Neural (S_token, S_type), lexical (L_token, L_type), and entropic (E_token, E_type) logography measures computed on the Bible corpora. Recall that lower values for S and L measures correspond to a lower degree of logography (inverse is true for E_type). The unexpectedly low values for some of the Chinese and Japanese encodings for L as opposed to the expected higher values for the same with the S measures suggests that on balance the neural attention-based metric is better at capturing the notion of logography. The fourth column gives the per-token spelling accuracy of the neural S model on the test data. For Russian and Swedish the † marker in the columns for the E measures indicates that the number is suspect because a very small number of training and testing verses (3,289 and 3,823, respectively) were left after removing verses that contained null pronunciations.