EMNLP2025 Ghonim: concept-ediq

a massive bank of concepts multi model semantically linked

EMNLP2025 Bai: understanding and leveraging expert specialization of context faithfulness

Two steps: step one is to use router tuning to prioritize experts that rely on context, step two is to especially hit those for fine-tuning for improved Qantas alliance. Big gainz and hot pot and other QA data set just by the router tuning

EMNLP2025 Vasu: literature grounded hypothesis generation

Use citation links to generate a Providence graph of hypothesis, then, fine tune a language model to reproduce this Providence graph, use resulting model to improve RAG that would be contextually grounded

EMNLP2025Li: enhancing RAG RESPONSE evaluator

Cast rag response evaluator as a reinforcement learning problem with accuracy, format, evidence use rewards. Perform reinforcement learning fine-tuning to improve metric

EMNLP2025 Wang: understanding expressivity of language models through the next token distribution

Language models have an inductive bias for low or very high entropy tokens where medium entropy are harder to illicit, this is because embedding is deficient in rank

EMNLP2025 Arad: SAEs are good for steering if you select the right features

There are some features that actually meaningfully preturb the output, intervening on those actually still work

EMNLP2025 Zhang: a survey on hallucination

A survey of hallucination