Beyond Pretraining Bias: Evaluating Language-Adaptive Fine-Tuning Strategies for Sentiment and Topic Classification in Three Nigerian Languages
Domaine:
natural language processingType de record:
paper
Créateur:
Olo
Éditeur:
Cen
Éditeur:
OSF
Hôte:
This paper presents a controlled empirical study comparing four adaptation strategies — zero-shot prompting, standard fine-tuning, language-adaptive fine-tuning (LAFT), and ensemble decoding — applied across sentiment classification and news-topic classification in Hausa, Yorùbá, and Igbo. Using AfriSenti, NaijaSenti, and MasakhaNEWS benchmarks with AfriBERTa-large and mBERT as base models, we find LAFT with curated monolingual pretraining data produces consistent F1 gains of 4.2–9.8 percentage points over standard fine-tuning. We document failure modes hidden by aggregate metrics: tonal morphology confusion in Yorùbá, dialectal token misalignment in Hausa, and orthographic inconsistency in Igbo.
Similaires
Multilingual language model Adaptive Fine-Tuning: A Study on African LanguagesUIO at SemEval-2023 Task 12: Multilingual fine-tuning for sentiment classification in low-resource languagesInvestigating the Impact of Language-Adaptive Fine-Tuning on Sentiment Analysis in Hausa Language Using AfriBERTaDomain-Adaptive Pre-Training for Arabic Aspect-Based Sentiment Analysis: A Comparative Study of Domain Adaptation and Fine-Tuning StrategiesAdapting Pre-trained Language Models to African Languages via Multilingual Adaptive Fine-TuningMULTILINGUAL ADAPTIVE FINE-TUNING (MAFT)