Code-switched & African Language Sentiment Classification 

Status: Completed |  Funding: Howard University

This set of projects explored sentiment classification in linguistically complex and low-resource contexts, specifically targeting code-switched and African languages. One project focused on classifying sentiment in code-switched text using multilingual embeddings, leading to a publication at NATL and a national conference presentation. A follow-up study introduced a novel translation-based preprocessing approach to enhance transformer model performance on code-switched data, culminating in a publication at CCSCE 2023. 

Additional research was conducted through participation in the AfriSenti-SemEval challenge, benchmarking transformer models across multiple African languages for sentiment analysis. This work produced competitive results and led to the development of a Language-Specific Model Fusion System that intelligently combines outputs from language-specific transformers. The system was detailed in a publication at SemEval-2023, contributing to the advancement of sentiment analysis in low-resource, multilingual settings. 

Contributors

Related Publications

• Sentiment analysis across multiple African languages: A current benchmark 

• Howard university computer science at semeval-2023 task 12: A 2-step system design for multilingual sentiment classification with language identification 

• Sentiment classification of code-switched text using pre-trained multilingual embeddings and segmentation 

• Zero-shot classification reveals potential positive sentiment bias in African languages translations 

• Evaluating Ensembled Transformers for Multilingual Code-Switched Sentiment Analysis 

• Baselining performance for multilingual codeswitching sentiment classification 

• Term frequency features vs transformers: A comparision for sentiment classification of african languages 

Progress Timeline

• 2024-2025