A Method for Detecting Determiner Errors Designed for the Writing of Non-native Speakers of English

Abstract

SUMMARYThis paper proposes a method for detecting determiner errors, which are highly frequent in learner English. To augment conventional methods, the proposed method exploits a strong tendency displayed by learners in determiner usage, i.e., mistakenly omitting determiners most of the time. Its basic idea is simple and applicable to almost any conventional method. This paper also proposes combining the method with countability prediction, which results in further improvement. Experiments show that the proposed method achieves an F-measure of 0.684 and significantly outperforms conventional methods. key words: learners' tendency, determiner error, error detection, article, English writing, learner corpus IntroductionDeterminer usage is one of the major difficulties that nonnative speakers of English are faced with in English writing. This is especially true for those whose mother tongue does not have a determiner system similar to that of English (e.g., Chinese and Japanese). It can easily be observed, among other errors, in an essay written by a Japanese learner of English:I became univercity student, I get up early every morning. I go to the school when I listening to music in train. Stady is very different. Especiary I think that programing and math doesn't know.The underlines indicate the noun phrases (NPs) that have a determiner error. Because of the difficulty inherent in using determiners, errors in determiners, including article errors, are one of the most frequent grammatical error types in learnerDeterminer errors are so frequent that they become problematic in several circumstances. For example, teachers have to identify and correct determiner errors in writing classrooms, which is time-consuming and costly. Similarly, raters have to identify a great number of determiner errors to evaluate writing skills in grammar in writing tests.Given propose a maximum entropy (ME) classifier-based method for predicting correct articles; if the prediction disagrees with the one actually used, then it is detected as an error. The features are based on lexical and syntactic information around the article in question. They report that their method achieves a recall of 0.40 with a precision of 0.90. It should be noted that these article-error detection methods can naturally be extended to determinererror detection. One can build an n-way classifier where n corresponds to the number of target determiners. The classifier selects the correct one out of the target determiners in determiner-error detection.As an alternative approach, Nagata et al.[9], [10] propose using countability prediction. Countability is highly related to determiner usage [11], [12]. For example, noncount nouns do not take the indefinite article whereas singular count nouns do not appear without a determiner. Their method first predicts the countability of the head noun from its surrounding context and then applies some rules to the prediction to examine whether the determiner that modifies the head noun is correct or not.Although performance has imp...