Papers with model predictions
On the Interpretability of Deep Learning Models for Collaborative Argumentation Analysis in Classrooms (2024.acl-srw)
Copied to clipboard
| Challenge: | Existing models for collaborative argumentation lack interpretability and teachers are skeptics about their use. |
| Approach: | They propose to use four explainable AI methods to provide models for automated analysis of argument moves and specificity levels within collaborative argumentation to cultivate trust among teachers. |
| Outcome: | The proposed models perform exceptionally well in analyzing word contributions and demonstrating that the models can be explained by a user-interface. |
HEIDL: Learning Linguistic Expressions with Deep Learning and Human-in-the-Loop (P19-3)
Copied to clipboard
| Challenge: | HITL-ML approaches are too low-level and far-removed from human’s conceptual models. |
| Approach: | They propose a prototype HITL-ML system that exposes the machine-learned model through high-level, explainable linguistic expressions formed of predicates representing semantic structure of text. |
| Outcome: | The proposed system exposes the machine-learned model through high-level, explainable linguistic expressions formed of predicates representing semantic structure of text. |
Progressive Fine-Tuning for Cost-Effective Structured Attribute Generation in E-commerce (2026.acl-industry)
Copied to clipboard
| Challenge: | Large language models excel at structured information generation but face cost and latency challenges when deployed at scale in user-facing products. |
| Approach: | They propose a parameter efficient supervised fine-tuning pipeline for adapting a small language model to structured attribute generation in e-commerce product listing. |
| Outcome: | The proposed model reduces inference costs by 98% and latency by 70% on a large-scale product listing service while preserving an 86.4% user acceptance rate. |
Improving the Faithfulness of Attention-based Explanations with Task-specific Information for Text Classification (2021.acl-long)
Copied to clipboard
| Challenge: | Recent studies show that attention cannot be considered as a faithful explanation across encoders and tasks. |
| Approach: | They propose a new family of Task-Scaling mechanisms that scale attention weights across tasks and two attention mechanisms. |
| Outcome: | The proposed models improve explanation faithfulness across two attention mechanisms, five encoders and five text classification datasets without sacrificing predictive performance. |
SELFEXPLAIN: A Self-Explaining Architecture for Neural Text Classifiers (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing models that explain text classification predictions are opaque and overfit to spurious artifacts. |
| Approach: | They propose a novel self-explaining model that explains a text classifier’s predictions using phrase-based concepts. |
| Outcome: | The proposed model shows that it is adequate, trustworthy and understandable by human judges compared to existing baselines. |
Confidence Modeling for Neural Semantic Parsing (P18-1)
Copied to clipboard
| Challenge: | Experimental results show that neural semantic parsers are difficult to interpret due to their complexity. |
| Approach: | They propose to use confidence models to estimate predictions for neural semantic parsers . they outline three major causes of uncertainty and use metrics to quantify them . |
| Outcome: | The proposed model outperforms a widely used method that relies on posterior probability and improves interpretation quality. |
How well can Text-to-Image Generative Models understand Ethical Natural Language Interventions? (2022.emnlp-main)
Copied to clipboard
| Challenge: | Text-to-image generative models can generate high-quality photo-realistic images conditional on natural language text descriptions in a zero-shot fashion. |
| Approach: | They propose an Ethical NaTural Language Interventions in Text-to-Image GENeration benchmark dataset to evaluate the change in image generation conditional on ethical interventions across three social axes – gender, skin color, and culture. |
| Outcome: | The proposed model generations cover diverse social groups while preserving image quality. |
Instance-Based Neural Dependency Parsing (2021.tacl-1)
Copied to clipboard
Hiroki Ouchi, Jun Suzuki, Sosuke Kobayashi, Sho Yokoi, Tatsuki Kuribayashi, Masashi Yoshikawa, Kentaro Inui
| Challenge: | Existing models that use instance-based inference for dependency parsing are difficult to understand for humans. |
| Approach: | They develop neural models that adopt an interpretable inference process for dependency parsing. |
| Outcome: | The proposed models achieve competitive accuracy with standard neural models and have plausibility of instance-based explanations. |
Sorting through the noise: Testing robustness of information processing in pre-trained language models (2021.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained language models have shown impressive performance on downstream NLP tasks, but we have yet to establish a clear understanding of their sophistication when it comes to processing, retaining, and applying information presented in their input. |
| Approach: | They examine how robustly pre-trained LMs retain and apply relevant context information in the face of distracting content. |
| Outcome: | The proposed models retain and use critical context information in the face of distracting content, while models are susceptible to factors of semantic similarity and word position. |
Toward Inclusive Language Models: Sparsity-Driven Calibration for Systematic and Interpretable Mitigation of Social Biases in LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a new method to mitigate stereotypical bias in large language models is needed . inherent biases from training on vast Internet datasets can amplify harmful stereotypes . |
| Approach: | They propose a method to identify stereotypical bias in decoder-only transformer models . they apply a localization mechanism that correlates internal activations with a new Context Influence score . |
| Outcome: | The proposed method reduces stereotypical biases on BBQ, StereoSet, and CrowS-Pairs while improving reasoning performance on MMLU by 10%. |
Exploring the Universal Vulnerability of Prompt-based Learning Paradigm (2022.findings-naacl)
Copied to clipboard
| Challenge: | Prompt-based learning inherits the vulnerability from pre-training, where model predictions can be misled by inserting triggers into the text. |
| Approach: | They propose a potential solution to mitigate this vulnerability by injecting triggers into pre-trained language models using only plain text. |
| Outcome: | The proposed learning paradigm inherits the vulnerability from the pre-training stage . it can totally control or severely decrease the performance of prompt-based models . |
LOGAN: Local Group Bias Detection by Clustering (2020.emnlp-main)
Copied to clipboard
| Challenge: | a number of machine learning models inherit and amplify the societal biases in data. |
| Approach: | a new bias detection technique based on clustering is proposed to detect local biases in data . authors propose to use LOGAN to analyze local bias in data. |
| Outcome: | The proposed technique detects bias in a local region and allows better analysis of model predictions. |
Comparing Explanation Faithfulness between Multilingual and Monolingual Fine-tuned Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Previous studies have investigated how different factors affect faithfulness of model explanations . |
| Approach: | They find that the larger the multilingual model, the less faithful FAs are compared to its counterpart monolingual models. |
| Outcome: | The results show that the larger the multilingual model, the less faithful the FAs are compared to its counterpart monolingual models. |
Re-Examining Calibration: The Case of Question Answering (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing calibration methods do not provide significant gains in accuracy. |
| Approach: | They propose a new calibration metric that better captures whether the model assigns low confidence to wrong predictions and high confidence to correct predictions. |
| Outcome: | The proposed calibration method better captures whether the model assigns low confidence to wrong predictions and high confidence to correct predictions. |
Cross-Domain Classification of Moral Values (2022.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods to identify moral values in text can be challenging for transferring knowledge between domains. |
| Approach: | They compare a deep learning model with a domain-specific value classifier to find out whether it can transfer knowledge to new domains. |
| Outcome: | The proposed model can generalize and transfer knowledge to novel domains, but introduce catastrophic forgetting. |
Efficient Nearest Neighbor based Uncertainty Estimation for Natural Language Processing Tasks (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods for uncertainty estimation are inadequate for safety-critical applications. |
| Approach: | They propose a method that uses the distances from neighbors and the ratio of labels in neighbors to estimate uncertainty. |
| Outcome: | The proposed method outperforms baseline and density-based methods in calibration and uncertainty metrics. |
ASPO: Adaptive Sentence-Level Preference Optimization for Fine-Grained Multimodal Reasoning (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances have extended DPO to multimodal scenarios, achieving strong performance. |
| Approach: | They propose to use a sentence-level preference optimization technique to optimize individual sentences for more precise preference optimization without additional models or parameters. |
| Outcome: | Experiments show that Adaptive Sentence-level Preference Optimization significantly improves the alignment of multimodal models. |
Is Attention Interpretable? (P19-1)
Copied to clipboard
| Challenge: | Attention mechanisms have recently boosted performance on a range of NLP tasks. |
| Approach: | They propose to manipulate attention weights in text classification models and analyze the resulting differences in their predictions. |
| Outcome: | The proposed approach improves models' predictions by using gradient-based rankings of attention weights. |
Prompt-based Connective Prediction Method for Fine-grained Implicit Discourse Relation Recognition (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to aid implicit discourse relation recognition (IDRR) lack explicit connectives and are difficult to implement on fine-grained IDRR. |
| Approach: | They propose a Prompt-based Connective Prediction method that instructs large-scale pre-trained models to use knowledge relevant to discourse relation and utilizes strong correlation between connectives and discourse relation to help the model recognize implicit discourse relations. |
| Outcome: | The proposed method surpasses the state-of-the-art model and achieves significant improvements on those fine-grained few-shot discourse relation classes. |
Asymmetric feature interaction for interpreting model predictions (2023.findings-acl)
Copied to clipboard
| Challenge: | Prior work on feature interaction attribution studies focus on asymmetric interaction that only explains the additional influence of a set of words in combination, which fails to capture asymmetry influence that contributes to model prediction. |
| Approach: | They propose an asymmetric feature interaction attribution explanation model that explores asymmetry higher-order feature interactions in the inference of deep neural NLP models. |
| Outcome: | The proposed model outperforms state-of-the-art models on two sentiment classification datasets. |
Does Robustness Improve Fairness? Approaching Fairness with Word Substitution Robustness Methods for Text Classification (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to reduce disparities in model outcomes have focused on data augmentation, debiasing model embeddings, or adding fairness-based optimization objectives during training. |
| Approach: | They propose to use certified word substitution robustness methods to improve equality of odds and equality of opportunity on multiple text classification tasks. |
| Outcome: | The proposed methods improve equality of odds and equality of opportunity on multiple text classification tasks. |
On the Lack of Robust Interpretability of Neural Text Classifiers (2021.findings-acl)
Copied to clipboard
Muhammad Bilal Zafar, Michele Donini, Dylan Slack, Cedric Archambeau, Sanjiv Das, Krishnaram Kenthapadi
| Challenge: | Several models have been proposed to interpret models with feature-based interpretability methods. |
| Approach: | They propose to quantify the robustness of neural text classifiers by using two randomization tests to compare models with identical initializations. |
| Outcome: | The proposed methods show surprising deviations from expected behavior . the results raise questions about the extent of insights that practitioners may draw from interpretations. |
Are Larger Pretrained Language Models Uniformly Better? Comparing Performance at the Instance Level (2021.findings-acl)
Copied to clipboard
| Challenge: | Larger models have higher out-of-distribution robustness, while smaller models have lower accuracy on rare subgroups. |
| Approach: | They develop statistically rigorous methods to investigate whether large models are better on every instance . they find that individual predictions are highly sensitive to noise in the randomness in training . |
| Outcome: | The proposed model is worse than BERT-MINI on 1-4% of instances across MNLI, SST-2, and QQP, compared to the overall accuracy improvement of 2-10%. |
Logic Traps in Evaluating Attribution Scores (2022.acl-long)
Copied to clipboard
| Challenge: | Modern deep learning models are notoriously opaque, which has motivated the development of methods for interpreting how deep models predict. |
| Approach: | They propose to review existing methods for evaluating attribution scores and summarize the logic traps in these methods. |
| Outcome: | The proposed methods show that they do not contain logic traps and that they are not reliable. |
LawBench: Benchmarking Legal Knowledge of Large Language Models (2024.emnlp-main)
Copied to clipboard
Zhiwei Fei, Xiaoyu Shen, Dawei Zhu, Fengzhe Zhou, Zhuo Han, Alan Huang, Songyang Zhang, Kai Chen, Zhixin Yin, Zongwen Shen, Jidong Ge, Vincent Ng
| Challenge: | LegalBench evaluated 20 LLMs in 162 legal tasks in 20 countries and jurisdictions. |
| Approach: | They present a comprehensive evaluation of 21 popular Large Language Models and the first comparative analysis of the empirical results. |
| Outcome: | The proposed benchmarks are based on the Bloom’s cognitive taxonomy and are compared to 21 popular LLMs. |
Code Vulnerability Detection via Nearest Neighbor Mechanism (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to learn code semantics from source code are difficult to identify. |
| Approach: | They propose a method which retrieves multiple neighbor samples and utilizes label information to provide help for model predictions. |
| Outcome: | Extensive experiments show that the proposed method can achieve obvious performance improvements compared to baseline models. |
Detecting and Reducing Bias in a High Stakes Domain (D19-1)
Copied to clipboard
| Challenge: | Existing research shows that a deep learning model can predict aggression and loss in posts by focusing on stop words such as “a” or “on”. |
| Approach: | They developed an approach to interpret a deep learning model that often bases its predictions on stop words such as "a" or "on" to tackle bias, they annotated the rationales and built models that drastically reduce bias. |
| Outcome: | The proposed model can predict aggression and loss in posts by using stop words such as "a" or "on" the new annotations enable us to quantitatively measure how justified the model predictions are, and build models that drastically reduce bias. |
A Little Annotation does a Lot of Good: A Study in Bootstrapping Low-resource Named Entity Recognizers (D19-1)
Copied to clipboard
| Challenge: | Named entity recognition models rely on large amounts of labeled data, making them challenging to extend to new, lower-resource languages. |
| Approach: | They propose a method for bootstrapping named entity recognition models in under-resourced languages . they use cross-lingual transfer learning and targeted annotation of only uncertain entities . |
| Outcome: | The proposed method achieves competitive accuracy with just one-tenth of training data. |
To Generate or to Retrieve? On the Effectiveness of Artificial Contexts for Medical Open-Domain Question Answering (2024.acl-long)
Copied to clipboard
| Challenge: | Medical open-domain question answering requires substantial access to specialized knowledge. |
| Approach: | They propose a framework that generates multiple-choice questions from a set of open-book parameters and a small-scale reader that can outcompete closed-book questions by 706x using fewer parameters. |
| Outcome: | The proposed framework outperforms closed-book models on MedQA-USMLE, MedMCQA, and MMLU while using up to 706x fewer parameters. |
Uncertainty Estimation of Transformer Predictions for Misclassification Detection (2022.acl-long)
Copied to clipboard
Artem Vazhentsev, Gleb Kuzmin, Artem Shelmanov, Akim Tsvigun, Evgenii Tsymbalov, Kirill Fedyanin, Maxim Panov, Alexander Panchenko, Gleb Gusev, Mikhail Burtsev, Manvel Avetisian, Leonid Zhukov
| Challenge: | Uncertainty estimation (UE) of model predictions is crucial step for a variety of tasks such as active learning, misclassification detection, adversarial attack detection, etc. |
| Approach: | They propose to modify UE methods for Transformer models for misclassification detection in named entity recognition and text classification tasks to improve model expressiveness and computational performance. |
| Outcome: | The proposed methods outperform computationally intensive methods on misclassification detection tasks and are based on a large dataset of simulated datasets. |
Instance-Based Learning of Span Representations: A Case Study through Named Entity Recognition (2020.acl-main)
Copied to clipboard
Hiroki Ouchi, Jun Suzuki, Sosuke Kobayashi, Sho Yokoi, Tatsuki Kuribayashi, Ryuto Konno, Kentaro Inui
| Challenge: | Recent neural networks can induce good span feature representations and achieve high performance in structured prediction tasks. |
| Approach: | They propose an instance-based learning method that learns similarities between spans . they aim to build models that have high interpretability without sacrificing performance . |
| Outcome: | The proposed method improves interpretability without sacrificing performance. |
Learning to Substitute Words with Model-based Score Ranking (2025.naacl-long)
Copied to clipboard
| Challenge: | Experimental results show that the proposed approach outperforms both masked language models and large language models. |
| Approach: | They propose a model-based scoring approach to quantify sentence quality . they propose 'loss function' that optimizes alignment between model predictions and sentence scores . |
| Outcome: | The proposed approach outperforms masked language models and large language models in the quantitative analysis of word substitutions. |
LBC: Language-Based-Classifier for Out-Of-Variable Generalization (2025.naacl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have great success in natural language processing tasks such as response generation, but their performance on tabular data tasks has been limited due to their inferior performance compared to traditional machine learning models (TMLs). |
| Approach: | They propose a Language-Based-Classifier (LBC) that maximizes the benefits of LLMs to outperform TMLs on OOV tasks. |
| Outcome: | The proposed model outperforms TMLs on OOV tasks by using three key methods. |
BERT for Coreference Resolution: Baselines and Analysis (D19-1)
Copied to clipboard
| Challenge: | Recent BERT-based models have reported dramatic gains on multiple semantic benchmarks including question-answering, natural language inference, and named entity recognition. |
| Approach: | They apply BERT to coreference resolution, achieving a new state of the art on the GAP and OntoNotes benchmarks. |
| Outcome: | A qualitative analysis of model predictions shows that BERT-large is better at distinguishing between related but distinct entities, but there is room for improvement in modeling document-level context, conversations, and mention paraphrasing. |
Measuring the Mixing of Contextual Information in the Transformer (2022.emnlp-main)
Copied to clipboard
| Challenge: | Experimentally, we show that ALTI provides more faithful explanations and increased robustness than gradient-based methods. |
| Approach: | They propose to measure token-to-token interactions within each layer and then use them to aggregate model predictions. |
| Outcome: | The proposed method provides more faithful explanations and increased robustness than gradient-based methods. |
What’s in a Name? Answer Equivalence For Open-Domain Question Answering (2021.emnlp-main)
Copied to clipboard
| Challenge: | A flaw in QA evaluation is that annotations often only provide one answer . therefore, model predictions semantically equivalent to the answer but superficially different are considered incorrect. |
| Approach: | They explore using alias entities from knowledge bases to extract additional answers . they incorporate additional answers for evaluation and model training with equivalent answers based on the results . |
| Outcome: | The proposed solution improves the accuracy of evaluation with additional answers and improves model training with equivalent answers. |
A Hierarchical Explanation Generation Method Based on Feature Interaction Detection (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing work on hierarchical attributions tends to limit text groups to a continuous text span, which is difficult for humans to read. |
| Approach: | They propose a method which captures feature interactions and converts non-hierarchical explanations into hierarchical versions. |
| Outcome: | The proposed method can convert ubiquitous non-hierarchical explanations into their corresponding hierarchical versions. |
Post-hoc Study of Climate Microtargeting on Social Media Ads with LLMs: Thematic Insights and Fairness Evaluation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Climate change communication on social media increasingly employs microtargeting strategies to effectively reach and influence specific demographic groups. |
| Approach: | They analyze social media ads using large language models to examine their performance . they find that LLMs perform well overall, but certain biases exist . |
| Outcome: | The results show that LLMs perform well overall, but certain biases exist in certain demographic groups. |
SeqMMR: Sequential Model Merging and LLM Routing for Enhanced Batched Sequential Knowledge Editing (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing research has demonstrated strong performance in single-instance or few-instantial sequential editing and one-time massive editing scenarios, but the batched sequential editing paradigm remains a significant challenge. |
| Approach: | They propose a framework for batched sequential knowledge editing that leverages **SeqMMR** and a model router to merge parameters from current batch-edited models with those of their predecessors. |
| Outcome: | The proposed framework iteratively merges parameters from current batch-edited models with those of their predecessors, ensuring that newly emerging knowledge is integrated while mitigating the forgetting of previously edited knowledge. |
Model Intrinsic Features of Fine-tuning based Text Summarization Models for Factual Consistency (2023.findings-acl)
Copied to clipboard
| Challenge: | a summarization model with relatively low factual consistency is more likely to model summaries that are not conditional to the documents. |
| Approach: | They analyze the model intrinsic features by varying the fine-tuning objectives and datasets. |
| Outcome: | The proposed models have a high inductive bias for factual consistency and are more likely to model summaries that are not conditional to the documents. |
Intent Discovery with Frame-guided Semantic Regularization and Augmentation (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing intent discovery methods focus on transferring prior knowledge of known intents to unknown ones. |
| Approach: | They propose to use frame knowledge as conceptual semantic guidance to bridge the gap between known intents representation learning and unknown intents clustering. |
| Outcome: | The proposed method outperforms solid baselines on two benchmark datasets. |
RRInf: Efficient Influence Function Estimation via Ridge Regression for Large Language Models and Text-to-Image Diffusion Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Influence function estimates the impact of training data on model predictions . high computational cost has hindered their applicability in large-scale applications. |
| Approach: | They propose a method to quantify the impact of training data on model predictions . they use a ridge regression problem to transform the estimation into a problem . |
| Outcome: | The proposed method outperforms existing methods on noisy data detection and influential data identification tasks. |
Parameter-Aware Contrastive Knowledge Editing: Tracing and Rectifying based on Critical Transmission Paths (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) encode vast amounts of knowledge in their parameters, but the acquired knowledge can be incorrect or outdated over time, necessitating rectification after pre-training. |
| Approach: | They propose a method that captures key information flows that influence model predictions . they propose 'critical transmission paths' to improve model editing . |
| Outcome: | The proposed method improves on two prominent datasets and three widely used LLMs. |
LLM-based Rumor Detection via Influence Guided Sample Selection and Game-based Perspective Analysis (2025.acl-long)
Copied to clipboard
Zhiliang Tian, Jingyuan Huang, Zejiang He, Zhen Huang, Menglong Lu, Linbo Qiao, Songzhu Mei, Yijie Wang, Dongsheng Li
| Challenge: | Existing methods for rumor detection on social media are limited by limited modeling capacity and insufficient training corpora. |
| Approach: | They propose an SFT-based rumor detection model with Influence guided Sample selection and Game-based multi-perspective analysis to address these issues. |
| Outcome: | The proposed model outperforms existing SOTA on three datasets. |
Towards Explainability and Fairness in Swiss Judgement Prediction: Benchmarking on a Multilingual Dataset (2024.lrec-main)
Copied to clipboard
| Challenge: | Using Swiss Judgement Prediction, we evaluate the explainability of state-of-the-art monolingual and multilingual LJP models. |
| Approach: | They propose an occlusion-based approach to evaluate the explainability performance of legal judgement prediction models using Swiss Judgement Prediction, the only available multilingual LJP dataset. |
| Outcome: | The proposed framework allows us to quantify the influence of lower court information on model predictions, exposing current models’ biases. |
Preference Heads in Large Language Models: A Mechanistic Framework for Interpretable Personalization (2026.acl-long)
Copied to clipboard
Weixu Zhang, Ye Yuan, Changjiang Han, Yuxing Tian, Zipeng Sun, Linfeng Du, Jikun Kang, Hong Kang, Xue Liu, Haolun Wu
| Challenge: | Large Language Models exhibit strong implicit personalization ability, but most approaches treat this behavior as a black box. |
| Approach: | They propose a mechanistic interpretation perspective and propose 'sparse' set of Preference Heads . they compute a Preference Contribution Score for each attention head and compare their predictions . |
| Outcome: | The proposed framework computes a Preference Contribution Score (PCS) for each attention head and measures its causal impact on user aligned outputs. |