Skip to content

๐Ÿ”„ Self-Supervised Learning

๐Ÿ’ฌ EMNLP2026 ยท 2 paper notes

๐Ÿ“Œ Same area in other venues: ๐Ÿง  NeurIPS2026 (63) ยท ๐ŸŽž๏ธ ECCV2026 (25) ยท ๐Ÿ“ท CVPR2026 (98) ยท ๐Ÿ”ฌ ICLR2026 (82) ยท ๐Ÿ’ฌ ACL2026 (1) ยท ๐Ÿงช ICML2026 (28)

CWoMP: Morpheme Representation Learning for Interlinear Glossing

CWoMP reformulates interlinear glossing from unconstrained character sequence generation into generative retrieval over a mutable lexicon of atomic form-meaning morpheme units, eliminating morpheme hallucinations, enforcing strict 1:1 segmentation-gloss alignment, reducing inference FLOPs by 8.5x, and allowing zero-retraining inference-time vocabulary expansion.

Unified Work Embeddings: Contrastive Learning of a Bidirectional Multi-Task Ranker

To overcome task fragmentation and extreme label cardinality in labor market intelligence, this paper introduces the WorkBench ranking evaluation suite and Unified Work Embeddings (UWE)โ€”a 109M-parameter bi-encoder that leverages bipartite many-to-many InfoNCE loss and bidirectional soft late interaction, surpassing 8B generalist models while exhibiting strong positive cross-task transfer.