Indonesian Swearing in App Reviews Is More Than Aggression, Corpus Study Finds

Swearing is usually treated as a simple signal of offensiveness, but a sweeping new study of Indonesian app reviews suggests that taboo words do far more than attack. Researchers at Universitas Sumatera Utara and Universitas Diponegoro analyzed 500,599 non-empty Google Play comments across three very different apps: the competitive game League of Legends: Wild Rift, the streaming service Netflix, and the social media platform X/Twitter. Their conclusion is striking: Indonesian digital swearing performs catharsis, complaint, humor, solidarity, evaluative intensification, and ideological stance-taking, and these functions often emerge simultaneously rather than sequentially.

The study, published in Frontiers in Language Sciences, adopts a mixed-method corpus-pragmatic design that integrates large-scale quantitative profiling with qualitative contextual interpretation. The researchers compiled three Google Play datasets spanning years of user feedback: League of Legends reviews covering October 2020 to August 2024, Netflix reviews covering December 2013 to July 2024, and X/Twitter reviews covering January 2018 to July 2024. After cleaning and normalization, the final analytical corpus contained 500,599 comments with more than 3.6 million tokens, providing a robust empirical foundation for comparative analysis.

Detection relied on a curated lexicon built through four procedures: consultation of prior swearing and impoliteness literature, corpus frequency mining, manual keyword-in-context review of candidate taboo forms, and iterative refinement during pilot coding. The lexicon captured non-standard variants such as anjg, anj, anjir, and njing, which were treated as connected to the animalized-reference family but interpreted contextually rather than automatically equated. Six non-exclusive binary categories operationalized swearing: cognitive devaluation, animalized reference, corporeal taboo, death/dirt/disgust, moral abuse, and expletive intensification. Each category was coded as present or absent per comment, allowing one comment to contain multiple functions at once.

The corpus profile reveals systematic platform differences. Swear-review percentages were similar for League of Legends and Netflix, at roughly 5.7 percent each, but X/Twitter showed a substantially lower rate of 2.6 percent. Yet density told a different story: Netflix had the highest swear density per 1,000 tokens at 8.22, compared with 6.72 for League of Legends and 5.54 for X/Twitter. This means Netflix swearing is more concentrated in shorter review texts, consistent with service-complaint intensification. League of Legends users swore nearly twice as often as X/Twitter users despite similar polarized rating distributions, indicating that swearing is not merely a function of negative sentiment but is shaped by platform-specific communicative norms.

Statistical evidence confirmed that swearing concentrates in low ratings. An odds ratio of 8.53 across all platforms showed that swear-containing reviews were far more likely to be negative, with a chi-square value of 18008.66 and p < 0.001. Spearman’s rho between swear count and star score was −0.18, confirming that more swearing accompanies lower ratings. But the platform-specific odds ratios diverged sharply: 4.37 for League of Legends, 14.30 for Netflix, and 17.56 for X/Twitter. The researchers interpret this asymmetry as follows: in League of Legends, taboo language is frequent and routinized as a channel for competence-based blame and cathartic release, so any single instance carries comparatively diffuse affective weight. On X/Twitter, users swear far less often and reserve taboo language for deliberate, marked stance-taking, so each occurrence concentrates greater evaluative and ideological force.

The most consequential reliability finding concerns the gap between lexical detection and contextual interpretation. Cohen’s kappa between the lexicon-based detector and contextual pragmatic coding was 0.3858 overall, which the researchers classify as fair agreement under conventional interpretation. League of Legends reached moderate agreement at 0.5091, whereas Netflix at 0.3555 and X/Twitter at 0.315 remained in the fair range, indicating that service complaints and social-media stance-taking create more ambiguous pragmatic environments. If lexical swearing were pragmatically transparent, the two detectors would converge strongly. Their partial divergence instead supports the central claim that Indonesian digital swearing is context-sensitive, multifunctional, and resistant to binary classification.

Qualitative keyword-in-context analysis using AntConc 4.3.1 illustrated how the same lexeme performs different pragmatic work across platforms. In League of Legends reviews, cognitive devaluation and animalized metaphor dominated, often targeting the competence of co-players rather than the platform itself, as when users blamed teammates who played like beginners. In Netflix reviews, corporeal taboo was proportionally highest, converting technical service failure into moralized complaint; one reviewer who paid but still encountered errors described the app with feces-related swearing and a vulgar corporeal term to intensify dissatisfaction rather than to identify a literal target. In X/Twitter reviews, swearing frequently expressed frustration toward access problems or platform changes, and forms such as anjir sometimes performed stance and amusement rather than direct aggression, which explains why lexicon-only classification would overclassify such comments.

Graphological strategies amplified pragmatic force throughout the corpus. Capitalization, repeated punctuation, and vowel or consonant lengthening such as anjingg or anjirrr functioned as written substitutes for prosody, intensity, and affect. Code-mixing expanded pragmatic range, especially in forms such as fak, which borrowed the affective force of English profanity while being localized through Indonesian orthography. The researchers describe pragmatic multifunctionality as emerging through inferential modulation: a lexeme carries a literal or historically recoverable meaning, but its situated function depends on target, co-text, platform genre, graphology, and shared community expectations. Thus anjing, which literally means dog, does not automatically mean the same thing in every review; its pragmatic value is constructed through the interpretive relation between word, target, and platform setting.

The study contributes a layered model in which no single layer determines meaning. The lexical-semantic layer supplies taboo potential, but pragmatic function is produced only when lexical meaning interacts with platform affordance and socio-cognitive inference. Platform affordances, including anonymity, public visibility, asynchronous review posting, star ratings, and the absence of immediate conversational repair, do not determine meaning but condition the kinds of pragmatic action that become salient and recognizable. League of Legends reviews import the interactional logic of competitive play, where blame, expertise, and rank are central; Netflix reviews import the logic of consumer entitlement and service accountability; X/Twitter reviews import a broader social-media repertoire including public critique, humor, identity display, and collective reaction to platform change.

The practical implications are concrete. A moderation system should not assign the same pragmatic weight to aggressive tolol targeting a teammate and solidarity-marking anjir in an X/Twitter review. App developers can use complaint-intensity patterns, such as corporeal taboo swearing in Netflix reviews, as a signal of severe service dissatisfaction rather than merely as removable toxicity. Sentiment-analysis models likewise require platform-specific lexicons because high-density swearing in League of Legends does not necessarily carry the same sentiment value as lower-density stance swearing on X/Twitter. The researchers acknowledge limitations: the study relies on a single primary annotator, Google Play comments represent a specific digital genre that should not be generalized uncritically to other communication forms, temporal coverage ends in 2024, and the statistical procedures require further robustness testing. Even so, the work offers a culturally situated, data-grounded model for interpreting Indonesian digital swearing beyond binary offensive-language classification, with direct relevance for platform-sensitive annotation, sentiment analysis, and content moderation.

Subject of Research: Pragmatic multifunctionality of Indonesian swearing in Google Play app reviews across gaming, streaming, and social media platforms

Article Title: Pragmatic multifunctionality of Indonesian swearing in google play app reviews: a mixed-method corpus study

Article References: Soraya, R., Nurlela, N., Zein, T. T., Prihantoro, Mulyadi, M., & Mahriyuni, M. (2026). Pragmatic multifunctionality of Indonesian swearing in google play app reviews: a mixed-method corpus study. Frontiers in Language Sciences, 5, Article 1879834. https://doi.org/10.3389/flang.2026.1879834

Image Credits: AI Generated

DOI: 10.3389/flang.2026.1879834

Keywords: swearing, pragmatics, Indonesian language, corpus linguistics, app reviews, content moderation, sentiment analysis, platform affordances, League of Legends, Netflix, X/Twitter, offensive language detection