approach is proposed in this paper to automatically recognize the subject-verb pairs and evaluate their agreement by obtaining the dependency relationship between the subjects and its predicate verbs. Phrase syntactic parsing and sentence simplification are used and proved to be effective in our routine. The rest of the paper is organized as follows: a concise survey of related works is presented in the next section; section 3 is the description of our method; section 4 illustrates the procedure of our experiments; and the experimental results with analysis are presented in section 5; section 6 is the conclusion. Copyright 2009 by Dongfeng Cai, Yonghua Hu, Xuelei Miao, and Yan Song 63 23rd Pacific Asia Conference on Language, Information and Computation, pages 63–71 2 Related Works For there are limited researches exclusively focus on subject-verb agreement, many related works are reported on dealing with grammatical errors, some of which includes subject-verb agreement case. Atwell (1987), Bigert and Knutsson (2002), Chodorow and Leacock (2000) proposed the n-gram error checking for finding grammatical errors. Hand-crafted error production rules (or “mal-rules”), with context-free grammar, are designed for a writing tutor for deaf students (Michaud et al., 2000). Similar strategies with parse trees are pursued in (Bender et al., 2004), …

Content truncated to 500 words