Automatic Assistance for Academic Word Usage

Proceedings of the 28th International Conference on Computational Linguistics, pages 2163–2168 Barcelona, Spain (Online), December 8-13, 2020 2163 Automatic Assistance for Academic Word Usage Dariush Saberi, John Lee, Jonathan Webster Department of Linguistics and Translation, The Halliday Centre for Intelligent Applications of Language Studies, City University of Hong Kong dsaberi2-c@my.cityu.edu.hk, jsylee@cityu.edu.hk, ctjjw@cityu.edu.hk Abstract This paper describes a writing assistance system that helps students improve their academic writ- ing by incorporating vocabulary that is more typical in the academic setting. Given an input text, the system suggests word substitutions according to an academic word list, and ranks them with a masked language model. Experimental results show that lexical formality analysis can improve the quality of the suggestions, in comparison to a baseline that relies on the masked language model only. Further, a user study demonstrate that students were able to use the system to improve text quality. 1 Introduction While most research on automatic writing assistance has focused on grammatical error correction (Ng et al., 2014), there has been increasing attention on analyzing academic writing, with respect to the writing style expected in the genre (Bailey, 2011). Examples of recent efforts include identification of argumentation structure (Zhang et al., 2017), automatic assistance for nominalization and sentence restructuring (Lee et al., 2019), and lexical substitution of academic vocabulary (Yimam et al., 2020). This paper focuses on the latter, an example of which is shown in Table 1. Lexical substitution (LS) is the subfield of natural language processing that aims to replace target words in a text without changing its meaning (McCarthy and Navigli, 2009). Our task — academic LS — may be viewed as a special form of LS, with the additional requirement that the substitution be typical for academic discourse. We describe and evaluate a writing assistance system for academic word usage that incorporates lexical formality analysis. Experimental results show that the incorporation of formality analysis helps raise the performance in identifying suitable substitutions. 2 Previous work The system reported in Yimam et al. (2020), which performs academic lexical substitution (LS) using the typical LS pipeline (Paetzold and Specia, 2016), is closely related to ours. In the first step, the system identifies the non-academic target words with a binary classifier. The second step, Substitution Gener- ation, proposes candidate substitutions for each target word. Finally, the system ranks these candidates to determine the optimal substitution. The candidate substitutions were harvested from PPDB (Pavlick et al., 2015) and WordNet (Fellbaum, 1998), and ranked with the TF-Ranking deep learning model (Pa- sumarthi et al., 2019). Automatically mined pairs of informal and formal words (Brooke et al., 2010) have been shown to be useful for natural language generation (Sheikha and Inkpen, 2011). Lexical formality analysis may also be beneficial for academic LS since academic vocabulary tends to be more formal. Although informal- formal text pairs are now publicly available in large volume (Pavlick and Tetreault, 2016), they have not yet been exploited in academic LS. This work is licensed under a Creative Commons Attribution 4.0 …

Content truncated to 500 words