Uppsala University at SemEval-2023 Task12: Zero-shot Sentiment Classification for Nigerian Pidgin Tweets

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1491–1497 July 13-14, 2023 ©2023 Association for Computational Linguistics Uppsala University at SemEval-2023 Task12: Zero-shot Sentiment Classification for Nigerian Pidgin Tweets Annika Kniele and Meriem Beloucif Department of Linguistics and Philology Uppsala University annika.kniele.4937@student.uu.se Abstract While sentiment classification has been consid- ered a practically solved task for high-resource languages such as English, the scarcity of data for many languages still makes it a challeng- ing task. The AfriSenti-SemEval shared task aims to classify sentiment on Twitter data for 14 low-resource African languages. In our par- ticipation, we focus on Nigerian Pidgin as the target language. We have investigated the ef- fect of English monolingual and multilingual pre-trained models on the sentiment classifi- cation task for Nigerian Pidgin. Our setup in- cludes zero-shot models (using English, Igbo and Hausa data) and a Nigerian Pidgin fine- tuned model. Our results show that English fine-tuned models perform slightly better than models fine-tuned on other Nigerian languages, which could be explained by the lexical and structural closeness between Nigerian Pidgin and English. The best results were reported on the monolingual Nigerian Pidgin data. The model pre-trained on English and fine-tuned on Nigerian Pidgin was submitted to Task A Track 4 of the AfriSenti-SemEval Shared Task 12, and scored 25 out of 32 in the ranking. 1 Introduction Nigerian Pidgin (pcm) is an English-based Creole language spoken in Nigeria by ca. 40 million na- tive speakers and ca. 80 million second-language speakers (Muhammad et al., 2022). Table 1 shows an example of a Nigerian Pidgin tweet: pcm E don tay wey I don dey crush on this fine woman. . . en I have had a crush on the beautiful woman for a while. . . Table 1: Example of a tweet in Nigerian Pidgin (Muhammad et al., 2022). We note that there are few words that an English speaker can easily understand and distinguish. Despite having a considerable number of speak- ers, it is considered a low-resource language. Due to lexical and structural similarities between En- glish (en) and Nigerian Pidgin (Muhammad et al., 2022), one might assume that English could be a good candidate language to be used in zero-shot experiments for Nigerian Pidgin sentiment classifi- cation. Since English is a high-resource language, this could go some way towards alleviating the resource scarcity problem for Nigerian Pidgin. Besides Nigerian Pidgin, two of the most widely spoken languages in Nigeria are Igbo (ig) and Hausa (ha). The Igbo language is part of the Benue- Congo group of the Niger-Congo languages, and Hausa is one of the Chadic (Afroasiatic) languages. This means that while the two languages are ge- ographically close to Nigerian Pidgin, no close linguistic ties exist between them and Nigerian Pidgin (Muhammad et al., 2022). However, code- switching between the two languages and Nigerian Pidgin, or English, is not uncommon. As an ex- ample, consider Table 2, which gives examples of tweets in Igbo and Hausa, both of which are …

Content truncated to 500 words