Editing Glottochronology (section)

==Modifications==
Somewhere in between the original concept of Swadesh and the rejection of glottochronology in its entirety lies the idea that glottochronology as a formal method of linguistic analysis becomes valid with the help of several important modifications. Thus, inhomogeneities in the replacement rate were dealt with by Van der Merwe (1966)<ref name=":2">van der Merwe, N. J. 1966 "New mathematics for glottochronology", Current Anthropology 7: 485–500</ref> by splitting the word list into classes each with their own rate, while Dyen, James and Cole (1967)<ref>Dyen, I., James, A. T., & J. W. L. Cole 1967 "Language divergence and estimated word retention rate", <Language 43: 150–171</ref> allowed each meaning to have its own rate. Simultaneous estimation of divergence time and replacement rate was studied by Kruskal, Dyen and Black.<ref name=kruskal1973/>

Brainard (1970) allowed for chance cognation, and drift effects were introduced by Gleason (1959). Sankoff (1973) suggested introducing a borrowing parameter and allowed synonyms.

A combination of the various improvements is given in Sankoff's "Fully Parameterised Lexicostatistics".  In 1972, Sankoff in a biological context developed a model of genetic divergence of populations. Embleton (1981) derives a simplified version of that in a linguistic context. She carries out a number of simulations using this which are shown to give good results.

Improvements in statistical methodology related to a completely different branch of science,  [[phylogenetics]]; the study of changes in DNA over time sparked a recent renewed interest. The new methods are more robust than the earlier ones because they calibrate points on the tree with known historical events and smooth the rates of change across them. As such, they no longer require the assumption of a constant rate of change ([https://web.archive.org/web/20120413143102/http://language.psy.auckland.ac.nz/publications/index.php?pub=Gray_and_Atkinson2003Nature Gray & Atkinson 2003]).

===Starostin's method===
Another attempt to introduce such modifications was performed by the Russian linguist [[Sergei Starostin]], who had proposed the following:
* Systematic [[loanword]]s, borrowed from one language into another, are a disruptive factor and must be eliminated from the calculations; the one thing that really matters is the "native" replacement of items by items from the same language. The failure to notice that factor was a major reason in Swadesh's original estimation of the replacement rate at under 14 words from the 100-wordlist per millennium, but the real rate is much slower (around 5 or 6). Introducing that correction effectively cancels out the "Bergsland & Vogt" argument since a thorough analysis of the Riksmal data shows that its basic wordlist includes about 15 to 16 borrowings from other Germanic languages (mostly [[Danish language|Danish]]), and the exclusion of those elements from the calculations brings the rate down to the expected rate of 5 to 6 "native" replacements per millennium.
* The rate of change is not really constant but depends on the time period during which the word has existed in the language (the chance of lexeme X being replaced by lexeme Y increases in direct proportion to the time elapsed, the so-called "aging of words" is empirically understood as gradual "erosion" of the word's primary meaning under the weight of acquired secondary ones).
* Individual items on the 100 word-list have different stability rates (for instance, the word "I" generally has a much lower chance of being replaced than the word "yellow").

The resulting formula, taking into account both the time dependence and the individual stability quotients, looks as follows:

: <math>t = \sqrt \frac{\ln(c)}{-Lc}</math>

In that formula, −''Lc'' reflects the gradual slowing down of the replacement process because of different individual rates since the least stable elements are the first and the quickest to be replaced, and the square root represents the reverse trend, the acceleration of replacement as items in the original wordlist "age" and become more prone to shifting their meaning. This formula is obviously more complicated than Swadesh's original one, but, it yields, as shown by Starostin, more credible results than the former and more or less agrees with all the cases of language separation that can be confirmed by historical knowledge. On the other hand, it shows that glottochronology can really be used only as a serious scientific tool on language families whose historical phonology has been meticulously elaborated (at least to the point of being able to distinguish between cognates and loanwords clearly).