COLLECTED BY : NLPGhana AUTHORS : Paul Azunre, Lawrence Adu-Gyamfi, Esther Appiah, Felix Akwerh, Salomey Osei, Cynthia Amoaba, Salomey Afua Addo, Edwin Buabeng-Munkoh, Nana Boateng, Franklin Adjei, Bernard Adabankah CONTACT : natural.language.processing.gh@gmail.com WEBSITE : https://GhanaNLP.org DATASET DESCRIPTION This dataset is bilingual machine translation training corpus for English and Akuapem Twi of 25,421 sentence pairs. A transformer-based machine translator was used to generate initial translations in Akuapem Twi, which were later verified and corrected where necessary by native speakers. The main idea of a typical use case for the dataset is for further training of machine translation models in Akuapem Twi. The data can also be used for other downstream NLP tasks such as Named Entity Recognition and POS tagging, with appropriate additional annotations. Another potential application is training unsupervised embeddings for the Akuapem Twi language. In addition a higher quality 697 crowdsourced sentences are provided for use as an evaluation set for the tasks highlighted above. It is recommended as a testing dataset for machine translation English to Twi and Twi to English models. _________________________________________________________________________________________________ ------------------------------------------------------------------------------------------------- Human Verified Corpus | ------------------------------------------------------------------------------------------------- | sentence count | word count | unique word count | ------------------------------------------------------------------------------------------------- English | 25421 | 163861 | 13058 | ------------------------------------------------------------------------------------------------- Akuapem Twi | 25421 | 170908 | 12314 | _________________________________________________________________________________________________ ------------------------------------------------------------------------------------------------- Crowdsourced data Corpus | ------------------------------------------------------------------------------------------------- English | 697 | 3525 | 1225 | ------------------------------------------------------------------------------------------------- Akuapem Twi | 697 | 6702 | 2541 | _________________________________________________________________________________________________