Conditional Entropy Measures Intelligibility among Related Languages

Abstract

The Scandinavian languages are so alike that their speakers often communicate, each using their own language, which Haugen (1966) dubbed SEMICOMMUNICATION. The success of semi-communication depends on the languages involved, and, moreover, can be asymmetric: for example, Swedish is more easily understandable for a Dane, than Danish for a Swede. It has been argued that non-linguistic factors could explain intelligibility, including its asymmetry. Gooskens (2006), however, found a high correlation between linguistic distance and intelligibility. This suggests that we need to seek linguistic factors that influence intelligibility, and that potentially asymmetric factors would be particularly interesting. Gooskens' distance techniques cannot capture asymmetry. The present paper attempts to develop a model of the success of semi-communication based on conditional entropy, in particular using the conditional entropy of the phonememapping in corresponding (cognate) words. Semantically corresponding words were taken from frequency lists and aligned, and the conditional entropy of the phoneme mapping in aligned word pairs was calculated. This gives us information about the difficulty of predicting a phoneme in a native language given a corresponding phoneme in the foreign language. We also examine the conditional entropy of selected word classes, such as native/loan and function/content words.