EduNLP at BEA 2026 Shared Task 1: Multi-Model Ensemble with Feature-Augmented Transformers for Vocabulary Difficulty Prediction

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

Summary

The EduNLP system, submitted to the BEA 2026 Shared Task on Vocabulary Difficulty Prediction for English Learners, employs a multi-model ensemble combining handcrafted linguistic features with fine-tuned XLM-RoBERTa transformers. This approach participated in both the closed and open tracks of the competition. The system significantly outperformed baselines across all three L1s, achieving best RMSEs of 1.058 in the closed track (for Chinese L1) and 0.992 in the open track (for Chinese L1). A post-hoc error analysis identified polysemous words used in rare senses and nominalized -ing forms as the primary failure modes for the model. This work was presented at the 21st Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2026) in San Diego, California.

Key takeaway

For NLP engineers developing educational tools, particularly for English language learning, you should consider augmenting transformer models like XLM-RoBERTa with handcrafted linguistic features. This ensemble approach demonstrably improves vocabulary difficulty prediction, as shown by RMSEs of 1.058 and 0.992. Focus your error analysis on polysemous words in rare senses and nominalized -ing forms to refine model performance and address specific learner challenges.

Key insights

Combining linguistic features with fine-tuned XLM-RoBERTa transformers in an ensemble significantly improves vocabulary difficulty prediction.

Principles

• Ensemble models enhance prediction accuracy.
• Linguistic features augment transformer performance.
• Error analysis reveals specific failure modes.

Method

A multi-model ensemble integrates handcrafted linguistic features with fine-tuned XLM-RoBERTa transformers for vocabulary difficulty prediction, evaluated on closed and open tracks.

In practice

• Integrate linguistic features into transformer models.
• Employ ensemble methods for robust NLP tasks.
• Analyze errors for polysemy and nominalized forms.

Topics

• Vocabulary Difficulty Prediction
• Multi-Model Ensemble
• XLM-RoBERTa Transformers
• Linguistic Features
• Natural Language Processing
• Educational Applications

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• 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.
• What Makes Words Hard? Sakura at BEA 2026 Shared Task on Vocabulary Difficulty Prediction — LLM fine-tuning with soft-target loss effectively predicts vocabulary difficulty while explainable models offer insights.
• Use of Machine Learning techniques and Large Language Models for automatic evaluation of Celpe-Bras exam texts — BERT adaptations achieved the best automatic evaluation for Celpe-Bras texts, but at a higher computational cost.
• Exploring Data Augmentation and Resampling Strategies for Transformer-Based Models to Address Class Imbalance in AI Scoring of Scientific Explanations in NGSS Classroom — Targeted data augmentation significantly improves transformer-based model performance in automated scoring of imbalanced educational text data.
• A Scalable Tool for Measuring Manner and Result Verbs in Developmental Language Research — A RoBERTa classifier, trained on LLM-generated annotations, accurately distinguishes manner and result verbs at scale.
• Transformer models for Urdu Language — Urdu's low-resource status in AI is improving through Transformer models and increased data development.

Open in AIssential →

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