Supplementary Material C: Borrowings and Chance Resemblances
============================================================
The files in this folder summarize all wrong decisions of the different methods
resulting from borrowing and chance resemblances in the KSL datset. Every
single wrong decision is listed in the files named after the schema
<borrowings-method.txt> or <chances-method.txt>. The files consist of three
columns. The first column gives the ID of the basic vocabulary item, the second
and third column contain the language entries (with the language name in
brackets) which were incorrectly judged to be cognate.

In the KSL dataset there are 499 cognate pairs out of a total of 5600 word
pairs. 72 of the word pairs are borrowings. 83 word pairs are chance
resemblances, i.e. they are neither cognates, nor borrowings, but their NED
score is below 0.6.

Comparing the cognate judgments of the four different methods yields the
following results:

                                                LexStat SCA     Turchin NED
Number of borrowings judged to be cognate       36      44      38      35
Perc. of borrowings judged to be cognate        50      61      53      49
Number of chance resemb. judged to be cognate   14      35      26      74
Perc. of change resemb. jugded to be cognate    17      42      31      89
