SATLab at BEA 2026 Shared Task 1: Predicting the Difficulty of English Words for Three L1 Learners Using Primarily Psycholinguistic Features

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

SATLab participated in the BEA 2026 shared task 1, focusing on predicting English word difficulty for L2 learners. Their system utilized features primarily derived from word frequency lists, lexical norms, and psychometric data, which were then processed by a gradient boosting decision tree model. While the system successfully outperformed the baseline, its performance was notably lower than that of the top-performing teams. The analysis included examining feature contributions through gain scores and Spearman rank correlations, alongside a brief review of the most significant errors encountered by the model.

Key takeaway

For AI Scientists and researchers developing L2 language learning tools, SATLab's approach highlights the value of psycholinguistic features in predicting word difficulty. You should prioritize integrating data from word frequency lists, lexical norms, and psychometric sources into your models. While gradient boosting decision trees offer a solid foundation, a thorough feature contribution analysis is essential to identify and refine the most impactful predictors, aiming to surpass baseline performance and approach top-tier accuracy.

Key insights

Psycholinguistic features are effective for predicting L2 English word difficulty but require careful selection for top performance.

Principles

• Psycholinguistic features enhance word difficulty prediction.
• Gradient boosting decision trees are suitable for this task.
• Feature contribution analysis is crucial for model understanding.

Method

The system uses features from word frequency, lexical norms, and psychometric data, fed into a gradient boosting decision tree model, with performance analyzed via gain scores and Spearman correlations.

In practice

• Utilize word frequency and psychometric data for L2 difficulty.
• Employ gradient boosting for predictive modeling.
• Analyze feature gain scores to refine models.

Topics

• English as a Second Language
• Word Difficulty Prediction
• Psycholinguistic Features
• Gradient Boosting
• Machine Learning Models
• Feature Engineering

Best for: AI Scientist, Research Scientist

Related on AIssential

• 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.
• Towards interpretable models for language proficiency assessment: Predicting the CEFR level of Estonian learner texts — Careful feature selection yields interpretable, generalizable models for language proficiency assessment with high accuracy.
• The Birth of Meaning — Word embeddings convert text into dense numerical vectors, capturing semantic meaning and context for AI models.
• What Makes Words Hard? Sakura at BEA 2026 Shared Task on Vocabulary Difficulty Prediction — Soft-target loss fine-tuning significantly improves LLM and MLM performance in continuous value prediction tasks.
• Probing for Reading Times — Early language model layers predict initial human reading effort better than scalar surprisal, which excels at later stages.
• 7 Readability Features for Your Next Machine Learning Model — Textstat provides diverse readability metrics as powerful features for machine learning models.

Open in AIssential →

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