McMaster NLP at SemEval-2026 Task 2: A Lightweight Multi-Feature System for Predicting Emotional Valence and Arousal over Time

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

Summary

McMaster NLP developed a lightweight, feature-based regression system for SemEval-2026 Task 2, designed to predict emotional valence (pleasantness) and arousal (activation) from longitudinal language data. This system processes diverse language inputs, from ecological essays to short affect-words, organized by user and time. Its approach integrates four distinct signals: sentence-level semantic embeddings, psycholinguistic category features, similarity measures with archetypal sentences, and trainable user-embeddings to account for individual differences. These combined features are fed into a multi-layer perceptron for joint prediction of valence and arousal. The system offers an interpretable baseline, enabling researchers to isolate the impact of semantic, psycholinguistic, similarity, and user-specific signals, with further analysis identifying informative feature groups and error patterns.

Key takeaway

For NLP Engineers developing affective computing models, this research highlights the value of a multi-feature, interpretable approach. You should consider integrating diverse linguistic signals, including psycholinguistic features and user-specific embeddings, to enhance the accuracy and explainability of valence and arousal predictions. This method provides a robust baseline for understanding feature contributions and concentrating error analysis.

Key insights

A lightweight, multi-feature system predicts emotional valence and arousal from longitudinal language data using diverse signals.

Principles

• Combining diverse linguistic signals enhances emotional prediction.
• User-specific embeddings improve cross-user affective modeling.
• Interpretable baselines clarify feature contributions.

Method

The system combines sentence-level semantic embeddings, psycholinguistic features, archetypal sentence similarity, and trainable user-embeddings into a feature vector. This vector is then processed by a multi-layer perceptron to jointly predict valence and arousal.

In practice

• Analyze feature group informativeness for error reduction.
• Isolate signal contributions for model interpretability.

Topics

• Emotional Valence Prediction
• Arousal Prediction
• Longitudinal Language Data
• Multi-Feature Systems
• Psycholinguistic Features
• Semantic Embeddings

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• From Affect Prediction to Affect Forecasting: Evidence for Distinct Information Sources in Longitudinal Text — Current affect prediction relies on text semantics, while future affective change forecasting benefits from numeric trajectory dynamics.
• 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.
• A Chinese Education Broadcast Emotion Corpus — A new Chinese education broadcast emotion corpus aids emotion recognition research.
• Moving Past Engagement Surveys to Truly Understand How Employees Feel at Work — Analyzing open-ended employee text with encoder-based transformers offers a more accurate and transparent alternative to traditional engagement surveys.
• Emotion Isn’t Universal: Rethinking Sentiment Analysis for a Global Customer Base — Cultural context is critical for accurate sentiment analysis, as emotion expression varies globally.
• Exploration of symbolic methods for emotion detection for Portuguese — Symbolic emotion detection in Portuguese shows stable relative performance despite domain-specific absolute variations.

Open in AIssential →

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