ICT-NLP at SemEval-2026 Task 3: Less Is More — Multilingual Encoder with Joint Training and Adaptive Ensemble for Dimensional Aspect Sentiment Regression

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

Summary

ICT-NLP's system for SemEval-2026 Task 3 Track A Subtask 1 addresses Dimensional Aspect Sentiment Regression (DimASR) with a lightweight, resource-efficient approach. It relies solely on multilingual pre-trained encoders, avoiding large language models or external corpora. Key techniques include joint multilingual and multi-domain training to enhance cross-lingual transfer and mitigate data sparsity, a bounded regression transformation for improved training stability and constrained predictions, and an adaptive ensemble strategy via subset search to reduce prediction variance. The system demonstrated strong, consistent performance, securing 1st place on zho-res, 2nd on zho-lap, and 3rd on jpn-hot, with all other datasets ranking within the top half of participating teams.

Key takeaway

For NLP engineers developing sentiment regression systems, this work demonstrates that resource-efficient multilingual encoders can achieve top-tier performance without relying on LLMs. You should consider implementing joint multilingual/multi-domain training and bounded regression transformations to enhance stability and cross-lingual transfer. An adaptive ensemble strategy can further reduce prediction variance, offering a robust alternative to larger, more complex models.

Key insights

Lightweight multilingual encoders with joint training and adaptive ensemble achieve strong dimensional aspect sentiment regression.

Principles

• Joint multilingual/multi-domain training aids cross-lingual transfer.
• Bounded regression improves training stability and constrains predictions.
• Adaptive ensemble via subset search reduces prediction variance.

Method

The system uses multilingual pre-trained encoders, applies joint multilingual and multi-domain training, incorporates a bounded regression transformation, and employs an adaptive ensemble strategy via subset search.

In practice

• Utilize joint training for cross-lingual data sparsity.
• Implement bounded regression for stable training and valid output.
• Apply adaptive ensemble to reduce prediction variance.

Topics

• SemEval-2026
• Dimensional Aspect Sentiment Regression
• Multilingual Encoders
• Joint Training
• Adaptive Ensemble
• Cross-lingual Transfer

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

Related on AIssential

• 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.
• Beyond Polarity: Multi-Dimensional LLM Sentiment Signals for WTI Crude Oil Futures Return Prediction — Multi-dimensional LLM sentiment, beyond polarity, improves WTI crude oil futures return prediction.
• From TF-IDF to Transformers: A Comparative and Ensemble Approach to Sentiment Classification — A soft voting ensemble of diverse machine learning and transformer models significantly enhances sentiment classification accuracy, particularly for nuanced text.
• F2LLM-v2: Inclusive, Performant, and Efficient Embeddings for a Multilingual World — F2LLM-v2 offers efficient, performant, and inclusive multilingual embeddings for over 200 languages.
• Adaptive Multimodal Sentiment Analysis with Stream-Based Active Learning for Spoken Dialogue Systems — Stream-based active learning enables cost-efficient, personalized multimodal sentiment analysis in dialogue systems by minimizing user queries.
• Telenor Nordics Customer Service self-help corpus — A new, ethically sourced, multilingual customer service self-help corpus for Nordic languages addresses a critical data scarcity for NLP research.

Open in AIssential →

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