emnlp2026

[EMNLP 2026 Main] Speaking While Listening: a survey and empirical audit of full-duplex spoken dialogue systems — L0–L3 architectural hierarchy, T×I×R interaction ontology, and a five-state decision machine, with a curated list of models, datasets, and benchmarks.

Official code release for the EMNLP 2026 paper "Med-Banana: Learning Quality-Controlled Medical Image Editing from Success-and-Failure Trajectories" — Med-Banana-80K dataset + edit–verify–refine system

[EMNLP 2026 Main] Steering Geometry: Validating Human Value Geometry in LLM Steering Space.

EMNLP 2026 Industry Track | A benchmark for shopping agents on long-horizon tasks with distributed hidden intent.

Official implementation of EMNLP 2026 Main paper HyQuant: Hybrid-Precision Quantization for LLM Attention

Phrase-Localized Language-Contrastive Guidance: Training-Free Localized Accent Control for Code-Switching TTS (EMNLP 2026)

Run 23 speech language models on your own audio task through one interface: vLLM, transformers, and API backends behind a single JSONL contract. Ships HEAR, the speaker-attribution benchmark from our EMNLP 2026 paper.

Induce structured task models from naturalistic computer-use traces

Source code for the EMNLP 2026 paper "Zero-Shot Tabular Learning via Hypergraph Reasoning with LLMs."

Official code for An Empirical Study on Preference Tuning Generalization and Diversity Under Domain Shift

[EMNLP 2026 Findings] UGTPHON: A multilingual grapheme-to-phoneme benchmark for user-generated text.

Paper list for 'Beyond Textual Chain-of-Thought: A Survey on Action-Grounded Reasoning in Autonomous Driving' (Findings of EMNLP 2026) — 130 methods across 13 reasoning-representation subtypes + 24 benchmarks

MemUse: evaluating whether conversational memory is actually used — benchmark, negatives split, judge harness, baselines (EMNLP 2026)

Refurbished Embeddings with Integrated Guidance Networks for Efficient Context-Length Scaling. EMNLP 2026 (Findings)

[EMNLP 2026] Length-Adaptive Decoding for Masked Diffusion Machine Translation — a training-free target-length selector for masked diffusion LMs.

Official code for 'Skill Reuse as Compression in Agentic RL' (EMNLP 2026)

Copy inflation in multi-turn search agents: 78-92% of generated tokens are copied from retrieved documents and carry inflated log-probabilities, breaking confidence-based methods. Diagnostic toolkit + Retrieval-Grounded Voting. Findings of EMNLP 2026.

[EMNLP 2026] DynaTokens: Controlling Token Dynamics for Continual Video-Language Understanding

Official code for "Every Answer Has Its Own Path: Agentic Routing for Retrieval-Augmented Table Question Answering" (EMNLP 2026)

Sparse probing and activation patching of AI-text detection neurons in frozen BERT (EMNLP 2026)