Research map / Large language models / Language, safety and interpretability
LS-Tree: Model Interpretation When the Data Are Linguistic
Jianbo Chen, Michael Jordan
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 BERT instead of BoW | Sentiment classification · SST test | accuracy: 0.82% -> 0.93% | Improved |
| Use BERT instead of BoW | Sentiment classification · IMDB test | accuracy: 0.94% -> 0.93% | No clear effect |
| Use BERT instead of BoW | Sentiment classification · Yelp test | accuracy: 0.94% -> 0.97% | Improved |
| Use BERT rather than CNN or LSTM | Models interpreted by LS-Tree · SST, IMDB, and Yelp | average correlation with BoW linear coefficients: SST: 0.591, 0.580 -> 0.465; IMDB: 0.442, 0.552 -> 0.321; Yelp: 0.683, 0.684 -> 0.476 | Mixed |
| Use BERT rather than CNN or LSTM | Models interpreted by LS-Tree · SST and IMDB adversative relations | parent-node interaction score relative to a generic node: SST average score: 1.238; IMDB average score: 1.159 | Improved |
| Increase training data | CNN and LSTM adversative-relation modeling · Yelp | parent-node interaction score relative to a generic node: Yelp average scores: CNN 0.161; LSTM 0.224; BERT 0.746 | Improved |
| Use parent-node interaction scores to distinguish adversative from non-adversative contexts | Models interpreting the word while · Yelp reviews | parent-node interaction score relative to a generic node: BERT: 2.167 (adversative) vs 0.082 (non-adversative) | Improved |
| Compare training and test interaction-score variance | Overfitted neural models · Training versus test data | average variance of interaction scores | Improved |
Closest papers by meaning
- Measuring Progress in Dictionary Learning for Language Model Interpretability with Board Game Models (NeurIPS 2024 · similarity 0.84)
- Automated Assessment of Fidelity and Interpretability: An Evaluation Framework for Large Language Models’ Explanations (Student Abstract) (AAAI 2024 · similarity 0.83)
- Linguini: A benchmark for language-agnostic linguistic reasoning (NeurIPS 2025 · similarity 0.83)
- Distinguishing the Knowable from the Unknowable with Language Models (ICML 2024 · similarity 0.83)
- Improving Unsupervised Constituency Parsing via Maximizing Semantic Information (ICLR 2025 · similarity 0.83)
- Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again (NeurIPS 2026 · similarity 0.83)
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.
