Frequency and word replacement
Pagel, Atkinson and Meade compared 200 basic meanings across 87 Indo-European languages. They estimated how often a form was replaced by an unrelated form on a language family tree and compared the rates with meaning frequencies in English, Spanish, Russian and Greek corpora. More frequently used meanings tended to be replaced more slowly; frequency did not explain every difference.
For each meaning, the study’s table gives an estimated average rate, with uncertainty, per 10,000 language-years. The half-life is the interval at which the model assigns a 50% probability of replacement by an unrelated form. It depends on the model and the dating of the tree; it does not measure how long an individual word will last.
For example, “who”, “two” and “five” each appear in a single class of related forms in the study data. The authors estimate roughly 76,530 years for this category by extrapolating their model; they did not observe the words for 76,530 years. For “dirty”, the published rate corresponds to about 750 years. Forms can change in sound while remaining related.
A different study: English irregular verbs
Lieberman and colleagues tracked 177 Old English irregular verbs: 145 remained irregular in Middle English and 98 in the Modern English used in the 2007 study. In their data, more frequent groups of verbs regularised more slowly. For those groups, half-life rose approximately with the square root of frequency. This relationship concerns English verb regularisation; it is not a universal formula for word replacement in every language.
In the study’s model, a group used 100 times as often has a half-life about ten times as long. This compares groups; it does not predict when a particular verb will regularise.
Compare frequencies
How many times more frequent is one group than the other?