LEXICOSTATISTICS AS A BASIS FOR LANGUAGE CLASSIFICATION: INCREASING THE PROS, REDUCING THE CONS George Starostin “Lexicostatistics”, a method originally proposed by Morris Swadesh to build rela- tive genetic classiications of languages based on percentages of related items in their basic lexicon, and “glottochronology”, used to assign absolute dates of split- ting to language groups based on the assumption of a regular rate of change, have not been overtly popular with mainstream comparative linguists, after an early set of critical works had undermined their general credibility. Since then, however, signiicant process has been achieved in understanding and correcting the laws of the original method. The current paper focuses on drawing attention to some of these corrections, such as (a) distinguishing between externally and internally trig- gered lexical change, and (b) factoring out independent semantic innovation. This improved methodology, without signiicantly cluttering up the formal apparatus, consistently yields results that are not only more credible than Swadesh’s original procedure, but are also much more in line with standard comparative-historical linguistics. INTRODUCTION As of today, the lexicostatistical method of evaluating degrees of genetic relation- ship between different languages, based on percentages of historically related items in their basic vocabularies, is already more than sixty years old 1 . Ever since pio- neered in the 1950s by Morris Swadesh (Swadesh 1952; 1955), lexicostatistics has had a long and troubled history, rife with criticism, rejection, sometimes even open derision of the method and its supporters. Nevertheless, despite all the problems, controversies, and misunderstandings, we can now state with certainty that the method itself has survived – due partially to the relative ease of its practical applica- tion, and partially to its original built-in lexibility, which has allowed researchers to try out different alternate approaches, depending on the scope and nature of the encountered issues. Furthermore, over the past decade, there has been a veritable explosion of stud- ies on the applicability of network-based models for linguistic classiication. Such 1 Certain embryonic forms of the same method can be traced to even earlier times (Hymes 1973), but only Swadesh may be credited for developing and popularizing a fully formalized ap- proach.