Skip to content

๐Ÿง‘ Human Understanding

๐Ÿ’ฌ EMNLP2026 ยท 3 paper notes

๐Ÿ“Œ Same area in other venues: ๐Ÿง  NeurIPS2026 (31) ยท ๐ŸŽž๏ธ ECCV2026 (171) ยท ๐Ÿ“ท CVPR2026 (218) ยท ๐Ÿ”ฌ ICLR2026 (55) ยท ๐Ÿงช ICML2026 (7) ยท ๐Ÿค– AAAI2026 (24)

๐Ÿ”ฅ Top topics: LLM ร—2

Beyond CTC Spikes: LLM-Driven Gloss-Temporal Supervision for Continuous Sign Language Recognition

Addressing the sparse spike phenomenon and weak visual-semantic correspondence caused by CTC loss in continuous sign language recognition, this paper introduces the Gloss-Temporal Module (GTM) to generate sustained gloss-to-frame interval supervision via LLM reasoning over multi-source video cues, substantially reducing WER on pure RGB baselines without extra parameters.

DiscoSign: Discourse-Aware Text to Sign Language Gloss Translation

DiscoSign is the first discourse-aware computational framework for text-to-sign language gloss translation, using explicit state registries and deterministic post-processing in a single LLM call to enforce spatial coreference, question-answer clause restructuring, and concept-gloss consistency across multi-sentence discourse.

sEMG Typing Transcription with Conformer Encoding and Beam-Aware Large Language Model Correction

This paper introduces a two-stage wrist-sEMG typing transcription framework that links a multi-scale bilateral Conformer CTC frontend with a fine-tuned Qwen3.5-9B corrector via beam-ordered candidate lists, slashing generic CER to 19.65% and personalized CER to 2.01% on the emg2qwerty benchmark.