UM6P at SemEval-2023 Task 12: Out-Of-Distribution Generalization Method for African Languages Sentiment Analysis

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1004–1010 July 13-14, 2023 ©2023 Association for Computational Linguistics UM6P at SemEval-2023 Task 12: Out-Of-Distribution Generalization Method for African Languages Sentiment Analysis Abdelkader El Mahdaouy1, Hamza Alami2, Salima Lamsiyah3 and Ismail Berrada2 1Modeling, Simulation and Data Analysis, Mohammed VI Polytechnic University, Morocco 2School of Computer Science, Mohammed VI Polytechnic University, Morocco 3Dept. of Computer Science, Faculty of Science, Technology and Medicine, University of Luxembourg, Luxembourg {firstname.lastname}@{um6p.ma1,2, uni.lu3} Abstract This paper presents our submitted system to AfriSenti SemEval-2023 Task 12: Sentiment Analysis for African Languages. The AfriSenti consists of three different tasks, covering mono- lingual, multilingual, and zero-shot sentiment analysis scenarios for African languages. To improve model generalization, we have ex- plored the following steps: 1) further pre- training of the AfroXLM Pre-trained Language Model (PLM), 2) combining AfroXLM and MARBERT PLMs using a residual layer, and 3) studying the impact of metric learning and two out-of-distribution generalization training objectives. The overall evaluation results show that our system has achieved promising results on several sub-tasks of Task A. For Tasks B and C, our system is ranked among the top six participating systems. 1 Introduction The widespread use of the internet and social me- dia platforms has enabled billions of users world- wide to communicate with each other, express their opinions, and share their experiences. This has led to the proliferation of content from various spoken languages and dialects. To deal with the huge amount of available textual corpora on the Web, various Natural Language Processing (NLP) tools and applications have been proposed. How- ever, most of these NLP tools and applications have been developed for high-resource languages where both labeled and unlabeled data are highly available, while low-resource languages, such as African languages and dialects, still suffering from data scarcity (Nekoto et al., 2020; Marivate et al., 2020). Hence, there is a large gap between what has been achieved for high-resource languages and low- resource languages in NLP (Wang et al., 2021b; Ogueji et al., 2021; Alabi et al., 2022; Adebara and Abdul-Mageed, 2022; Adebara et al., 2022a). In recent years, there has been an increasing in- terest in NLP for African languages and dialects. On the one hand, various research works have been introduced to leverage existing unlabeled data for training or adapting existing multilingual lan- guage models to African languages and dialects (Ogueji et al., 2021; Alabi et al., 2022; Adebara et al., 2022b). On the other hand, several studies have been published on collecting, curating, and building labeled resources and corpora for African languages. These studies have tackled several NLP applications such as machine translation (Emezue and Dossou, 2021), named entity recognition (Ade- lani et al., 2021), language and dialect identifica- tion (Adebara et al., 2022a), and sentiment analysis (El Mahdaouy et al., 2021; El Mekki et al., 2021; Mabokela and Schlippe, 2022; Muhammad et al., 2022). Nevertheless, the existing studies remain limited to a few African languages and dialects (Adebara and Abdul-Mageed, 2022; Adebara …

Content truncated to 500 words