akniele/sentiment_classification_Nigerian_Pidgin

Project on zero-shot sentiment classification for the low-resource language Nigerian Pidgin # Zero-shot sentiment classification for Nigerian Pidgin Tweets This is a project on sentiment classification. A system description paper on the project, written by me and my professor, was published in the SemEval 2023 proceedings. Here is a link to the paper. The aim of the project is to investigate how using different languages for pre-training and fine-tuning affects the performance of zero-shot sentiment analysis for Twitter data in Nigerian Pidgin. Nigerian Pidgin is an English-based Creole language. As a proficient speaker of English, one can easily recognize some of the words: Nigerian Pidgin is considered low-resource, which is why sentiment classification is still a challenging task for this language, despite being considered practically solved for high-resource languages such as English. In this project, I use two pre-trained language models: bert-base-uncased (Devlin et al., 2019) and bert-base-multilingual-uncased (Devlin et al., 2019). I then fine-tuned the models separately on three different languages (or language pair in one case), to see how each language would affect the performance on sentiment classification in Nigerian Pidgin. The three languages are English Nigerian Pidgin Igbo & Hausa Note that the setup fine-tuning on Nigerian Pidgin is not zero-shot. It is included as a baseline. The data for Nigerian Pidgin, Igbo and Hausa is taken from Muhammad et al. (2023), while the data for English is provided by (Rosenthal et al., 2017). Igbo and Hausa are both Nigerian languages, but unlike English, they share no close linguistic ties with Nigerian Pidgin. ## References Shamsuddeen Hassan Muhammad, David Adelani, Sebastian Ruder, Ibrahim Sa’id Ahmad, Idris Abdulmumin, Shehu Bello Bello, Monojit Choudhury, Chris Chinenye Emezue, Saheed Salahuddeen Abdullahi, Anuoluwapo Aremu, Alipio Jeorge, and Pavel Brazdil. 2022. Naijasenti: A Nigerian Twitter Sentiment Corpus for Multilingual Sentiment Analysis. In Proceedings of the 13th Langu …