Research map / Label-efficient and robust learning / Semi-supervised learning
Towards Interpretable Natural Language Understanding with Explanations as Latent Variables
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| add variational EM joint training (vs M-step only) (ablation) | ELV · SemEval supervised | micro-F1: 80.2 -> 80.7 | Helped |
| add variational EM joint training (vs M-step only) (ablation) | ELV · TACRED supervised | micro-F1: 65.4 -> 65.9 | Helped |
| add variational EM joint training (vs M-step only) (ablation) | ELV · Restaurant supervised | macro-F1: 76.2 -> 77.8 | Helped |
| add variational EM joint training (vs M-step only) (ablation) | ELV · Laptop supervised | macro-F1: 74.1 -> 75.2 | Helped |
| add explanations via ELV to BERT-base | BERT-base · TACRED supervised | micro-F1: 64.7 -> 65.9 | Helped |
| add explanations via ELV to BERT-base | BERT-base · SemEval supervised | micro-F1: 78.3 -> 80.7 | Helped |
| explanation-based self-training instead of plain self-training | BERT-base · TACRED semi-supervised | micro-F1: 36.9 -> 42.5 | Helped |
| explanation-based self-training instead of plain self-training | BERT-base · Laptop semi-supervised | macro-F1: 42.3 -> 63.6 | Helped |
| explanation-based self-training instead of data programming | BERT-base · SemEval semi-supervised | micro-F1: 47.9 -> 66.4 | Helped |
| corrupt 80% of words in retrieved explanations (ablation) | ELV prediction model · Restaurant | macro-F1: 77.8 -> 73.2 | Hurt |
| mix original and corrupted explanations 1:1 (ablation) | ELV prediction model · Laptop | macro-F1: 75.2 -> 74.0 | Hurt |
| ELV joint training for explanation generator vs fine-tuned UniLM | explanation generator · human eval | informativeness: 3.48 -> 3.87 | Helped |
Closest papers by meaning
- Think in Latent, Explain in Language: Self-Explainable Latent Reasoning (ICML 2026 · similarity 0.88)
- Training Language Models to Explain Their Own Computations (NeurIPS 2026 · similarity 0.88)
- Faithful Explanations of Black-box NLP Models Using LLM-generated Counterfactuals (ICLR 2024 · similarity 0.86)
- Predictive Concept Decoders: Training Scalable End-to-End Interpretability Assistants (NeurIPS 2026 · similarity 0.86)
- Towards More Faithful Natural Language Explanation Using Multi-Level Contrastive Learning in VQA (AAAI 2024 · similarity 0.86)
- Automated Assessment of Fidelity and Interpretability: An Evaluation Framework for Large Language Models’ Explanations (Student Abstract) (AAAI 2024 · similarity 0.86)
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.