Olfactory-Inspired Sparse Combinatorial Coding for Low-Resource Named Entity Recognition

The paper introduces a novel olfactory-inspired architecture for Named Entity Recognition (NER) in low-resource languages, utilizing a receptor-glomerular bottleneck between standard token embeddings and a BiLSTM-CRF model. Evaluated on six multilingual datasets, the approach demonstrates significant F1 score improvements, particularly in low-resource scenarios, achieving a +6.23% F1 increase in Bangla and +4.43% in Telugu under strict data limitations. This architecture provides an effective inductive bias and regularization mechanism, enhancing representation learning in environments with limited supervision.