The predictability of name pronunciation errors in four South African languages

Abstract

Personal names are often pronounced in very different ways depending on the language background of the speaker. We seek to determine whether some of these pronunciations 'errors' are systematic and if so, in which ways. Specifically, we analyze some of the typical errors made by speakers from four South African languages (Setswana, English, isiZulu) when producing names from the same four languages. We compare these results with the pronunciations generated by four language-specific grapheme-to-phoneme (G2P) predictors trained on generic words from the four languages. We find that the G2P predictors are able to predict at least some of the typical errors humans make and, in fact, that these errors are slightly more predictable than the correct pronunciations themselves.