On detecting borrowing: Distance-based and character-based approaches

Two computational methods for detecting borrowing among a family of genetically related languages are proposed. One method, based on the detection of branches with negative length in lexicostatistical trees, is shown to work poorly. As we demonstrate, this method is similar to another recently proposed method for detecting borrowing based on skewing in lexicostatistical data. A second method, using character-based classification techniques in common use in the classification of biological taxa, is shown to be more effective. This method allows borrowed characters and the languages among which the borrowing may have taken place to be identified — in some cases, the most likely direction of the borrowing can also be specified.