Phoneme-Based English-Yorùbá Machine Transliteration

One of the challenges in translating English to Yorùbá language is the foreign names and technical terms used in news articles and scientific documents. Many of these names and terms contain letters not used in the orthography of Yorùbá language. We present a rule-based model for the transliteration of English noun words to Yorùbá such that the output respects the morphology and phonology of the target language. The model, which is phoneme-based, relies on the CMU pronouncing dictionary to derive the phoneme for each word. After the implementation of the model, evaluation of the system developed on standardized test set of 55 words yielded an accuracy of 72.7%, recall of 0.98, precision of 0.965 and F score of 0.972. A second test which included works with non-standardized set references has accuracy of 40.7%, recall of 0.91, precision of 0.925 and F score of 0.912. The low accuracy is an indication of several violation of standard Yorùbá orthography found in this bigger set. A few challenges identified with the model include inability to correctly render some of the vowels as required by the phonology of the target language.