Appendix 1

Letter usage in English, French, Latin, Greek, and epic poetry.

To get an idea of the difference in the frequency in which European languages and authors use their letters, I performed a principal components (PC) analysis (Kassambara, 2017) of the frequencies of Greek and Latin letters in English, French, Latin, Modern Greek, Ancient Greek, and each of the four major Greek epic poems two by Homer (the Odyssey and Iliad) and two by Hesiod (Theogony and Works and Days). The greatest proportion of the difference among languages is obvious and trivial. We know that some letters are used more frequently than others in all of these languages (see section The Proto-Canaanite scripts). The vowels, for example, are more frequent than most consonants. Some consonants are significantly rarer than others. The same is true for letter combinations, i.e., bigrams or trigrams. Clusters of consonants with vowels are more frequent than clusters of two or three consonants in all modern and ancient languages.

This trivial variation is taken into account and measured by the first PC (PC1). There is, however, variation in letter usage that is specific to a linguistic group or to a language. Obviously, the Latin and Greek alphabets are not identical. Some Greek letters are absent from Latin and vice versa. Even letters and clusters that exist in all languages within a group may present significantly different frequencies of usage. For example, K and Ck are quite frequent in English, rare in French (they use Qu instead), and hardly ever seen in Latin. The morphemes of each language, e.g. the English past -ed and gerund -ing endings, cause their respective letters to increase in usage within that language. Such language-specific or, even, text-specific variation is quantified by the subsequent principal components, PC2, PC3, etc., which are numbered by the decreasing amount of variance they explain (labeled as Dimension 1-8).

I measured the frequencies of single letters and 2 or 3-letter clusters in lexica (Balota et al., 2007; Gregory R. Crane (ed.), 2012; Kyparissiadis, n.d.; Kyparissiadis et al., 2017; Liddell & Scott, 1996; New & Pallier, 2017) where each word is only accounted for once. The number of entries in each lexicon is shown in Table 1. To make the Greek and Latin-born vocabularies comparable, I transliterated the Greek words using the LSJ beta code and removed all the diacritics for all languages. Words that differ only in tonic accent or other diacritics remain separate words and are accounted for independently for the calculation of letter or cluster frequencies. The overlap between the lexica is shown in Fig. 1, for Latin-alphabet languages, Fig. 2 for Greek languages, and Fig. 3, for Ancient Greek lexica. In the latter figure, the individual epic texts are pooled by their presumed author.

Table 1. Lexica and number of words used in the principal components analysis of letter usage frequencies.

Lexicon	Words
English	40480
French	46898
Latin	16969
Ancient Greek	36486
Modern Greek	35282
Iliad	19245
Odyssey	15456
Theogony	2848
Works and Days	2618

Figure 1. Venn diagram of the lexica of languages using the Latin alphabet

Figure 2. Venn diagram of the lexica of languages (or dialects) using the Greek alphabet

Figure 3. Venn diagram of the Ancient Greek lexica with epic dialects grouped by author.

The principal components analysis is summarised in Table 2, showing the proportion of variance explained by each PC, and Fig. 4 to 8, showing the correlations of each language with each PC and the contribution of each leater or letter-cluster (bigram or trigram) to language separation.

Table 2. Principal components of letter and letter-cluster frequencies.

Dimension	Eigenvalue	Variance (%)	Cumulative variance (%)
Dim.1	8.401	93.34	93.345
Dim.2	0.320	3.56	96.900
Dim.3	0.170	1.88	98.785
Dim.4	0.058	0.65	99.432
Dim.5	0.027	0.30	99.732
Dim.6	0.014	0.16	99.889
Dim.7	0.006	0.06	99.953
Dim.8	0.004	0.04	99.995