Figure 1
Map of 993 societies represented in Cantometrics, sized by the number of songs (a total of 5,484 songs, the maximum number within one society is 72, and the minimum is 1). The inset histogram shows the global distribution of song unusualness, with a dotted line showing the top 3% of unusual songs.
Figure 2
The six steps for calculating Unusualness for any particular song are as follows. 1) Identify a region of interest (South America). 2) Identify a society of interest (Canela). 3) For all songs, excluding songs by Canela performers, calculate the probability of each state within each feature (Note the values in the figure are for illustrative purposes only). 4) For each Canelo song, take the state probability for the value coded within that song and sum it with all other state probabilities. For example: in song 3240, the Vocal Organisation variable is coded as 13 (Polyphony, red). This occurs in 10% of all South American songs, or 0.1. For interval width, song 3240 is coded as a 4 (Narrow intervals), this occurs in 0.2 of all other South American songs (blue). Nasality is coded as 10 (Slight nasalization), which occurs in 0.1 of all other South American songs. We sum these state probabilities together and take the log of the summed value to calculate unusualness (the log-likelihood). 5) Repeat from step 2 for all societies in South America. 6) Repeat from step 1 once all South American societies have been calculated.
Figure 3
Boxplots showing the spread of unusualness within the top and bottom 10 unusual societies (from societies represented by 5 songs or more).
Figure 4
This figure shows a modified profile plot of all Malay songs (grey) and the modal profile of Malay songs (red). There are 11 songs in total. The left-hand side names each feature of Cantometrics. The right-hand side offers a written description of the modal state for each feature within Malay songs. Within each row shows a point for each possible coding of each feature, transformed to a 13-point scale. Each circle is sized by the proportion of which that coding is used in Southeast Asia (larger is more common).
Figure 5
A) Posterior distributions of the regression coefficient estimates for the four predictor variables. Blue fill indicates 89% confidence intervals. Only the Nearest phylogenetic neighbour and Kinship unusualness show 89% confidence intervals not containing zero. B) Scatter plot between the unusualness score for a particular song, and the average unusualness score for all other songs within a society (Leave One Out). Scales are reversed so that further to the right of the x-axis, or higher up the y-axis is more unusual. Quantitatively smaller values are more unusual.