When perceiving a sentence like “the mother gave the candle the daughter”, we must infer whether the intended message was (1) the literal, implausible one, or (2) the plausible alternative “the mother gave the candle to the daughter” but the word “to” dropped during transmission over a noisy channel. Such inferences rely on assumptions about the channel, the language, and likely meanings, but also about the producer’s identity (e.g., their tendency to misspeak or refer to different events). We tested whether noisy-channel comprehension is influenced by a speaker’s
perceived gender, given that gender stereotypes influence many cognitive processes. Do implausible messages made by men vs. women differ in how often they are interpreted non-literally as plausible alternatives?
We used the stimuli and paradigm from Gibson et al. (2013, PNAS). Participants (N=365 on Prolific) read 20 implausible sentences (60 fillers) each presented alongside a question (“did the daughter receive something/someone?”). Their answer indicated whether they interpreted the sentence literally. Before the experiment, participants saw a photo of either a man or a woman (out of 8 possible photos) with text introducing them as the research assistant running the study. The photo remained visible throughout the experiment with the text “this is my experiment”. Participant gender was crossed with the producer’s gender (“man photo” condition: 97 men, 84 women; “woman photo” condition: 101 men, 83 women).
To replicate a central prior finding, each implausible item had two versions: a double-object (“the mother gave the candle the daughter”, suggesting an omission of “to”) and a prepositional-object (“the mother gave the daughter to the candle”, suggesting an addition of “to”) construction. Each participant read 10 items of each kind. Because omissions are more likely than additions, non-literal interpretations should be more frequent for DO vs. PO sentences.
Interpretation (literal vs. non-literal) was modeled via a mixed-effects logistic regression with the maximal converging structure. Fixed effects included participant gender, producer (photo) gender, construction, and all interactions.
Replicating prior findings, participants made more non-literal inferences for DO than PO sentences (b=1.47, SE = 0.12, c2 (1)=15.5, p<10-4). Critically, a man photo led to more non-literal interpretations compared to a woman photo (b=-0.79, SE=0.23 c2 (1)=11.8, p<10-3). Thus, when producing implausible sentences, men are “given the benefit of the doubt” more often: their implausible utterances are interpreted non-literally as plausible alternatives. We avoid drawing conclusions about the reason (perhaps participants think men are more likely to mistype; perhaps they think women are more likely to say implausible things; etc.). Regardless of the reason, we demonstrate a gender bias in noisy-channel comprehension.