Predicting intelligibility and perceived linguistic distance by means of the Levenshtein algorithm

In this article, we investigate the predictive value of so-called Levenshtein distances for both intelligibility scores and perceived linguistic distances. Additionally, we compare two measuring methods, namely normalised and non-normalised Levenshtein distances. The Levenshtein algorithm is a string edit distance measure that quantifies the distance between the pronunciations of corresponding words in different dialects or closely related languages. It calculates the minimal costs required to change a string of segments into another by means of insertions, deletions or substitutions. Kessler (1995) introduced the algorithm for measuring distances between Irish Gaelic dialects. Since then it has been applied successfully to Dutch dialects (Heeringa 2004, 213–278), Sardinian dialects (Bolognesi & Heeringa 2002) and German dialects (Nerbonne & Siedle 2005).