Discourse Mode Identification in Essays

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 112–122 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1011 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 112–122 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1011 Discourse Mode Identification in Essays Wei Song†, Dong Wang‡, Ruiji Fu‡, Lizhen Liu†, Ting Liu§, Guoping Hu‡ †Information Engineering, Capital Normal University, Beijing, China ‡iFLYTEK Research, Beijing, China §Harbin Institute of Technology, Harbin, China {wsong, lzliu}@cnu.edu.cn, {dongwang4,rjfu, gphu}@iflytek.com, tliu@ir.hit.edu.cn Abstract Discourse modes play an important role in writing composition and evaluation. This paper presents a study on the manual and automatic identification of narration, ex- position, description, argument and emo- tion expressing sentences in narrative es- says. We annotate a corpus to study the characteristics of discourse modes and de- scribe a neural sequence labeling model for identification. Evaluation results show that discourse modes can be identified au- tomatically with an average F1-score of 0.7. We further demonstrate that discourse modes can be used as features that im- prove automatic essay scoring (AES). The impacts of discourse modes for AES are also discussed. 1 Introduction Discourse modes, also known as rhetorical modes, describe the purpose and conventions of the main kinds of language based communi- cation.Most common discourse modes include narration, description, exposition and argument. A typical text would make use of all the modes, although in a given one there will often be a main mode. Despite their importance in writing composition and assessment (Braddock et al., 1963), there is relatively little work on analyzing discourse modes based on computational models. We aim to contribute for automatic discourse mode identification and its application on writing assessment. The use of discourse modes is important in writ- ing composition, because they relate to several as- pects that would influence the quality of a text. First, discourse modes reflect the organization of a text. Natural language texts consist of sen- tences which form a unified whole and make up the discourse (Clark et al., 2013). Recognizing the structure of text organization is a key part for dis- course analysis. Meurer (2002) points that dis- course modes stand for unity as they constitute general patterns of language organization strate- gically used by the writer. Smith (2003) also pro- poses to study discourse passages from a linguistic view of point through discourse modes. The orga- nization of a text can be realized by segmenting text into passages according to the set of discourse modes that are used to indicate the functional re- lationship between the several parts of the text. For example, the writer can present major events through narration, provide details with description and establish ideas with argument. The combi- nation and interaction of various discourse modes make an organized unified text. Second, discourse modes have rhetorical significance. Discourse modes are closely related to rhetoric (Connors, 1981; Brooks and Warren, 1958), which offers a principle for learning …

Content truncated to 500 words