Results for the phonemic diversity <i>t</i>_<i>F</i>.

At the start of the simulation, only the central node of the grid is populated with a population of N = 10 tribes, speaking a mix of two languages (5 tribes speak Ewe and 5 speak Ngizim) with 40 phonemes each. The maximum number of tribes per node (N_s) is set to 10 tribes/node and the demographic threshold below which phonemes are lost is also set to 10 tribes/node. Net fecundity R₀ = 1.2. Time = 2,280 generations or about 70 kyr, corresponding to the present time from the Out-of-Africa dispersal. a) The blue and green squares are the simulation results for the phonemic diversity (measured as t_F) versus distance, computed over simulated languages binned in 1000-km intervals for two phonemic loss times, namely 60 generations/phoneme (blue line, slope = -(0.9–1.1)·10⁻³ km^-1, r = -0.983) and 120 generations/phoneme (green line, slope = -(0.4–0.57)·10⁻³ km^-1, r = -0.922). The observed phonemic diversity cline (measured as t_F) is also shown (histogram, see S1 Text, Sec A for details). The linear fit to the observed histogram (red line) has slope = -(0.5–2.1)·10⁻³ km^-1 (all slopes are reported with 95% confidence level). b) Dependence of the slope of the linear fit to the simulated phonemic diversity cline on the time of phonemic loss (e.g., if the phonemic loss time is 80 generations/phoneme, then 1 phoneme disappears every 80 generations). Each error bar has been obtained from 100 numerical simulations, and shows the 95% confidence-level interval. The dotted and dash-dotted lines give the mean, upper and lower bounds to the slope of the observed phonemic diversity cline (obtained from Fig 3A).