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

Abstract

To test the capacity of the Levenshtein algorithm to predict intelligibility & perceived linguistic distance between closely related language varieties, nonnormalized & normalized Levenshtein distances were calculated between Standard Danish & 17 other Scandinavian varieties, including Standard Swedish, Norwegian, & Faroese & representative regional dialects of Swedish, Norwegian, & Danish. Calculations were based on cognate pairs among 58 transcribed words of a text translated into each of the varieties under study & read aloud by native speakers of the variety; nonnormalized values consisted of the raw number of differences between symbols within cognate pairs, & normalized values represent %s of alignment slots with differences within pairs. Native speakers of Standard Danish (N = 351, aged 15-19) listened to the recordings of the text in other varieties, translated them into Standard Danish as an intelligibility test, & gave subjective ratings of the linguistic distance between Standard Danish & each other variety. Results indicate that normalized Levenshtein distances are highly predictive of intelligibility & have a significantly lower correlation with perceived distances; the difference between intelligibility & perceived distance correlations is not significant for the nonnormalized Levenshtein distances. J. Hitchcock