|
nif:isString
|
-
This study was conducted according to the appropriate ethical guidelines and approved by the Institutional Review Board (IRB) at the University of Wisconsin—Madison. Participants were at least 18 years old and were fully informed of what the study involved. Because obtaining signed consent was impractical in the online study, the IRB approved a waiver for signed consent. No sensitive information was collected, and all data were confidential. We analyzed only anonymous data. We report all data exclusions, all manipulations, and all measures. The data and analysis files and all laughter clips are available online (https://osf.io/ca66s/). We recruited 768 online participants on Amazon’s Mechanical Turk and TurkPrime [47] to “rate 50 very brief audio clips of people laughing” in exchange for $2 (all participation occurred May 11–12, 2017). Five participants reported audio malfunctions and one participant reported that he did not listen to the sounds before rating them; excluding these participants resulted in a sample of 762 (see Table 1 for participant demographics).
Table data removed from full text. Table identifier and caption: 10.1371/journal.pone.0183811.t001 Participant demographics. After reading the consent information, participants were randomly assigned to judge the degree to which the laugh samples communicated a meaning related to one of the four dimensions (spontaneity n = 172, reward n = 254, affiliation n = 166, dominance n = 170). Each participant evaluated the laughs on just one of the four rating scales so experimental demands would not lead them to rate each laugh as high on only one dimension. Due to a programming error, the reward condition was oversampled. Each participant rated a subsample of 50 laughs randomly drawn from the entire pool of 400 laughs. Each laugh was rated on a given dimension approximately 24 times (762 participants *50 judgments) / (400 laughs * 4 rating dimensions). Instructions asked participants to rely on their “spontaneous impressions” to “rate the extent to which you think the…description fits this clip”. The descriptions, which varied across conditions, were accompanied by a 10-point Likert scale (1 = “not at all”, 10 = “very much”): Spontaneity condition: “Laughter can sometimes be spontaneous. You could feel that someone’s laughter is unintentional and is occurring outside of their control.”Reward condition: “Laughter can sometimes be rewarding. You could feel that someone’s laughter means they like something that you did or said.”Affiliation condition: “Laughter can sometimes be reassuring. You could feel that someone’s laughter means they are acknowledging you and want you to know they are not threatening.”Dominance condition: “Laughter can sometimes be mocking. You could feel that someone’s laughter means at this moment they feel superior to or dominant over you.”After rating 50 laughs, participants answered several demographic and task feedback questions.
To maximize the variability of our laughter sample, we obtained our stimuli from Sound Snap, a professional online sound library (soundsnap.com). Sound Snap’s voice recordings are licensed by sound designers and producers; as such, they are largely produced in recording studios and often sound artificial. This is particularly important to consider in laughter, as spontaneity strongly influences perceiver judgments. However, we think it is appropriate to use these somewhat artificial stimuli in the current study for two reasons. Firstly, our social functional account is agnostic about the feeling states underlying an expression, instead seeking to identify common social consequences. Secondly, posed and synthetic facial expressions have been instrumental in identifying the action units relevant to certain emotions or social functions [48], and distilled, sometimes caricatured expressions often exaggerate the most essential features of an expression [49]. On April 19, 2017, we used the following keywords in a Sound Snap search, which returned 598 audio clips: LAUGH* -*BOY* -*GIRL* -CARTOON* -GROUP* -CROWD* -ANIMAL -WOMEN -MEN -LADIES -KID -BAB* -TODDLER* -TALK* -SPEECH -SPEAK* -MANIC (dashes precede excluded keywords). Clips were then eliminated from the initial search return for the following reasons: contained no adult human laughter; contained speech, ambient noise, or multiple speakers; were low-quality vintage recordings; or were tagged with the words “ghost,” “clown,” “cartoon,” or “crazy.” This resulted in 400 relevant laughter samples (256 male, 144 female). We then trimmed any silence from the beginning and end of the samples.
Eleven acoustic features were extracted from the 400 laugh samples using PRAAT [50] (see Table 2 for descriptive statistics). We describe the variables and the motivation for their inclusion in the current study below: Duration: The duration of the laughter sample in seconds, log-transformed to correct for positive skew. In at least one study, spontaneous laughter bouts were longer than volitional bouts [30, cf 3].Intensity: The mean intensity, or loudness, in dB. Greater intensity may be an indicator of reduced inhibition [51] or increased laughter spontaneity [3].Pitch variables:F0 mean refers to mean fundamental frequency, or pitch, as calculated using PRAAT’s auto-correlation technique. F0 range is the difference between the lowest and highest F0 for each sample. Standard deviation of F0 divided by the total duration (SD F0 / duration) of the sample captures the average moment-to-moment variability in pitch; this variable was log-transformed to correct for positive skew. Slope is the mean absolute F0 slope, which measures how sharply the pitch changes occur by dividing the difference between a local F0 maximum and minimum (at intervals of .01 seconds) by the duration it takes to go from one to the other. Raised F0 and greater SD F0 / duration and F0 range are associated with spontaneity in laughter [3,30]. Steeper F0 slopes are associated with high arousal emotion states [52]. To correct for the skewed distribution of pitch variables on a Hertz scale, F0 mean, slope, and F0 range were transformed from Hertz to a semitone scale (12*log(X)), with F0 range calculated as a ratio of the maximum to minimum F0 (12*log(maximum/minimum)) [3].Spectral variables:Center of gravity refers to the spectral centroid, which accounts for the weighting of noise across the sample (log-transformed). Changes in center of gravity can correspond to the oral-nasal distinction in vowels [53] and the perception of vowel height in nasal vowels [54]. More generally, center of gravity is an indicator of the timbre, or brightness, of a sound, with higher centers sounding brighter [55]. Spontaneous laughs in one study had higher centers of gravity than volitional laughs [30]. Harmonics-to-noise ratio is the average degree of periodicity in dB; a higher value indicates a purer, more tonal sound, and a lower value indicates a noisier vocalization. Proportion voiced is the proportion of frames that are voiced as opposed to unvoiced. Voiced segments are nearly periodic, while unvoiced segments are noisier, and include exhalations and snorts [12]. Previous work showed that spontaneous laughs have more unvoiced segments [30] and longer intervals between voiced bursts [3] compared to volitional laughs. Laughs intended to portray teasing and schadenfreude have lower harmonics-to-noise ratios than laughter intended to portray tickling [29]. The spectral variables should be interpreted cautiously due to the laughter samples’ unknown recording environments and possible compression at some point in the editing process.Formant variables:F1 mean and F2 mean, or the first and second formants (transformed to semitones), are peaks in the sound spectrum that help determine the vowel sound of a vocalization. Lowering F1 and raising F2 results in a “higher” vowel (e.g., shifting from /aː/ to /iː/). Spontaneous and rewarding laughter may be expected to feature high F1 means based on previous research [56], as a higher F1 is associated with higher arousal [57]. Raised F2 can convey increased positivity [58]. A general raising of the vowel sound, which involves increasing the relative dispersion of the first and second formants, creates the illusion of a smaller body size [19], as formant spacing is much more strongly related to body size than F0 [59,60]. Furthermore, open vowel sounds are associated with high-arousal calls in monkeys [61]. Formant positioning therefore has the potential to predict perceptions of all four social dimensions in laughter [21]. Table data removed from full text. Table identifier and caption: 10.1371/journal.pone.0183811.t002 Summary statistics for the 11 acoustic measures and subjective ratings for the 395 laughs included in the key analyses, separated by the sex of the actor (female n = 142, male n = 253). Note. Where indicated, pitch variables have been transformed to a semitone scale (ST) for analyses, although F0 Mean is also reported in Hertz. F0 Range is the change in semitones from the minimum to the maximum pitch. *Indicates that a variable was subsequently log-transformed for analyses due to non-normality, identified via visual inspection.
Five laugh samples were removed from subsequent analyses because they had no voiced frames and were therefore missing values for pitch variables. Inspecting the summary statistics suggests participants rated these unvoiced laughs as lower on reward (M = 2.84, SD = 2.23), affiliation (M = 3.04, SD = 2.29), and dominance (M = 4.26, SD = 3.23) than the other 395 laughs (mean spontaneity ratings are not noticeably different, M = 4.82, SD = 3.05, see Table 2 for descriptives of included laughs).
|
![[This page as RDF]](http://data.gesis.org/somesci/static/rdf-icon.gif){kind=icon}