Sagarmatha at SemEval-2026 Task 9: Heterogeneous Ensembling and Hierarchical Task Conditioning for Multilingual Latent Distributional Divergence Modeling

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

Summary

The Sagarmatha system, developed for SemEval-2026 Task 9, addresses affective polarization by identifying discursive strategies such as dehumanization and vilification across 22 languages. This system employs a heterogeneous ensemble architecture that integrates mDeBERTa-v3, ReMBERT, LaBSE, mmBERT, and XLM-RoBERTa. Its design features two primary architectural pillars: learnable weighted layer pooling and hierarchical task conditioning. While its final submission (R3) demonstrated high leaderboard stability, the primary configuration (Weighted Polyglot, R1) achieved superior performance in complex multi-label tasks. Sagarmatha ranked 1st globally in English and Hausa manifestation identification, and 1st in Telugu detection, securing 2nd in categorization. All code and resources are publicly available.

Key takeaway

For NLP Engineers developing multilingual systems to detect nuanced discursive strategies, Sagarmatha's approach offers a robust blueprint. You should consider integrating heterogeneous transformer models like mDeBERTa-v3 and XLM-RoBERTa with learnable weighted layer pooling. Implementing hierarchical task conditioning can significantly boost performance on complex multi-label classification, as demonstrated by its top rankings in SemEval-2026 Task 9.

Key insights

Sagarmatha combines heterogeneous ensembling and hierarchical task conditioning for superior multilingual discursive strategy identification.

Principles

• Heterogeneous ensembling improves multilingual NLP.
• Hierarchical task conditioning enhances multi-label performance.
• Weighted layer pooling optimizes model integration.

Method

Sagarmatha integrates mDeBERTa-v3, ReMBERT, LaBSE, mmBERT, and XLM-RoBERTa using learnable weighted layer pooling and hierarchical task conditioning to identify discursive strategies across 22 languages.

In practice

• Combine diverse transformer models for robustness.
• Apply weighted layer pooling for ensemble integration.
• Use hierarchical conditioning for complex multi-label tasks.

Topics

• Multilingual NLP
• Ensemble Learning
• Transformer Models
• SemEval-2026 Task 9
• Affective Polarization
• Discursive Strategies

Code references

• SUJAL390/SagarmathaatSemevaltask9

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Global vs. Local Sentence Embeddings for Brazilian Portuguese: Revisiting Monolingual Models in the Age of Foundation Models — Optimal model choice for PT-BR sentence embeddings balances vocabulary coverage and adaptation flexibility.
• Comparing LLM and Fine-Tuned Model Performance on NVDRS Circumstance Extraction with Varying Prompt Complexity — LLMs, guided by a context-aware "Complexity Score" for prompt selection, significantly outperform fine-tuned models on inferentially complex, low-prevalence data extraction.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification — Iterative imbalance-aware fine-tuning of LLMs, combined with augmentation and post-processing, significantly improves rare-class NLP performance.
• Optimal Splitting of Language Models from Mixtures to Specialized Domains — Optimizing compute allocation between general pretraining and specialized continued pretraining improves language model performance.

Open in AIssential →

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