Papers with interpretability analysis
TRACE: Training and Inference-Time Interpretability Analysis for Language Models (2025.emnlp-demos)
Copied to clipboard
| Challenge: | Existing tools for interpretability analysis of transformer models are post hoc, rely on scalar metrics or require nontrivial integration effort. |
| Approach: | They propose a modular toolkit for training and inference-time interpretability analysis of transformer models. |
| Outcome: | Experiments with autoregressive transformers show that TRACE reveals developmental phenomena overlooked by traditional scalar metrics such as loss or accuracy. |
Listening to Affected Communities to Define Extreme Speech: Dataset and Experiments (2022.findings-acl)
Copied to clipboard
| Challenge: | XTREMESPEECH dataset contains 20,297 social media passages from Brazil, Germany, India and Kenya . |
| Approach: | They propose a hate speech dataset containing 20,297 social media passages from Brazil, Germany, India and Kenya. |
| Outcome: | The proposed dataset contains 20,297 social media passages from Brazil, Germany, India and Kenya. |
Assessing Robustness of Text Classification through Maximal Safe Radius Computation (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Neural network NLP models are vulnerable to small modifications of the input that maintain the original meaning but result in a different prediction. |
| Approach: | They propose to provide a measure of robustness against word substitutions by computing a safe radius for a given input text. |
| Outcome: | The proposed methods are compared with LIME and CNN-Cert and show that they perform well on sentiment analysis and news classification models. |
Why and How LLMs Benefit from Knowledge Introspection in Commonsense Reasoning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) can improve commonsense reasoning by generating intermediate knowledge, but the effectiveness of this knowledge introspection is not always guaranteed. |
| Approach: | They propose a training-free strategy that optimizes introspection via two stages: Knowledge Detection and Knowledge Regeneration. |
| Outcome: | The proposed approach mitigates the limitations of standard introspection and has consistent performance gains across all settings. |
Understanding New-Knowledge-Induced Factual Hallucinations in LLMs: Analysis and Interpretation (2026.findings-acl)
Copied to clipboard
| Challenge: | Prior studies have shown that fine-tuning on new knowledge can induce factual hallucinations in large language models (LLMs), leading to incorrect outputs when evaluated on previously known information. |
| Approach: | They propose to conduct a fine-grained analysis of large language models using a dataset Biography-Reasoning and QA and knowledge reasoning tasks to understand their findings. |
| Outcome: | The proposed model is able to perform a range of downstream tasks without requiring a large amount of knowledge and is compared with a control dataset. |
Difference in Task Performance on Sparse Speech Representations (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for learning speech representations that are useful for a variety of downstream tasks have been extensively investigated in different domains. |
| Approach: | They propose to train Autoencoders with varying sparsity levels using three SSL features and evaluate them on six tasks of SUPERB: speech enhancement, speaker identification, speech Emotion Recognition, phone recognition, automatic speech recognition and slot filling. |
| Outcome: | The proposed model can be used to learn speech representations that are useful for a variety of downstream tasks. |
ICL CIPHERS: Quantifying ”Learning” in In-Context Learning via Substitution Ciphers (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies suggest that In-Context Learning operates in dual modes . however, disentangling these modes remains a challenging goal . |
| Approach: | They propose a class of task reformulations based on substitution ciphers borrowed from classic cryptography. |
| Outcome: | The proposed model can solve tasks with a BIJECTIVE mapping, but it requires 'deciphering' the latent cipher. |