Rule-based conversion of closely-related languages: a Dutch-to-Afrikaans convertor

Abstract

For fast-tracking the development of resources for resource-scarce languages, one could transfer existing technologies from one language to another well-sourced, closely-related language. In this contribution, the authors describe the development and performance of a rule-based Dutch-to-Afrikaans converter, with the aim to transform Dutch text so that it looks more like an Afrikaans text (even though it might not even be a good Dutch translation). The rules we used is based on systematic orthographic, morphosyntactic and lexical differences between the two languages. The authors report on an accuracy of 71% on word-level, after minor optimisation with regard to iteration of rules. In a small-scale evaluation on running text, the authors obtain a BLEU score of 0.2519. They conclude that such a rule-based approach to conversion of closely-related languages holds much promise, with potential application in technology transfer (or even machine translation) between such languages.