Bantu Machine Translation Proof-of-Concept Using ciNyanja and ciYawo

Bantu machine translation language pairs are few and far in-between. This study succeeds in adding the ciNyanja-ciYawo pair to the list. A bootstrap approach is demonstrated that identifies the minimum required language data – both in terms of lexical items and grammatical rules – for use in building an Apertium rule-based machine translation package. A limited lexicon is used in this preliminary study in order to demonstrate a proof-of-concept. Grammatical categories are also restricted. Grammatical scope is further limited in tense, aspect and mood. Despite these reduced scopes, translation quality is evaluated and a first WER score for this Bantu language translation pair is presented. A benchmark evaluation score will, however, only be available at the end of the full study, which will have less restrictions in scope.