Work

uogal at BEA 2026 Shared Task 1: Ensemble of Multilingual Encoders with NMT Augmentation for L1-Aware Vocabulary Difficulty Prediction

Year: 2026

Type: conference-paper

Abstract: This paper describes a system submission to the closed track of the BEA 2026 shared task on L1-aware vocabulary difficulty prediction (Spanish, German, Mandarin Chinese).We explored three approaches: hand-crafted tabular features with tree-based regressors, fine-tuned multilingual encoders, and fine-tuned decoders (artificial learner simulation with LoRA-tuned Pythia models).Each was examined with and without an English context produced by neural machine translation (NMT), and we also combined all three families in a single ensemble.In the end, ensembling only the fine-tuned multilingual encoders (four base architectures together with four NMT-augmented variants, combined through per-language stacking) gave the best results while remaining the simplest configuration to train.

Source: Workshop on Innovative Use of NLP for Building Educational Applications

DOI: 10.18653/v1/2026.bea-1.75

Authors: Bernardo Stearns, John P. McCrae, Thomas Gaillat, Jefkine Kafunah

Institutions: Ollscoil na Gaillimhe – University of Galway, Insight SFI Research Centre for Data Analytics, Linguistique Ingénierie et Didactique des Langues, Université Rennes 2

Language: English

Topic: Natural Language Processing Techniques

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

SDG: Quality education

Open Access status: hybrid

OpenAlex ID: W7166851265