Learning to Identify Sentence Parallelism in Student Essays

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 794–803, Osaka, Japan, December 11-17 2016. Learning to Identify Sentence Parallelism in Student Essays Wei Song†, Tong Liu†, Ruiji Fu‡, Lizhen Liu†, Hanshi Wang†, Ting Liu§ †Information Engineering, Capital Normal University, Beijing ‡Iflytek Research Beijing, Beijing §Harbin Institute of Technology, Harbin {wsong, tongliu, lzliu, hswang}@cnu.edu.cn, rjfu@iflytek.com, tliu@ir.hit.edu.cn Abstract Parallelism is an important rhetorical device. We propose a machine learning approach for automated sentence parallelism identification in student essays. We build an essay dataset with sentence level parallelism annotated. We derive features by combining generalized word alignment strategies and the alignment measures between word sequences. The experimental results show that sentence parallelism can be effectively identified with a F1 score of 82% at pair-wise level and 72% at parallelism chunk level. Based on this approach, we automatically identify sentence parallelism in more than 2000 student essays and study the correlation between the use of sentence parallelism and the types and quality of essays. 1 Introduction Parallelism is an important rhetorical device. It can be defined as two or more coherent text spans (phrases or sentences), which have similar syntactic structures and related semantics, and express relevant content or emotion together. Each text span is a parallelism unit and the parallel units form a parallelism chunk. The following two sentences segmented by the semicolon form an example of sentence parallelism. The inherent vice of capitalism is the unequal sharing of blessing; the inherent virtue of socialism is the equal sharing of miseries. by Churchill. Parallelism adds balance and rhythm to make speeches and writings more vivid and powerful. Moreover, parallelism also adds clarity to the sentence or even the discourse. Several sentences are expressed similarly to show that the content in the sentences are equal in importance. Therefore, properly using parallelism may improve the quality of texts. On the other hand, identifying parallelism in essays would potentially help to evaluate the quality of writings and benefit applications like essay scoring and organization evaluation. In this paper, we study the problem of identifying parallelism in student essays. We focus on identifying sentence parallelism. A parallelism unit is a sentence, and several parallel sentences form a parallelism chunk. Parallelism identification is a task to find the parallelism chunks within essays. This task is nontrivial. There are several factors to be considered. Since the parallel sentences should have similar structures and related semantics, they can be seen as forming a certain kind of alignment between each other. However, such alignment can exist in various levels, from surface lexical patterns to syntactic structures, semantics and even emotions. Moreover, the alignment can occur at various granularity (words, phrases, clauses or sentences). Therefore, it is difficult to design manual rules to identify sentence parallelism. We propose a learning based framework for sentence parallelism identification. We annotate a sentence parallelism dataset consisting of about 500 student essays. This dataset allows us to derive features to model sentence parallelism and utilize machine learning to learn a …

Content truncated to 500 words