GHaLIB: A Multilingual Framework for Hope Speech Detection in Low-Resource Languages

TL;DR

Hope speech detection remains underexplored in NLP, especially for Urdu and other low-resource languages. The authors introduce GHaLIB, a multilingual framework that combines language-specific encoders with a multilingual backbone to detect hope speech across Urdu, English, German, and Spanish, using weighted training and Optuna-based hyperparameter tuning. The model achieves strong Urdu results (Urdu binary F1 of 95.2% and multiclass F1 of 65.2% on PolyHope-M 2025) and competitive cross-language performance, demonstrating the value of language-adaptive components in multilingual settings. By releasing open-source code and resources, the work provides a practical, scalable path toward inclusive hope-speech tooling in low-resource environments.

Abstract

Hope speech has been relatively underrepresented in Natural Language Processing (NLP). Current studies are largely focused on English, which has resulted in a lack of resources for low-resource languages such as Urdu. As a result, the creation of tools that facilitate positive online communication remains limited. Although transformer-based architectures have proven to be effective in detecting hate and offensive speech, little has been done to apply them to hope speech or, more generally, to test them across a variety of linguistic settings. This paper presents a multilingual framework for hope speech detection with a focus on Urdu. Using pretrained transformer models such as XLM-RoBERTa, mBERT, EuroBERT, and UrduBERT, we apply simple preprocessing and train classifiers for improved results. Evaluations on the PolyHope-M 2025 benchmark demonstrate strong performance, achieving F1-scores of 95.2% for Urdu binary classification and 65.2% for Urdu multi-class classification, with similarly competitive results in Spanish, German, and English. These results highlight the possibility of implementing existing multilingual models in low-resource environments, thus making it easier to identify hope speech and helping to build a more constructive digital discourse.

Figures (8)

• Figure 1: Uncertain prediction caused by shared wording in hope sentiments.
• Figure 2: English dataset label distribution
• Figure 3: Urdu dataset label distribution
• Figure 4: Average text length per sentiment category.
• Figure 5: Average word count per sentiment category.