Kafi Noonoo To English Machine Translation Using Deep Learning Approaches

Abstract

Kafi Noonoo is one of the Ethiopian languages that is spoken by the Kaffa people in the southwestern part of Ethiopia. Additionally, it is a morphologically rich language and has an indigenous name for prestige, cultural place, and cultural dejectedness, which has no equivalent meaning in other languages. Machine translation is a technique that automatically translates text’s or speech’s meaning from one language to another without human involvement to resolve information gaps. Various machine translation studies have been conducted for resource-rich languages like English, French, German, and others. However, the variety of linguistic patterns, the dominance of technologically developed languages, and the lack of machine translation from Kafi  Noonoo to English will lead to the disappearance of Kafi Noonoo indigenous words among native speakers. To tackle such a problem, this article designed a Kafi Noonoo to English and vice versa machine translation solution by using deep learning approaches. The bidirectional long short-term memory, bidirectional gated recurrent unit with and without attention, and transformer were applied. In order to train the model, the bilingual parallel sentences were collected from Kafi Noonoo linguistic-related sources. Different experiments were applied to find out the optimal value of the proposed model. Based on the experiment’s result, the transformer performed better with an accuracy of 89% and a BLEU score of 6.34 and 5.42 for Kafi Noonoo to English and English to Kafi Noonoo, respectively. According to our experiment result, the transformer model was suitable for morphologically rich languages like Kafi Noonoo to English and vice versa, for machine translation. For a better result, there is a necessity to generate parallel corpora in order to conduct comparable research.

Keywords: Kaffa, Kafi Noonoo, Low-resource machine Translation, Transformer