Number Sense Disambiguation for Yorùbá Language using Multi-Layered Perceptron

Numbers are an important part of any language and they exist in different formats especially numbers represented using the Hindu-Arabic symbols. In the Yorùbá text, some of the formats of usage are cardinal, ordinal, year, phone number, date, and currency as covered in this study. This study offers to build a model to disambiguate these number senses or to differentiate them when they are found within Yoruba text. The disambiguation of the sense of these formats of usage is very important for effective text-to-speech synthesis. Yorùbá data text with numbers were collected and a Multilayered Perceptron model was trained and tested. 431 text cases were used because the Yorùbá language is still a low-resource language even though it has a reasonable number of speakers. The model has an accuracy of 83.99%, precision of 83.52%, and recall of 83.99% and F1-Score of 83.63%.