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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use RISE geodesic rotation instead of Procrustes global rotation (ablation) | sentence embedding transformation · SICK | cosine similarity: 0.67 -> 0.84 | Improved |
| use RISE instead of mean difference vectors on the sphere (MDV) (ablation) | sentence embedding transformation · SICK | cosine similarity: 0.83 -> 0.84 | No clear effect |
| use RISE instead of Procrustes on BLiMP (ablation) | sentence embedding transformation · BLiMP | cosine similarity: 0.99 -> 0.97 | No 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 -> 0.74-0.89 | Improved |
| switch phenomenon from politeness to negation | RISE · cross-language transfer | mean rotor alignment score: 0.762 -> 0.788 | Improved |
| switch embedding model from mBERT to bge-m3 | RISE · Synthetic Multilingual | rotor alignment score: 0.709 -> 0.782 | Improved |
| switch embedding model from TE3L to mBERT | RISE · SICK | rotor alignment score: 0.623 -> 0.736 | Improved |
| evaluate on SICK instead of BLiMP | RISE · BLiMP vs SICK | rotor alignment score: 0.949 -> 0.663 | Hurt |
| increase embedding dimensionality (768 to 3072) | RISE · Validation Avg | mean cosine: 0.802 -> 0.774 | No clear effect |
| use mBERT instead of bge-m3 for cross-language transfer | RISE · Cross-lang/monolingual ratio, politeness | ratio: 0.95 -> 0.77 | Hurt |
| transfer prototypes cross-model to bge-m3 for Zulu instead of English | RISE with Morris mapping · TE3L to bge-m3 transfer | cosine similarity: 0.80-0.82 -> 0.63-0.66 | Hurt |
Closest papers by meaning
- Invariant Features in Language Models: Geometric Characterization and Model Attribution (NeurIPS 2026 · similarity 0.89)
- Diff-eRank: A Novel Rank-Based Metric for Evaluating Large Language Models (NeurIPS 2024 · similarity 0.86)
- ROSE: Rotate Your Large Language Model to See (CVPR 2026 · similarity 0.86)
- A Polar coordinate system represents syntax in large language models (NeurIPS 2024 · similarity 0.86)
- What Rotary Position Embedding Can Tell Us: Identifying Query and Key Weights Corresponding to Basic Syntactic or High-level Semantic Information (NeurIPS 2024 · similarity 0.85)
- Tracing the Representation Geometry of Language Models from Pretraining to Post-training (NeurIPS 2025 · similarity 0.85)
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.
