Efficacy in deceptive vocal exaggeration of human body size

Abstract

How can deceptive communication signals exist in an evolutionarily stable signalling system? To resolve this age-old honest signalling paradox, researchers must first establish whether deception benefits deceivers. However, while vocal exaggeration is widespread in the animal kingdom and assumably adaptive, its effectiveness in biasing listeners has not been established. Here, we show that human listeners can detect deceptive vocal signals produced by vocalisers who volitionally shift their voice frequencies to exaggerate or attenuate their perceived size. Listeners can also judge the relative heights of cheaters, whose deceptive signals retain reliable acoustic cues to interindividual height. Importantly, although vocal deception biases listeners' absolute height judgments, listeners recalibrate their height assessments for vocalisers they correctly and concurrently identify as deceptive, particularly men judging men. Thus, while size exaggeration can fool listeners, benefiting the deceiver, its detection can reduce bias and mitigate costs for listeners, underscoring an unremitting arms-race between signallers and receivers in animal communication.

Fig. 1. Human vocal communication of body size.

a Sagittal MRI image of human vocal anatomy illustrating vocal tract length (VTL) (in red) during production of the vowel /u:/, and permanently descended larynx housing the vocal folds (blue circle). Human VTL scales fairly allometrically with body size due to anatomical constraints, whereas vocal fold length does not^,,,. b Resonances of the vocal tract termed formants (labelled F₁–F₄) and their relative spacing (∆F) inversely predict human height controlling for sex and age^,, whereas fundamental frequency (labelled f_o), related to vocal fold length, does not. c Human listeners robustly associate both voice frequencies (low ∆F and f_o) with large body size in regular speech^, and in nonverbal vocalisations (e.g., roars), but can humans judge size from deceptive vocal signals? Panels a, b were reproduced and reformatted with permission; parts of panel c were designed using resources from freepik.com and rawpixel.com (https://www.freepik.com/free-vector/illustration-business-people_2609966.htm#query=man%20black%20silhouette&position=17).

Fig. 2. Vocal size deception biases judgements of body size (Experiment 1).

a Bias in height judgements shown as the mean difference (±SEM) between perceived and actual heights of vocalisers, in cm, for honest vocal signals (central blue bars) and deceptive vocal signals (attenuating size = grey bars, exaggerating size = red bars), where 0 indicates accurate height judgements, positive values indicate overestimation, and negative values indicate underestimation. Estimated marginal means and pairwise comparisons derive from linear mixed models, LMMs (see Supplementary Table 5), where all ***p < 0.001 following Šidák correction for multiple comparisons. Tests are two-tailed. Error bars represent standard errors of the mean, ±SEM. See also Supplementary Fig. 1 for dot plots illustrating the distribution of data. b Linear regressions showing relationships between perceived and actual heights of vocalisers, where each dot represents a vocaliser, and height judgements are averaged across listeners for each vocaliser and each size condition (exaggerating size = red circles, honest = blue squares, attenuating = grey triangles), where Cook’s Di < 0.20 (see Methods). 1 cm on the x-axis is equal to 1 cm on the y-axis. Spearman’s rho (r_s) correlation coefficients are given for each regression line, where *p < 0.05, one-tailed (males exaggerating r = 0.48, p = 0.02, n = 18; honest r = 0.44, p = 0.03, n = 19; attenuating r = −0.17, p = 0.25, n = 18; females exaggerating r = 0.35, p = 0.06, n = 20; honest r = 0.54, p = 0.01; n = 18; attenuating r = 0.50, p = 0.02; n = 18). All data were derived from Experiment 1 based on 120 vocal stimuli produced by n = 40 vocalisers (20 males, 20 females) in each of three size conditions (honest, attenuating, exaggerating) and judged by n = 97 listeners, where each vocal stimulus was rated by an average of 50 listeners (see Methods). Source data are provided as a Source Data file.

Fig. 3. Listeners can detect deception, but remain deceived by deceptive signals (Experiment 1).

a Percentages of vocalisers that listeners perceived as deceptively exaggerating (red bars) or attenuating (light grey bars) their size, or as producing honest vocal signals (blue bars, centre) are shown along the y-axis as a function of the intended size deception indicated along the x-axis. Estimated marginal means and pairwise comparisons derive from LMMs (see Supplementary Table 7), where ***p < 0.001, **p < 0.01, following Šidák correction for multiple comparisons. Tests are two-tailed. b, c Bias in listeners’ size assessments as a function of whether a listener failed to detect (dark grey bars) or correctly detected (white bars) a vocal signal as deceptive or honest, where panel b shows ‘error’ in height judgements (mean difference between perceived vs actual heights of vocalisers), and panel c shows ‘deception gain’ in height judgements (mean difference between perceived height from honest signals and perceived height from deceptive signals). Estimated marginal means and pairwise comparisons derive from LMMs (see Supplementary Tables 8 and 9), **p < 0.01, *p < 0.05 following Šidák correction. Tests are two-tailed. Error bars ± SEM. Acronyms: Att. attenuating, Hon. honest, Exg. exaggerating. All data derive from Experiment 1, based on 120 vocal stimuli produced by n = 40 vocalisers (20 males, 20 females) in each of three size conditions (honest, attenuating, exaggerating) and judged by n = 97 listeners, where each vocal stimulus was rated by an average of 50 listeners (see Methods). See also Supplementary Fig. 2 for dot plots illustrating distribution of data. Source data are provided as a Source Data file.

Fig. 4. Awareness reduces bias: Listeners recalibrate height judgements for signals correctly and concurrently detected as deceptive (Experiment 2).

a Bias in height judgements shown as the mean difference (±SEM) between perceived and actual heights of vocalisers, in cm, for honest vocal signals (blue bars) and deceptive vocal signals (attenuating = grey bars, exaggerating = red bars). Estimated marginal means and pairwise comparisons derive from LMMs (see Supplementary Table 11), where all ***p < 0.001 following Šidák correction for multiple comparisons. Error bars ±SEM. b Percentages of vocalisers that listeners perceived as deceptively exaggerating (red bars) or attenuating (light grey bars) their size, or as producing honest vocal signals (blue bars, centre) are shown along the y-axis as a function of the intended size deception indicated along the x-axis. Estimated marginal means and pairwise comparisons derive from LMMs (see Supplementary Table 12), where ***p < 0.001, **p < 0.01, *p < 0.05 following Šidák correction. Tests are two-tailed. c, d Bias in listeners’ size assessments as a function of whether a listener failed to detect (dark grey bars) or correctly detected (white bars) a vocal signal as deceptive or honest, where panel c shows ‘error’ in height judgements (mean difference between perceived and actual heights of vocalisers), and panel d shows ‘deception gain’ in height judgements (mean difference between perceived height from honest signals and perceived height from deceptive signals). Panel d also illustrates that deception gain was lower for male vocalisers (left side) when correctly detected as cheating by other male listeners (labelled with ‘m’) compared to when detected by female listeners (‘f’). Estimated marginal means and pairwise comparisons derive from LMMs (see Supplementary Tables 13 and 14), ***p < 0.001, **p < 0.01, *p < 0.05 following Šidák correction. Tests are two-tailed. Error bars ± SEM. Acronyms: Att. attenuating, Hon. honest, Exg. exaggerating, m male listeners, f female listeners. All data derive from Experiment 2 based on 120 vocal stimuli produced by n = 40 vocalisers (20 males, 20 females) in each of three size conditions (honest, attenuating, exaggerating) and judged by n = 98 listeners, each of whom rated all 120 vocal stimuli (see Methods). See also Supplementary Fig. 3 for dot plots illustrating the distribution of data. Source data are provided as a Source Data file.