Historical Linguistics & Lexicostatistics

In the present book several aspects of statistical application for comparative-historical linguistics are investigated; this includes attempts to provide a precise genetic classification of languages. Most papers in the book were written in English for this volume; a few have been translated from Russian.

S. Starostin in his paper on lexicostatistics maintains that it would be incorrect to stop glottochronological studies just because M. Swadesh’s methodology shows many inadequacies (one of them being the assumption that a language loses 14 words from the basic 100 over a millennium: the figure seems much too high). One can improve the methodology by both excluding borrowings from the standard diagnostic lists of basic words and amending formulas used in the studies. These formulas should reflect the fact that different basic words of any language have different life expectancies (being, at a certain point in time, replaced by other words with the same meaning).

When properly applied, glottochronology provides, for a given period of time, similar figures for changes in the lexicon both for languages without writing and for languages with an established literary tradition.

Improved glottochronology provides rather precise absolute dating both for separation of any two dialects of a given language, and for separation of any two languages which belong to the given group or family of languages. Starostin also proposes to use a somewhat different method of glottochronology, namely, root glottochronology (which provides better retention rates when compared with word glottochronology). It is applicable to languages which may be exposed to a reliable etymological analysis in order to determine appropriate roots. Only inherited roots can be chosen for such a study. Among the applications of this method a study of a written text can be used. It is important that the style of the text (colloquial; literary; scholarly) has no bearing on the results.

To illustrate the above method Starostin investigates the 1st page of the original Russian text of his present paper. This text includes 100 roots of Indo-European (IE) origin; they are being compared with appropriate roots in Polish, Lithuanian, German, and French. Using this analysis, as well as comparative data from similar analyses of other Russian texts, Starostin shows that for every 100 Russian roots one can find in Polish approximately 95 related roots; in Lithuanian—74; in German—54; in French—51. These figures are ultimately determined by the time of the "split" of proto-Slavic into its daughter languages; of proto-Balto-Slavic; of proto-Indo-European.

When taking texts in other languages and comparing them with Russian words which contain appropriate IE roots, Starostin obtains these figures: for 100 German roots—55 Russian; for 100 French roots—50 Russian; for 100 Old Greek roots—52 Russian; for 100 Vedic Sanskrit roots—54 Russian. This means that "the age of the text does not influence the result of the statistical analysis". Of course, in each case this similarity reflects the distance between proto-Indo-European (of which 100 roots are present in a text written in this or that language: it doesn’t actually matter in what language) and Russian.

For the next experiment Starostin takes the Russian version of the Swadesh 100 word list and compares appropriate IE roots which those retained in Polish (getting 91 Polish cognates), Lithuanian (67—68 matches), German (50—51 matches), and French (50 matches). These results are very similar to those obtained above, when written texts were analysed. Thus, "the distribution of individual frequency characteristics of roots in the Swadesh list coincides with their usual distribution in texts." Starostin shows here that the dating of language divergence as obtained by word glottochronology practically coincides with the dating obtained when employing root glottochronology. Naturally, this methodology is not applicable if the languages