AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Neologism Learning for Controllability and Self-Verbalization

John Hewitt, Oyvind Tafjord, Robert Geirhos, Been Kim

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 10 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
learn a neologism embedding (frozen model) for a simple conceptGemma-3-4B-IT · 7 simple concepts (LIMA)concept score increase (%): 92% of base-to-training gap closed on averageImproved
use single-word synonym self-verbalizations in place of the neologism (ablation)Gemma-3-4B-IT · 7 simple conceptsconcept score increase (%): 92% (neologism) -> 39% (best synonym), 25% (first synonym)Mixed
use long questionnaire-based verbalization in place of the neologism (ablation)Gemma-3-4B-IT · 7 simple conceptsconcept score increase (%): 92% -> 83% average; use-like 103% -> 32%Mixed
train with APO-up (DPO variant) instead of plain likelihood (ablation)neologism learning · 7 simple conceptsconcept score increase (%): 70% -> 92%Improved
train with multiple prompt templates (ablation)neologism learning · multi-template evaluationconcept score increase (%): 77% -> 86%Improved
add hinge loss on embedding norm to multi-template training (ablation)neologism learning · multi-template evaluationconcept score increase (%): 86% -> 89%Improved
add hinge loss to multi-template training (ablation)neologism learning · 2 concepts combinedconcept score increase (%): 76% -> 54%Hurt
learn neologisms instead of few-shot prompting for compositionsGemma-3-4B-IT · Short+Numerical+Likely all threegoal score (harmonic mean): 0.387 -> 0.482Improved
learn neologism instead of few-shot for the complex Likely conceptGemma-3-4B-IT · Likely onlysuccess rate: 0.281 -> 0.667Improved
learn neologism instead of few-shot for simple Short conceptGemma-3-4B-IT · Short onlygoal score: 0.922 -> 0.736Hurt
in-context learning of the neologism with 10 examples instead of embedding training (ablation)Gemma-3-4B-IT · 5 simple conceptsconcept score increase (%): 46% average for in-context on GemmaHurt
learn neologisms for complex AxBench conceptsGemma-3-4B-IT · AxBench (5 concepts)overall score (0-2): 0.0-0.4 (no concept) -> 1.54-1.91Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy