Multilingual Multi-Label Emotion Classification at Scale with Synthetic Data

2026-04-14

AI Summary

This study addresses the scarcity of annotated data in multilingual emotion classification, where existing corpora are predominantly English, single-label, and limited in linguistic coverage. The authors propose a culturally adapted generative approach combined with procedural quality filtering to construct, for the first time, a million-scale multilingual, multi-label synthetic dataset spanning 23 languages and 11 emotion categories—without requiring manual annotation. Multilingual Transformer models (from DistilBERT to XLM-R-Large) trained on this dataset achieve strong in-domain performance, with 0.868 F1-micro and 0.987 AUC-micro. In zero-shot transfer evaluations on GoEmotions and SemEval-2018 benchmarks, the models attain an AUC-micro of 0.810, rivaling or surpassing English-only counterparts. The best-performing base model has been publicly released.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Emotional Intelligence

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Emotion classification in multilingual settings remains constrained by the scarcity of annotated data: existing corpora are predominantly English, single-label, and cover few languages. We address this gap by constructing a large-scale synthetic training corpus of over 1M multi-label samples (50k per language) across 23 languages: Arabic, Bengali, Dutch, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Mandarin, Polish, Portuguese, Punjabi, Russian, Spanish, Swahili, Tamil, Turkish, Ukrainian, Urdu, and Vietnamese, covering 11 emotion categories using culturally-adapted generation and programmatic quality filtering. We train and compare six multilingual transformer encoders, from DistilBERT (135M parameters) to XLM-R-Large (560M parameters), under identical conditions. On our in-domain test set, XLM-R-Large achieves 0.868 F1-micro and 0.987 AUC-micro. To validate against human-annotated data, we evaluate all models zero-shot on GoEmotions (English) and SemEval-2018 Task 1 E-c (English, Arabic, Spanish). On threshold-free ranking metrics, XLM-R-Large matches or exceeds English-only specialist models, tying on AP-micro (0.636) and LRAP (0.804) while surpassing on AUC-micro (0.810 vs. 0.787), while natively supporting all 23 languages. The best base-sized model is publicly available at https://huggingface.co/tabularisai/multilingual-emotion-classification

Problem

• multilingual
• multi-label
• emotion classification
• data scarcity
• synthetic data

Innovation

• synthetic data
• multilingual emotion classification
• multi-label learning
• culturally-adapted generation
• zero-shot transfer