Politeness and Indirectness: When Sexism Hides Behind Advice in Workplace Statements

Abstract

Sexism in workplaces is often framed as ordinary guidance or evaluation rather than as overt hostility, which can cause sentiment and toxicity filters to miss it. A corpus analysis was conducted using the Sexist Workplace Statements dataset (1,142 statements; 627 labeled sexist). A pragmatics-informed operationalization was applied to classify sexist statements as benevolent or hostile and to label each statement’s primary speech act as advice, evaluation, insult, joke, or complaint. Benevolent sexism was estimated to constitute 73.8% of sexist statements, while hostile sexism constituted 26.2%. Benevolent sexism was concentrated in evaluation and advice, whereas hostile sexism was concentrated in insults. A sentiment-or-profanity toxicity proxy achieved high precision but low recall for sexism, capturing most hostile sexism while missing most benevolent sexism. A supervised baseline (TF–IDF plus logistic regression) performed well on the binary label but still showed false negatives dominated by benevolent evaluations. The findings were interpreted through ambivalent sexism theory, speech act theory, and politeness theory, highlighting how indirectness and face-work enable discriminatory norms to be advanced under the guise of help.  These results make explicit that sexism detection systems should incorporate pragmatics- and speech-act-aware features to reliably identify benevolent, “helpful”-framed workplace sexism that standard sentiment/toxicity signals systematically overlook.