LogSigma at SemEval-2026 Task 3: Uncertainty-Weighted Multitask Learning for Dimensional Aspect-Based Sentiment Analysis

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

Summary

LogSigma is a system developed for SemEval-2026 Task 3, focusing on Dimensional Aspect-Based Sentiment Analysis (DimABSA). Unlike traditional ABSA, DimABSA requires predicting continuous Valence and Arousal (VA) scores on a 1–9 scale, presenting a challenge due to varying prediction difficulty across languages and domains. LogSigma addresses this by employing learned homoscedastic uncertainty, where the model automatically balances regression objectives during training using task-specific log-variance parameters (log σ²). The system integrates language-specific encoders and multi-seed ensembling. LogSigma secured 1st place on five datasets across both tracks. A key finding was that learned variance weights differed substantially by language, ranging from 0.66× for German to 2.18× for English, indicating that optimal task balancing is language-dependent and cannot be determined in advance.

Key takeaway

For NLP engineers developing multilingual sentiment analysis models, particularly for continuous VA scores, you should integrate dynamic task weighting. Relying on fixed objective balancing is suboptimal, as task difficulty varies significantly across languages, with learned variance weights ranging from 0.66× to 2.18×. Implement homoscedastic uncertainty learning to automatically adjust task contributions, improving model performance and adaptability across diverse linguistic contexts.

Key insights

Optimal task balancing in multitask learning for continuous sentiment prediction is language-dependent and requires dynamic weighting.

Principles

• Task difficulty varies by language and domain.
• Dynamic task weighting improves multitask learning.
• Homoscedastic uncertainty can balance objectives.

Method

The system learns task-specific log-variance parameters (log σ²) to automatically balance Valence and Arousal regression objectives during training, combined with language-specific encoders and multi-seed ensembling.

In practice

• Apply uncertainty weighting to balance regression tasks.
• Use language-specific encoders for multilingual tasks.
• Employ multi-seed ensembling for robust results.

Topics

• Dimensional Aspect-Based Sentiment Analysis
• Multitask Learning
• Homoscedastic Uncertainty
• Valence Arousal Scores
• SemEval-2026
• Natural Language Processing

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

Related on AIssential

• Consistency Analysis of Sentiment Predictions using Syntactic & Semantic Context Assessment Summarization (SSAS) — SSAS improves LLM sentiment prediction consistency by pre-processing noisy data into structured, high-signal prompts.
• A Multitask Transformer for Offensive Language Detection and Target Identification in HateBR — Multitask learning improves offensive language detection consistency but may reduce fine-grained target attribution.
• When Surface Form Changes Moderation Decisions: A Paired Study of Code-Mixed Workflow Instability — Code-mixed inputs cause significant moderation workflow instability, increasing review burden and non-hate false-flags.
• Discovering Semantic Latent Structures in Psychological Scales: A Response-Free Pathway to Efficient Simplification — Semantic analysis of questionnaire items can efficiently simplify psychological scales without requiring respondent data.
• 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.
• Beyond Sentiment Classification: A Generative Framework for Emotion Intensity Evaluation in Text — Generative LLMs fine-tuned on continuous emotion intensity scores outperform classification and generalize to unseen affective dimensions.

Open in AIssential →

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