Corpus-driven insights into the discourse of women survivors of Intimate Partner Violence

Abstract

Despite its ubiquity, Intimate Partner Violence (IPV) is still under-researched from a Critical Discourse Studies (CDS) perspective. Thus, this paper investigates the discourse of women survivors of IPV focusing on a corpus-driven examination of the data. This is done after applying the text-analysis software tool LIWC (Linguistic Inquiry and Word Count) to a 120,000-word corpus collected from an anonymised, public, online forum available to IPV survivors. I contrast a plethora of linguistic phenomena in three online communities embedded within this forum (‘Is it Abuse?’, ‘Getting out’ and ‘Life after abuse’) in the attempt to sketch out how the discursive output varies across these three stages. This paper shows how pronominal distribution plays a role in the forging of collective identity. Differences in the emotional tone across the three explored groups are also identified. Useful though these corpus-driven pointers may be, this study also warns of the precaution with which findings solely deriving from quantitative analyses need to be treated.

A Systematic Literature Review of the Use of Computational Text Analysis Methods in Intimate Partner Violence Research

“…Six studies used unsupervised topic modelling or exploration as their primary approach (More & Francis, 2021;Rodriguez & Storer, 2020;Sanchez-Moya, 2017;Xu, Zeng, Tai, & Hao, 2022;Xue et al, 2020;Xue et al, 2019). Here we use 'unsupervised' to mean that a dataset has no labels or annotations-it is simply a collection of instances of raw text data (for example, a collection of Tweets without any categories or labels assigned to each Tweet).…”

“…Evaluation of the studies which used unsupervised approaches was much more variable, reflecting the difficulties in evaluating unsupervised methods more broadly (Zhao et al, 2015). Some unsupervised studies did not include any explicit evaluation of their technique (Xu et al, 2022) or were using tools developed and tested in previous research (such as LIWC (Sanchez-Moya, 2017)). Other studies which used unsupervised topic modelling attempted to evaluate the optimal number of topics, using methods such as Rate of Perplexity Change (RPC) (Xue et al, 2019).…”

“Lock, Load, n’ Thank the Driver”: The Positive Influence of Prosocial Activity on Language in Online Social Groups

“…The LIWC (Pennebaker et al, 2007) is a dictionary-based Linguistic marker detection tool. It has been used extensively in the analysis of mood and tone in online text-based observational research (Bäck et al, 2018; Klein et al, 2018; Sánchez-Moya, 2018). LIWC contains 70+ dictionaries that have been compiled by several groups cross coding and “selecting for fit” from larger correlated dictionaries (Pennebaker et al, 2015).…”