MUCS@Dravidianlangtech@ACL2026: Hope Speech Detection in Code-Mixed Tulu Language Using Multiple Features

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Advanced, medium

Summary

Team MUCS developed a system for hope speech detection in code-mixed Tulu, a low-resource language, for the DravidianLangTech@ACL 2026 shared task. Without pretrained language models for Tulu, the system uses multiple hand-crafted features. These include word n-grams (n=1, 3), character n-grams (n=1, 3), syllable n-grams (n=1, 3), and sub-words. These features train two ensembles of classical Machine Learning models. One ensemble combines Multinomial Naive Bayes (MNB) and Logistic Regression (LR) classifiers. The second pairs k Nearest Neighbor (kNN) and Decision Tree (DT) classifiers; both ensembles use soft-voting. Experimental results show feature integration effectively captures lexical, sub-lexical, and phonological cues in noisy code-mixed text. The system achieved competitive performance on development and test datasets. An ablation study also evaluated feature set contributions.

Key takeaway

For NLP Engineers developing solutions for low-resource languages like Tulu, where pretrained models are unavailable, you should prioritize robust feature engineering. Focus on hand-crafted features such as word, character, and syllable n-grams, alongside sub-words. These effectively capture linguistic nuances in code-mixed text. Consider implementing ensemble classical Machine Learning models with soft-voting, like MNB/LR or kNN/DT combinations, to achieve competitive performance. This approach offers a viable path for building effective language processing systems in data-scarce environments.

Key insights

Feature-based ML ensembles effectively detect hope speech in low-resource, code-mixed languages without pretrained models.

Principles

• Feature engineering is crucial for low-resource NLP.
• Ensemble methods enhance classification robustness.
• Sub-lexical cues aid noisy text analysis.

Method

The proposed system uses word, character, syllable n-grams (n=1, 3), and sub-words as features. These train soft-voting ensembles of MNB/LR and kNN/DT classifiers for hope speech detection.

In practice

• Use n-grams for code-mixed text.
• Combine MNB/LR for robust classification.
• Explore ablation studies for feature impact.

Topics

• Hope Speech Detection
• Code-Mixed Language
• Tulu Language
• Low-Resource NLP
• Feature Engineering
• Ensemble Machine Learning
• N-grams

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• The TechBeat: The Complete Guide to Implementing Healthcare Voice Agents (3/22/2026) — The provided content showcases a strong emphasis on artificial intelligence, detailing its diverse applications, development methodologies, and significant market impact.
• Speech-Driven End-to-End Language Discrimination towards Chinese Dialects — Speech-driven features significantly enhance fine-grained language discrimination for similar dialects, outperforming text-based methods.
• TeamHerald@CHIPSAL 2026: Hate Speech Detection and Sentiment Analysis of Nepali Memes using Transformer-based Architectures and Ensemble Learning — Ensemble strategies for meme analysis perform differently for binary vs. multi-class tasks, requiring task-specific aggregation.
• Hard to Be Heard: Phoneme-Level ASR Analysis of Phonologically Complex, Low-Resource Endangered Languages — Data scarcity, not phonological complexity, primarily drives ASR errors in low-resource, typologically complex languages.
• Code Is More Than Text: Uncertainty Estimation for Code Generation — Code generation uncertainty estimation requires specialized methods accounting for code's unique properties, outperforming natural language-derived baselines.
• Robust Language Identification for Romansh Varieties — The paper titled "Robust Language Identification for Romansh Varieties," authored by Charlotte Model, Sina Ahmadi, and Jannis Vamvas, was presented at the 11th Edition of the Swiss Text Analytics Conf…

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.