AUTOMATING THE ANALYSIS OF REPORTING VERBS IN ACADEMIC CITATIONS: AN NLP-ASSISTED CORPUS STUDY

Abstract

Verbs that are included in reporting like to argue, suggest, claim and demonstrate are the main part of academic writing that can allow authors to command the knowledge, make judgement, and place positions in the conversation concerning disciplinary issues. Although the idea of reporting verbs as resources of stance in citation situations has long been studied through the application linguistics discipline, the prevailing analytic traditions are labor intensive and only scalable to large and heterogeneous corpora in general (Hyland, 1999, 2002; Thompson and Ye, 1991). Recent methods in natural language processing (NLP) provide practical targets to automate part of this analysis, thus being able to make reporting-verb behavior descriptions more systematic, replicable, and large-scale (Jurafsky and Martin, 2023; Teufel and Moens, 2002). This paper is a corpus-based research on demonstrating a corpus approach to reporting verbs in citation sentences and involving both quantitative method of distributional analysis and qualitative methodology of interpretation of rhetorical purpose. It uses a large computational linguistics citation dataset (the ACL Anthology Network resources) by combining it with citation sentiment annotations as well as explore the distribution of verb choice on positive, negative, and neutral citation situations (Athar, 2011; Radev, Muthu Krishnan, Qazvinian, and Abu-Jbara, 2013). A frequency profiling and polarity-conditioned comparison as well as expert validation are both supported by an NLP pipeline consisting of tokenization, part-of-speech tagging, stemming and verb-phrase extraction, and gives insight into the quality of automated extraction. The results, formulated during the present paper show that the automated processing could retrieve effective reporting verb patterns at scale and introduces logical variation attributed to evaluative contexts, and also persists in revealing difficulties facing NLP in discerning polyfunctionality, hedging and context dependent rhetorical strength.