Detecting Censorship and Self-Censorship: NLP Analysis of Political Discourse in Iraqi Social Media and Blogs
Bibliographic record
Abstract
Objectives: This article explores how censorship and self-censorship shape political discourse in Iraqi social media posts and blogs. It focuses on how users avoid or encode taboo topics—such as criticism of powerful parties, militias, and security institutions—through euphemisms, omissions, and indirect reference. Methods: Using a corpus of 1.2 million posts and comments from Iraqi Twitter, Facebook pages, and independent blogs (2018–2024), the study applies named-entity recognition, keyword-based topic modelling, and pattern mining to identify both explicit and obfuscated references to political actors and events. It constructs lexicons of “direct terms” (official names, titles) and “evasions” (nicknames, code words, metaphors) through iterative annotation. Temporal analysis compares the use of direct versus obfuscated language around key political events, while qualitative close reading grounds the statistical patterns in concrete examples of online speech. Results: The analysis reveals systematic strategies of self-censorship: users avoid naming certain actors directly, substitute descriptive phrases or initials, and rely on shared code words that outsiders might miss. Peaks of euphemistic language coincide with periods of heightened repression, harassment, or online trolling. At the same time, overtly critical discourse persists in semi-anonymous and diasporic spaces. Conclusions: The article argues that combining NLP techniques with grounded reading can help map the “shadow vocabulary” of Iraqi political talk online and clarify how fear, risk, and creativity shape digital expression. It also highlights the ethical responsibility of researchers working with such sensitive data, especially regarding anonymity, re-identification risks, and potential harm to vulnerable speakers.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.016 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.005 | 0.005 |
| Science and technology studies | 0.002 | 0.002 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.001 | 0.003 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".