IUST_NLP at SemEval-2023 Task 10: Explainable Detecting Sexism with Transformers and Task-adaptive Pretraining

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 50–57 July 13-14, 2023 ©2023 Association for Computational Linguistics HausaNLP at SemEval-2023 Task 12: Leveraging African Low Resource TweetData for Sentiment Analysis Saheed Abdullahi Salahudeen1, Falalu Ibrahim Lawan1, Ahmad Mustapha Wali2, Amina Abubakar Imam3, Aliyu Rabiu Shuaibu4, Aliyu Yusuf5, Nur Bala Rabiu6, Musa Bello1, Shamsuddeen Umaru Adamu1, Saminu Mohammad Aliyu6, Murja Sani Gadanya6, Sanah Abdullahi Muaz6, Mahmoud Said Ahmad, Abdulkadir Abdullahi7, Abdulmalik Yusuf Jamoh8 ∀HausaNLP, 1Kaduna State University, 2University of Bucharest, 3University of Abuja, 4Nile University, 5Universiti Teknologi PETRONAS, 6Bayero University Kano, 7Shehu Shagari College of Education, 8Ahmadu Bello University, Zaria a.salahudeen@kasu.edu.ng Abstract We present the findings of SemEval-2023 Task 12, a shared task on sentiment analysis for low-resource African languages using Twitter dataset. The task featured three subtasks; subtask A is monolingual sentiment classifica- tion with 12 tracks which are all monolingual languages, subtask B is multilingual sentiment classification using the tracks in subtask A and subtask C is a zero-shot sentiment classi- fication. We present the results and findings of subtask A, subtask B and subtask C. We also release the code on github. Our goal is to leverage low-resource tweet data using pre- trained Afro-xlmr-large, AfriBERTa-Large, Bert-base-arabic-camelbert-da-sentiment (Arabic-camelbert), Multilingual-BERT (mBERT) and BERT models for sentiment analysis of 14 African languages. The datasets for these subtasks consists of a gold standard multi-class labeled Twitter datasets from these languages. Our results demonstrate that Afro- xlmr-large model performed better compared to the other models in most of the languages datasets. Similarly, Nigerian languages: Hausa, Igbo, and Yoruba achieved better performance compared to other languages and this can be attributed to the higher volume of data present in the languages. 1 Introduction Social media offers an opinionated platform of data content on many topics of interest, such as product reviews, feedback on purchases or ser- vices, political interest, etc., that is dynamically created by users in different languages. Sentiment analysis in many languages is required due to the need for effective classification of these contents. However, the majority of research on sentiment analysis has been done in high-resource languages (Jha et al., 2016; Vilares et al., 2016; Heikal et al., 2018; Elhadad et al., 2019; Abdullahi et al., 2021; Carvalho and Plastino, 2021) while several low- resource African languages receive little attention in Natural Language Processing (NLP) applica- tion due to insufficient of publicly available data. Although, recently, some considerable efforts are made for the development of sentiment analysis for some low-resource African languages (Devlin et al., 2018; Yimam et al., 2020; Ogueji et al., 2021; Abubakar et al., 2021; Muhammad et al., 2022), nonetheless, sentiment analysis for low- resourced African languages still is a misrepre- sented research area. The AfriSenti1 shared task 12 (Muhammad et al., 2023b) aims at building sentiment analy- sis for 14 low-resource African languages using a Twitter dataset which include Hausa, Yoruba, Igbo, Nigerian Pidgin from Nigeria, Amharic, Tigrinya, and Oromo from Ethiopia, Swahili from Kenya and Tanzania, Algerian Arabic dialect from Algeria, Kinyarwanda …

Content truncated to 500 words