Deep Learning-Based Discourse Analysis of English Social Media Communication
- Authors
-
-
Suman Thapa
Author
-
- Keywords:
- discourse analysis, english language education, social media communication, computer-mediated discourse, second-language pragmatics, deep learning, discourse markers
- Abstract
-
The use of English in social media has evolved as one of the main places where informal English is used, but the discourse of social media is under-researched when it comes to the field of English language education within a systematic, mixed-methods framework. The aim of this study is to examine the discourse-pragmatic aspect of English-medium communication on social media platforms (SMS) of three selected platforms and its implication for ELT. A corpus consisting of 1255 discourse units was gathered from public posts and comment threads on Facebook, X/Twitter and Instagram and analysed at the structural, pragmatic and interactional level of description, following Herring's computer-mediated discourse analysis (CMDA). A lightweight bidirectional long short-term memory (BiLSTM) classifier was used in a strictly auxiliary capacity to pre-tag discourse units according to their likely speech-act categories, and trained human coders, who discussed their disagreements, then applied the full CMDA scheme, so the automated layer did not replace but aided the qualitative interpretation. Results reveal that there is a patterning of lexical reduction, stance taking, hedging, code mixing, phatic framing and the use of discourse markers specific to each platform, with X/Twitter being more marker rich and Facebook more lengthy and hedged, and using more phatic exchanges. Instagram was somewhere in the middle, with particularly high visual-stance signalling. The inter-rater agreement for the human-coded layer was high (Cohen's κ = 0.79) and the auxiliary classifier achieved an F1-score of 0.82, indicating that categories were applied to a reasonable extent of consistency. The patterns are placed in the context of second-language pragmatic competence, and it is suggested that social media discourse is an underused authentic register for hedges, politeness and register-shift in the English language classroom. Discourse excerpts derived from the corpus, which are short, are recommended for being incorporated into the teaching materials of English without replacing the usual registers of the curriculum. The study adds to the literature an empirically grounded, platform-comparative discourse account of English social media that is directly applicable to the work of English language teachers and curriculum designers.
- Downloads
- Published
- 2025-06-30
- Issue
- Vol. 14 No. 2 (2025)
- Section
- Articles