Dependency locality as an explanatory principle for word order Supplemental Material

Figure 2 (p. 390).

Histograms of languages by proportion of sentences with shorter dependencies than average baseline reorderings. One histogram per baseline. The vertical line indicates 50% on the x-axis. Darker gray blocks indicate languages where the significant majority of sentences have shorter dependencies than the baseline. Light gray indicates languages where the proportion of such sentences is not significantly different from one half. Black indicates languages where the significant majority of sentences in the corpus have longer dependency length than the baseline. The black exceptions are corpora of Uyghur (proportion of optimized sentences = 0.46), Latin (0.45), and Northern Kurdish (0.39) when compared to the fixed random baseline, and Korean when compared to the fixed (0.46) and free (0.47) head-consistent random baselines.