DH 2026

Daejeon, July 27–31

Poster

In-Group Language in Antidepressant Withdrawal Support Communities: An NLP Analysis

Sue Young Chung
Korea University, Korea, Republic of (South Korea) · suechung@gmail.com
Eugene Chung
Korea University, Korea, Republic of (South Korea) · echung2@korea.ac.kr

Antidepressant prescribing has risen significantly, yet clinical guidance and support for discontinuing these drugs when patients no longer benefit from them remain inadequate (Boland et al., 2025; Read et al., 2023). Surveys indicate that approximately half of individuals taking antidepressants report experiencing withdrawal symptoms when attempting to reduce or discontinue, yet many feel that healthcare professionals are ill-informed and unprepared to provide effective guidance (Boland et al., 2025; Read et al., 2023). Surveys indicate that approximately half of individuals taking antidepressants report experiencing withdrawal symptoms when attempting to reduce or discontinue, yet many feel that healthcare professionals are ill-informed and unprepared to provide effective guidance (Read et al., 2023). Evidence further suggests that limited high-quality research and insufficient training leave many clinicians reluctant to support tapering, contributing to patients’ sense of being misinformed or disbelieved (Read et al., 2023). Evidence further suggests that limited high-quality research and insufficient training leave many clinicians reluctant to support tapering, contributing to patients’ sense of being misinformed or disbelieved (Boland et al., 2025). As a result, patients turn to peer-led online support communities where individuals validate one another’s symptoms and share detailed, experience-based advice. In these digital spaces, where physical cues are limited, language does more than convey information; it helps members build alignment, group cohesion, and shared understanding (Boland et al., 2025). As a result, patients turn to peer-led online support communities where individuals validate one another’s symptoms and share detailed, experience-based advice. In these digital spaces, where physical cues are limited, language does more than convey information; it helps members build alignment, group cohesion, and shared understanding (Pennebaker & Chung, 2012; Pérez-Sabater, 2021). This study examines how these communities employ specific vocabulary and distinctive linguistic patterns to construct a shared understanding of withdrawal.(Pennebaker & Chung, 2012; Pérez-Sabater, 2021). This study examines how these communities employ specific vocabulary and distinctive linguistic patterns to construct a shared understanding of withdrawal.

The study draws on sociolinguistic theories of how group identity and solidarity are expressed through language. Eastman (1985) explains how group-specific language establishes social identity through culturally specific vocabulary, recognizable topics, and shared attitudes. By learning this language, outsiders may become insiders, gaining access to the group’s informational and emotional support. Eastman (1985) explains how group-specific language establishes social identity through culturally specific vocabulary, recognizable topics, and shared attitudes. By learning this language, outsiders may become insiders, gaining access to the group’s informational and emotional support. Eckert (2019) further shows that linguistic variation helps speakers signal social alignment and construct recognizable group identities. Eckert (2019) further shows that linguistic variation helps speakers signal social alignment and construct recognizable group identities. Pennebaker and Chung (2012) identify related linguistic markers that signal group cohesion and engagement, such as function words and first-person plural pronouns (e.g., ‘we’).Pennebaker and Chung (2012) identify related linguistic markers that signal group cohesion and engagement, such as function words and first-person plural pronouns (e.g., ‘we’).

The study asks: “What distinct in-group language characteristics does the antidepressant withdrawal community exhibit compared to general online discourse?”

Three hypotheses guide the analysis:

H1 (Lexical Distinctiveness): The in-group employs a statistically distinct vocabulary compared to general English.

H2 (Semantic Distinctiveness): Common core words acquire specialized meanings within the in-group, as reflected in altered semantic neighborhoods.

H3 (Identity Markers): The linguistic differences identified in H1 and H2 function as identity markers that classify text as in-group or out-group.

To investigate these hypotheses, two corpora are used: an in-group corpus of 500,000 tokens from survivingantidepressants.org and an out-group corpus of 500,000 tokens from the Corpus of Global Web-Based English (GloWbE; Davies, 2013). For H1, lexical distinctiveness is assessed using Keyness analysis implemented in (GloWbE; Davies, 2013). For H1, lexical distinctiveness is assessed using Keyness analysis implemented in quanteda (Benoit et al., 2018). This method identifies terms that are unique to the in-group and rare in general English (e.g., ‘windows and waves’). For H2, semantic neighborhoods are visualized with Principal Component Analysis (PCA) via (Benoit et al., 2018). This method identifies terms that are unique to the in-group and rare in general English (e.g., ‘windows and waves’). For H2, semantic neighborhoods are visualized with Principal Component Analysis (PCA) via FactoMineR (Le et al., 2008). This approach provides an interpretable view of how core lexical items (e.g., ‘doctor’, ‘prescriber’) shift in meaning as a result of shared community experiences. For H3, sentence-level embeddings are generated using Sentence-BERT (Le et al., 2008). This approach provides an interpretable view of how core lexical items (e.g., ‘doctor’, ‘prescriber’) shift in meaning as a result of shared community experiences. For H3, sentence-level embeddings are generated using Sentence-BERT (all-MiniLM-L6-v2; Reimers & Gurevych, 2019) and then classified with a linear Support Vector Machine, a model chosen for its interpretability and stability with moderately sized datasets. This pipeline tests whether in-group linguistic markers create a measurable linguistic boundary between withdrawal discourse and broader English usage.(all-MiniLM-L6-v2; Reimers & Gurevych, 2019) and then classified with a linear Support Vector Machine, a model chosen for its interpretability and stability with moderately sized datasets. This pipeline tests whether in-group linguistic markers create a measurable linguistic boundary between withdrawal discourse and broader English usage.

By applying NLP methods to a specialized, community-generated corpus rather than to large-scale aggregated corpora, this study aims to offer a clearer picture of how shared language supports connection and mutual understanding during antidepressant withdrawal. It links in-group linguistic style to digital solidarity by examining how community members use recurring vocabulary and linguistic patterns to support one another (Eckert, 2019; Pérez-Sabater, 2021). In doing so, the study contributes to the conference theme of “Engagement” by showing how small, domain-specific datasets can help explain how people support one another and build shared meaning in digital spaces.(Eckert, 2019; Pérez-Sabater, 2021). In doing so, the study contributes to the conference theme of “Engagement” by showing how small, domain-specific datasets can help explain how people support one another and build shared meaning in digital spaces.

Keywords: in-group language; group identity; digital solidarity; online support communities; Natural Language Processing

References
  1. Benoit, K., Watanabe, K., Wang, H., Nulty, P., Obeng, A., Müller, S., & Matsuo, A. (2018). quanteda: An R package for the quantitative analysis of textual data. Journal of Open Source Software, 3, 774, Article 30.
  2. Boland, M., Higgins, A., Kwak, S., & Cadogan, C. (2025). ‘I Wish It Were More Often Told to People Before They Are Prescribed These Medications How Hard It Is to Get Off Them’: A Qualitative Descriptive Analysis of Free-Text Responses to a Survey on Reducing and Stopping Psychiatric Medication. Health Expectations, 28(4), e70384.
  3. Davies, M. (2013). Corpus of Global Web-Based English. Available online at https://www.english-corpora.org/glowbe/.
  4. Eastman, C. M. (1985). Establishing social identity through language use. Journal of Language and Social Psychology, 4(1), 1–20.
  5. Eckert, P. (2019). The individual in the semiotic landscape. Glossa: a journal of general linguistics, 4(1).
  6. Le, S., Josse, J., & Husson, F. (2008). FactoMineR: An R Package for Multivariate Analysis. Journal of Statistical Software, 25, 1–18, Article 1.
  7. Pennebaker, J. W., & Chung, C. K. (2012). Language and Social Dynamics (Technical Report ARI/FB-TR-1318). U.S. Army Research Institute for the Behavioral and Social Sciences.
  8. Pérez-Sabater, C. (2021). Moments of sharing, language style and resources for solidarity on social media: A comparative analysis. Journal of Pragmatics, 180, 266–282.
  9. Read, J., Lewis, S., Horowitz, M., & Moncrieff, J. (2023). The need for antidepressant withdrawal support services: Recommendations from 708 patients. Psychiatry Research, 326, 115303.
  10. Reimers, N., & Gurevych, I. (2019). Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks. arXiv preprint arXiv:1908.10084.