AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Steering Embedding Models with Geometric Rotation: Mapping Semantic Relationships Across Languages and Models

Michael Freenor, Lauren Alvarez

ICLR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining

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
use RISE geodesic rotation instead of Procrustes global rotation (ablation)sentence embedding transformation · SICKcosine similarity: 0.67 -> 0.84Improved
use RISE instead of mean difference vectors on the sphere (MDV) (ablation)sentence embedding transformation · SICKcosine similarity: 0.83 -> 0.84No clear effect
use RISE instead of Procrustes on BLiMP (ablation)sentence embedding transformation · BLiMPcosine similarity: 0.99 -> 0.97No clear effect
use RISE instead of Procrustes for cross-language transfer (ablation)sentence embedding transformation · Cross-language transfer (all phenomena)cosine similarity: 0.25-0.62 -> 0.74-0.89Improved
switch phenomenon from politeness to negationRISE · cross-language transfermean rotor alignment score: 0.762 -> 0.788Improved
switch embedding model from mBERT to bge-m3RISE · Synthetic Multilingualrotor alignment score: 0.709 -> 0.782Improved
switch embedding model from TE3L to mBERTRISE · SICKrotor alignment score: 0.623 -> 0.736Improved
evaluate on SICK instead of BLiMPRISE · BLiMP vs SICKrotor alignment score: 0.949 -> 0.663Hurt
increase embedding dimensionality (768 to 3072)RISE · Validation Avgmean cosine: 0.802 -> 0.774No clear effect
use mBERT instead of bge-m3 for cross-language transferRISE · Cross-lang/monolingual ratio, politenessratio: 0.95 -> 0.77Hurt
transfer prototypes cross-model to bge-m3 for Zulu instead of EnglishRISE with Morris mapping · TE3L to bge-m3 transfercosine similarity: 0.80-0.82 -> 0.63-0.66Hurt

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