Fig. 6: Evidence that natural languages are configured in a way that reduces predictive information, in phonotactics, morphology and syntax.

a, Predictive information calculation for phonological forms in selected languages, comparing the attested forms against forms that have been deterministically shuffled while preserving manner of articulation. b, Letter-level predictive information of noun morphology (vertical black line), compared against predictive information values for four random baselines (densities of 10,000 samples; see text). P values indicate the proportion of baseline samples with lower predictive information than the attested forms. c, Letter-level predictive information of adjective–noun pairs from 12 languages, compared with baselines. Non-local baselines always generate much higher predictive information than the attested forms and are not shown.