Figure 3.

Statistically estimated rates of lexical replacement for the Swadesh list in the Indo-European languages (from [6]) correlated with the first principal component loadings obtained from the frequency of use of the Swadesh list among 17 languages, and with part of speech. The relationship accounts for 46% of the variance in rates of lexical replacement (p < 0.0001), and holds separately within part-of-speech category: prepositions (‘in’, ‘with’), conjunctions (‘and’, ‘because’), adjectives (‘white’, ‘thin’), verbs (‘to throw’, ‘to eat’), nouns (‘hand’, ‘hair’), special adverbs (‘here’, ‘some’), pronouns (‘I’, ‘they’) and numerals (‘one’, ‘five’). Regression lines were allowed to have separate intercepts but constrained to have the same slope. Allowing these slopes to vary increased the overall R² to 0.47, not a significant change.