Rationalizing Neural Predictions
Open accessThe DOI matches its Crossref record; 35 of the 35 references it cites were found in a registry or an index.
Abstract
Prediction without justification has limited applicability.As a remedy, we learn to extract pieces of input text as justifications -rationales -that are tailored to be short and coherent, yet sufficient for making the same prediction.Our approach combines two modular components, generator and encoder, which are trained to operate well together.The generator specifies a distribution over text fragments as candidate rationales and these are passed through the encoder for prediction.Rationales are never given during training.Instead, the model is regularized by desiderata for rationales.We evaluate the approach on multi-aspect sentiment analysis against manually annotated test cases.Our approach outperforms attention-based baseline by a significant margin.We also successfully illustrate the method on the question retrieval task. 1
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures






Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“A lot of work has also been done on argumentation and dialog games [6, 11, 51, 53] but the focus in these areas is generally the logical structure of the framework to express and to relate arguments or the protocol to exchange arguments.”
From Beyond explainability: justifiability and contestability of algorithmic decision systems (2022) · cites this paper as [53]
Experimental — accuracy not yet measuredIn the citing paper
“This model can achieve competitive accuracy while having built-in interpretability because the prediction is made exclusively based on rationales.”
From Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation (2022) · cites this paper as [52]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
"Why Should I Trust You?": Explaining the Predictions of Any Classifier
Ribeiro, M. T., Singh, S., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939778 · Cited by 17,555
Neural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213 · cited 2×
“Attention based models offer another lens to the inner workings of neural models (Bahdanau et al., 2015; Cheng et al., 2016; Martins and Astudillo, 2016; Chen et al., 2015; Xu and Saenko, 2015; Yang et al., 2015).”
BackgroundConvolutional Neural Networks for Sentence Classification
Kim, Y. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1181 · Cited by 14,831
Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
Xu, K., Ba, J., Kiros, R., Cho, K., Courville, A., Salakhutdinov, R., Zemel, R., & Bengio, Y. (2015) · arXiv · doi:10.48550/arxiv.1502.03044 · Cited by 8,248
Bag of Tricks for Efficient Text Classification
Joulin, A., Grave, E., Bojanowski, P., & Mikolov, T. (2017) · Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers · doi:10.18653/v1/e17-2068 · Cited by 4,545
Bag of Tricks for Efficient Text Classification
Joulin, A., Grave, E., Bojanowski, P., & Mikolov, T. (2016) · arXiv · doi:10.48550/arxiv.1607.01759 · Cited by 4,542
A Convolutional Neural Network for Modelling Sentences
Kalchbrenner, N., Grefenstette, E., & Blunsom, P. (2014) · Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.3115/v1/p14-1062 · Cited by 3,601 · cited 1×
“For instance, we could use a convolutional network (Kim, 2014; Kalchbrenner et al., 2014), deep averaging network (Iyyer et al.”
MethodA Neural Attention Model for Abstractive Sentence Summarization
Rush, A. M., Chopra, S., & Weston, J. (2015) · Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d15-1044 · Cited by 2,656
Recurrent Models of Visual Attention
Mnih, V., Heess, N., Graves, A., & Kavukcuoglu, K. (2014) · arXiv (Cornell University) · doi:10.48550/arxiv.1406.6247 · Cited by 2,568 · cited 1×
“space of possible rationales in terms of their interaction with the encoder. We could also apply variance reduction techniques to increase stability of stochastic training (cf. (Weaver and Tao, 2001; Mnih et al., 2014; Ba et al., 2015; Xu et al., 2015)). 7 Acknowledgments We thank Prof. Julian McAuley for sharing the…”
MethodTeaching Machines to Read and Comprehend
Hermann, K. M., Kočiský, T., Grefenstette, E., Espeholt, L., Kay, W., Suleyman, M., & Blunsom, P. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1506.03340 · Cited by 2,521
Research works printed but not linked: 6
- [6]6. Cicero dos Santos, Luciano Barbosa, Dasha Bogdanova, and Bianca Zadrozny. 2015. Learning hybrid rep- resentations to retrieve semantically equivalent ques- tions. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Lan- guage Processing (Volume 2: Short Papers), pages 694-699, Beijing, China, July. Association for Com- putational Linguistics.
- [7]7. Manaal Faruqui, Jesse Dodge, Sujay K. Jauhar, Chris Dyer, Eduard Hovy, and Noah A. Smith. 2015a. Retrofitting word vectors to semantic lexicons. In Pro- ceedings of NAACL.
- [9]9. Aurélie Herbelot and Eva Maria Vecchi. 2015. Build- ing a shared world: mapping distributional to model- theoretic semantic spaces. In Proceedings of the 2015 Conference on Empirical Methods in Natural Lan- guage Processing. Association for Computational Lin- guistics.
- [12]12. Mohit Iyyer, Varun Manjunatha, Jordan Boyd-Graber, and Hal Daumé III. 2015. Deep unordered compo- sition rivals syntactic methods for text classification. In Proceedings of the 53rd Annual Meeting of the As- sociation for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers).
- [27]27. Tim Rocktäschel, Edward Grefenstette, Karl Moritz Her- mann, Tomáš Kočiskỳ, and Phil Blunsom. 2016. Rea- soning about entailment with neural attention. In In- ternational Conference on Learning Representations.
- [29]29. Sebastian Thrun. 1995. Extracting rules from artifi- cial neural networks with distributed representations. In Advances in neural information processing systems (NIPS).
Cited by
Items 601–700 of 900
Kavumba, P., Brassard, A., Heinzerling, B., & Inui, K. (2023) · Findings of the Association for Computational Linguistics: EACL · doi:10.18653/v1/2023.findings-eacl.162 · Cited by 19
This Reads Like That: Deep Learning for Interpretable Natural Language Processing
Fanconi, C., Vandenhirtz, M., Husmann, S., & Vogt, J. E. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.869 · Cited by 6
“Our results demonstrate improved faithfulness in capturing the essential factors driving the model’s predictions compared to Lei et al. (2016).”
BackgroundMethodResultInfluentialDEFIne: A Fluent Interface DSL for Deep Learning Applications
Dethlefs, N., & Hawick, K. A. (2017) · doi:10.1145/3039895.3039898 · Cited by 4
“While this is an active area of ongoing research [45, 37], DEFIne provides a basic set of visualisation options to shed some light on the learning process and outcomes.”
BackgroundMaking Efficient Use of a Domain Expert's Time in Relation ExtractionPreprint
Adilova, L., Giesselbach, S., & Stefan Rüping, A. (2018) · arXiv (Cornell University) · doi:10.24406/publica-fhg-407874 · Cited by 4
What do Deep Networks Like to Read?Preprint
Pfeiffer, J., Kamath, A., Gurevych, I., & Ruder, S. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.04547 · Cited by 4
“Lei et al. (2016) jointly train a generator with the model and extract rationales by forcing the model’s prediction based on the rationale to be close to the model’s prediction on the original input.”
BackgroundMethodInfluentialExplainable Deep Modeling of Tabular Data using TableGraphNetPreprint
Terejanu, G., Chowdhury, J., Rashid, R., & Chowdhury, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.05205 · Cited by 4
“For natural language processing, (Lei et al., 2016) propose to extract pieces of text as justifications for predictions using a modular architecture containing a generator and encoder.”
MethodRelation Extraction with ExplanationPreprint
Shahbazi, H., Fern, X. Z., Ghaeini, R., & Tadepalli, P. V. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2005.14271 · Cited by 4
“Prior work studied learning from human provided rationales (Lei et al., 2016; Ross et al., 2017; Bao et al., 2018; Ghaeini et al., 2019) in order to improve model explain-ability.”
BackgroundA Weighted Solution to SVM Actionability and InterpretabilityPreprint
Denton, S. M., & Salleb-Aouissi, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.03372 · Cited by 4
“, [3, 10, 18]).”
Explaining the Road Not TakenPreprint
Shen, H., & Huang, T.-H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.14973 · Cited by 4
“We started with four commonly used forms, including “feature attribution [6, 26],” “tuple/graph [32, 47],” “free text [25, 39],” and “example [16, 54].”
BackgroundMethodRationale production to support clinical decision-makingPreprint
Taylor, N., Sha, L., Joyce, D. W., Lukasiewicz, T., Nevado‐Holgado, A. J., & Kormilitzin, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.07611 · Cited by 6
ReAGent: A Model-agnostic Feature Attribution Method for Generative Language ModelsPreprint
Zhao, Z., & Shan, B. (2024) · doi:10.22541/au.170709121.16176681/v1 · Cited by 4
Noiser: Bounded Input Perturbations for Attributing Large Language ModelsPreprint
Madani, M. R. G., Gema, A. P., Sarti, G., Zhao, Y., Minervini, P., & Passerini, A. (2025) · arXiv.org · doi:10.48550/arxiv.2504.02911 · Cited by 4
“The underlying goal is to identify which parts of the input contribute more toward the prediction ˆ y (Atanasova et al., 2020; Wallace et 2020; Madsen et al., 2022; Chrysostomou & Aletras, 2022; Lei et al., 2016; Chan et al., 2022a; Ghasemi Madani & Minervini, 2023).”
Learning from the Best: Rationalizing Prediction by Adversarial Information CalibrationPreprint
Sha, L., Camburu, O.-M., & Lukasiewicz, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.08884 · Cited by 4
“For NLP tasks, it is often desirable that a rationale is formed of fluent subphrases (Lei, Barzilay, and Jaakkola 2016).”
InfluentialExDDV: A New Dataset for Explainable Deepfake Detection in Video
Hondru, V., Hogea, E., Onchiş, D., & Ionescu, R. T. (2026) · 2026 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) · doi:10.1109/wacv61042.2026.00416 · Cited by 3
Yue, L., Liu, Q., Du, Y., Wang, L., An, Y., & Chen, E. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3592313 · Cited by 5
“For a fair comparison, we replicate the pre-process of [9] and [13] where the dataset has been divided into three aspects including appearance, smell, and palate.”
InfluentialCan LLM-Generated Textual Explanations Enhance Model Classification Performance? An Empirical Study
Dhaini, M., Vladika, J., Erdoğan, E., Attaoui, Z., & Kasneci, G. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-04549-2_16 · Cited by 5
A Prism Module for Semantic Disentanglement in Name Entity Recognition
Liu, K., Li, S., Zheng, D., Lu, Z., Gao, S., & Si, L. (2019) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p19-1532 · Cited by 3
“Hard attention has been successfully applied in computer vision (Ba et al., 2014; Mnih et al., 2014) but the application is limited in NLP. Lei et al. (2016) proposed a novel type of hard attention and apply it to improve the interpretability of models.”
BackgroundMethodQu, J., Arguello, J., & Wang, Y. (2023) · Conference on Human Information Interaction and Retrieval · doi:10.1145/3576840.3578315 · Cited by 10
“, words or sentences) that strongly influenced the classifier to predict the target category [16, 18, 23].”
BackgroundStyLEx: Explaining Style Using Human Lexical Annotations
Hayati, S. A., Park, K., Rajagopal, D., Ungar, L., & Kang, D. (2023) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.208 · Cited by 6
“Annotating explanations as rationales (part of input) (Lei et al., 2016) through expert annotations (Zaidan and Eis-ner, 2008) is widely used to model explanations in NLP when external annotations are available.”
BackgroundMethodSingh, G. V., Ghosh, S. K., Verma, A., Painkra, C., & Ekbal, A. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.275 · Cited by 3
“An encoder-generator system for quality rationales without annotations was presented in Lei et al. (2016).”
Explainability of Text Processing and Retrieval Methods
Anand, A., Saha, S., Sen, P., & Mitra, M. (2023) · Annual Meeting of the Forum for Information Retrieval Evaluation (FIRE) · doi:10.1145/3632754.3632944 · Cited by 4
“Specifically, there in the last five years, there has been a lot of progress in rationale-based models [16, 17, 46] that claim to be interpretable by design.”
BackgroundExploring the Interpretability of Sequential Predictions Through Rationale Model
Saidat, M. R. A., Salloum, S. A., & Shaalan, K. F. (2024) · Studies in big data · doi:10.1007/978-3-031-52280-2_2 · Cited by 3
Boosting Explainability through Selective Rationalization in Pre-trained Language Models
Yuan, L., Hu, S., Yu, K., & Wu, L. (2025) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3690624.3709212 · Cited by 5
Iterative Counterfactual Data Augmentation
Plyler, M., & Chi, M. (2025) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v39i19.34195 · Cited by 3
Liu, W., Deng, Z., Niu, Z., Wang, J., Wang, H., & Li, R. (2025) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00758 · Cited by 6
“We compare with RNP (Lei et al., 2016), DMR (Huang et al., 2021), Inter RAT (Yue et al., 2023), NIR (Storek et al., 2023), CR (Zhang et al., 2023), and A2I (Liu et al., 2024b).”
Scalable Explanation of Inferences on Large Graphs
Chen, C., Liu, Y., Zhang, X., & Xie, S. (2019) · Proceedings · doi:10.1109/icdm.2019.00111 · Cited by 7
“Explainable RNNs [23] handles linear chains but not general MRFs.”
BackgroundAspect-Based Sentiment Analysis with Adjustments to Irrelevant Sentimental-Related Features
Jiang, H., Wu, W., & Ren, J. (2019) · IEEE International Conference on Tools with Artificial Intelligence (ICTAI) · doi:10.1109/ictai.2019.00049 · Cited by 3
“The attention mechanism has proven to be very effective in machine translation [1], question answering [14] and many other tasks [2,15,16].”
BackgroundHierarchical POI Attention Model for Successive POI Recommendation
Li, L. (2021) · Transactions on computational science and computational intelligence · doi:10.1007/978-3-030-71704-9_11 · Cited by 3
Method for Generating Explainable Deep Learning Models in the Context of Air Traffic Management
Rudd, K., Eshow, M. M., & Gibbs, M. (2022) · Lecture notes in computer science · doi:10.1007/978-3-030-95467-3_17 · Cited by 3
The Problem of Sentiment Analysis
Liu, B. (2020) · Cambridge University Press eBooks · doi:10.1017/9781108639286.003 · Cited by 4
QUASER: Question Answering with Scalable Extractive Rationalization
Ghoshal, A., Iyer, S. V., Paranjape, B., Lakhotia, K., Yih, S., & Mehdad, Y. (2022) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3477495.3532049 · Cited by 7
“, rationales, as supporting evidence, has emerged as an effective solution [20, 40].”
BackgroundMethodInfluentialTraceNet : Tracing and locating the key elements in sentiment analysis
Zhao, Q., Liu, J., Kang, Z., & Zhou, Z. (2023) · Knowledge-Based Systems · doi:10.1016/j.knosys.2023.110792 · Cited by 3
An algorithm for training a class of polynomial models
Popescu, M. C., Grama, L., & Rusu, C. (2023) · Digital Signal Processing · doi:10.1016/j.dsp.2023.104168 · Cited by 3
A Compare-and-contrast Multistage Pipeline for Uncovering Financial Signals in Financial Reports
Ju, J.-H., Huang, Y.-S., Lin, C.-W., Lin, C. T., & Wang, C. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.800 · Cited by 6
“For example, Lei et al. (2016) proposed a method for learning the rationale (words) to justify a model’s prediction by selecting a subset of text inputs.”
MethodPruning Literals for Highly Efficient Explainability at Word Level
Yadav, R. K., Bhattarai, B., Jana, A., Jiao, L., & Yimam, S. M. (2024) · International Symposium on the Tsetlin Machine (ISTM) · doi:10.1109/istm62799.2024.10931519 · Cited by 3
“[20] advocated creating two models, each with its own set of parameters, one to extract rationales and the other to consume them to make a prediction.”
Desmoulin‐Canselier, S., & Métayer, D. L. (2018) · European journal of law and technology · Cited by 10
DISK-CSV: Distilling Interpretable Semantic Knowledge with a Class Semantic Vector
Bashier, H. K., Kim, M., & Goebel, R. G. (2021) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2021.eacl-main.263 · Cited by 3
“There are other methods, which focused on constructing a self-explainable network (Bastings et al., 2019) and (Lei et al., 2016).”
MethodLeveraging Causal Inference for Explainable Automatic Program Repair
Wang, J., Si, S., Zhu, Z., Qu, X., Hong, Z., & Xiao, J. Z. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892168 · Cited by 6
“[14], [28] proposes a model that justifies the prediction results using fragments of the input.”
BackgroundRecent Development on Extractive Rationale for Model Interpretability: A Survey
Wang, H., & Dou, Y. (2022) · International Conference on Cloud Computing, Big Data and Internet of Things (3CBIT) · doi:10.1109/3cbit57391.2022.00080 · Cited by 4
Explaining Transition Systems through Program InductionPreprint
Penkov, S., & Ramamoorthy, S. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1705.08320 · Cited by 3
“Methods for textual rationalisation of the predictions made by deep models have also been proposed [17–19].”
BackgroundKnowBias: A Novel AI Method to Detect Polarity in Online ContentPreprint
Saligrama, A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.00724 · Cited by 3
“There has been research on filtering text for the purposes of deriving justifiable predictions [4], but not for domain adaptation for our target problem.”
BackgroundDSReg: Using Distant Supervision as a RegularizerPreprint
Meng, Y., Li, M., Li, X., Wu, W. Q., & Li, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.11658 · Cited by 3
Regularizing Black-box Models for Improved Interpretability (HILL 2019 Version)Preprint
Plumb, G., Al-Shedivat, M., Xing, E. P., & Talwalkar, A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.01431 · Cited by 3
Teaching AI to Explain its Decisions Using Embeddings and Multi-Task LearningPreprint
Codella, N., Hind, M., Ramamurthy, K. N., Campbell, M. S., Dhurandhar, A., Varshney, K. R., Wei, D., & Mojsilović, A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.02299 · Cited by 4
“Work in the natural language processing and computer vision domains that generate rationales/explanations derived from input text (e.g., Lei et al. (2016); Ainur et al. (2010); Hendricks et al. (2016)).”
BackgroundExplainability-aided Domain Generalization for Image ClassificationPreprint
Schmidt, R. M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.01742 · Cited by 3
“The additional task usually tries to provide either text explanations [26, 79, 81, 207], an explanation association [6, 44, 90, 106], or prototypes [32, 114] which differ in the provided explainability type as well as the degree of insight.”
BackgroundYou Can Do Better! If You Elaborate the Reason When Making PredictionPreprint
Li, D., Tao, J., Chen, Q., & Hu, B. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.14919 · Cited by 3
“For example, previous works simulate the intractable sampling step by proposing optimization procedures based on reinforcement learning approaches [12, 25] and reparameterization techniques [9, 26].”
Multi-Dimensional Explanation of Target Variables from DocumentsPreprint
Antognini, D., Musat, C., & Faltings, B. V. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.11386 · Cited by 3
“1 Each should contain words that altogether are short, coherent, and alone sufficient for the prediction as a substitute of the input (Lei et al., 2016).”
BackgroundInfluentialExplainable Prediction of Text Complexity: The Missing Preliminaries for Text SimplificationPreprint
Gârbacea, C., Guo, M., Carton, S., & Mei, Q. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.15823 · Cited by 4
“Explainable classification can also be solved simultaneously through a neural network, using hard attentions to select individual words into the “rationale” behind a classification decision (Lei et al., 2016).”
BackgroundLearning to Rank Rationales for Explainable RecommendationPreprint
Xu, Z., Han, Y., Yang, T., Tran, A. T., & Ai, Q. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2206.05368 · Cited by 4
“also referred to as human-grounded evaluation by [24] (3)also referred to as Extractive Rationales by [20] and Highlights by [39] (4)interested readers can refer to [24] for a more detailed taxonomy.”
UNIREX: A Unified Learning Framework for Language Model Rationale ExtractionPreprint
Chan, A., Sanjabi, M., Mathias, L., Tan, L., Nie, S., Peng, X., Ren, X., & Firooz, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2112.08802 · Cited by 5
“predictor module uses only the extracted tokens to solve the task (Jain et al., 2020; Paranjape et al., 2020; Yu et al., 2021; Bastings et al., 2019; Lei et al., 2016).”
BackgroundMethodInfluentialAdvancing fairness in natural language processing : from traditional methods to explainability
Jourdan, F. (2024) · theses.fr (ABES) · doi:10.70675/b8535a05z82f1z46f7z9688z3cd35cea8381 · Cited by 5
Nomological Deductive Reasoning for Trustworthy, Human-Readable, and Actionable AI Outputs
Hakizimana, G., & Ledezma Espino, A. (2025) · Algorithms · doi:10.3390/a18060306 · Cited by 2
Multi-View Cooperative Learning with Invariant Rationale for Document-Level Relation Extraction
Lin, R., Fan, J., He, Y., Yang, Y., Li, J., & Guo, C. (2024) · Cognitive Computation · doi:10.1007/s12559-024-10322-z · Cited by 2
Explainable Multi-Hop Question Answering: A Rationale-Based Approach
Han, K., Jang, Y., & Kim, H. (2025) · Big Data and Cognitive Computing · doi:10.3390/bdcc9110273 · Cited by 3
“To address these limitations, we adopt a pointer-style rationale modeling approach that explicitly selects discrete spans or sentences from the input, yielding text-aligned evidence and providing transparency at the decision level [27,28].”
Shmuel, A., Lazebnik, T., & Glickman, O. (2026) · IEEE Access · doi:10.1109/access.2026.3657793 · Cited by 5
“While certain ML models like Random Forests (RF) are considered relatively interpretable [8], [9], others, such as Neural Networks (NN), are often viewed as complete ‘black boxes’ [10], [11].”
Robust Text Classifier on Test-Time Budgets
Parvez, M. R., Bolukbasi, T., Chang, K.-W., & Saligrama, V. (2019) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/d19-1108 · Cited by 3
“Similar to us, Lei et al. (2016) propose a selector - classifier framework to find text snip-pets as justification for text classification but their selector and classifier have similar complexity and require similar processing times; therefore, it is not suitable for computation gain.”
BackgroundMethodSentence Subjectivity and Sentiment Classification
Liu, B. (2020) · Cambridge University Press eBooks · doi:10.1017/9781108639286.005 · Cited by 3
Sentiment Analysis Tools and Techniques: A Review
Singh, H., Attwal, K. P. S., & Lal, M. (2022) · IEEE Annual Information Technology, Electronics and Mobile Communication Conference (IEMCON) · doi:10.1109/iemcon56893.2022.9946514 · Cited by 2
“Lei , Barzilay, and Jaakkola [12] to extract bits of input sentence as reviewing rating justifications (reasons), a neural network approach was presented.”
BackgroundModel Interpretability and Rationale Extraction by Input Mask Optimization
Brinner, M., & Zarrieß, S. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.867 · Cited by 8
Learning to Ignore Adversarial Attacks
Zhang, Y., Zhou, Y., Carton, S., & Tan, C. (2023) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.216 · Cited by 7
IAEval: A Comprehensive Evaluation of Instance Attribution on Natural Language Understanding
Gu, P., Shen, Y., Wang, L., Wang, Q., Wu, H., & Mao, Z. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.801 · Cited by 7
“Meanwhile, many studies give their views on the properties that an evidence should satisfy, such as compactness and sufficiency (Lei et al., 2016; DeYoung et al., 2020), comprehen-siveness (Kass et al., 1988; Yu et al., 2019), plausibility and faithfulness(Ding and Koehn, 2021; Jacovi and Goldberg,…”
Explainable Event Detection with Event Trigger Identification as Rationale Extraction
Hettiarachchi, H., & Ranasinghe, T. (2023) · International conference Recent advances in natural language processing · doi:10.26615/978-954-452-092-2_056 · Cited by 5
“To achieve this, we propose addressing event trigger detection as a rationale extraction task (Lei et al., 2016).”
MethodInterpretable Deep Image Classification Using Rationally Inattentive Utility Maximization
Pattanayak, K., Krishnamurthy, V., & Jain, A. (2024) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2024.3381335 · Cited by 2
“Anotherapproachseekstoprovidelocalapprox-imations to the trained model, local w.r.t the input image [26], [27].”
BackgroundFoundations and Obstacles of Deep Learning in Healthcare
Lakshmi Devi, S., S., V. L. P., & Nataraj, S. K. (2025) · IGI Global eBooks · doi:10.4018/979-8-3693-7868-7.ch006 · Cited by 2
Chakrapani, K., Safa, M. I., Moorthy, S. G., Kumaraswamy, M., & Parimala, G. (2025) · Wiley eBooks · doi:10.1002/9781394249312.ch24 · Cited by 3
Sentence Subjectivity and Sentiment Classification
Liu, B. (2015) · Cambridge University Press eBooks · doi:10.1017/cbo9781139084789.005 · Cited by 3
Identifying the Relative Importance of Customer Issues on Product Ratings through Machine Learning
Tiwari, H., Sait, S., Rizvi, M. I. H., & Damera-Venkata, N. (2018) · ACM Symposium on Document Engineering (DocEng) · doi:10.1145/3209280.3229113 · Cited by 2
“[6] develop the concept of “hard” attention to identify sentences that best explain a particular rating.”
BackgroundDSDR: Dynamic Semantic Discard Reader for Open-Domain Question Answering
Wang, X., Huang, Z., Zhang, Y., Tan, L., & Yang, L. (2018) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2018.8489553 · Cited by 2
“[34] Tao Lei, Regina Barzilay, and Tommi Jaakkola.”
BackgroundMethodConcept Matching for Low-Resource Classification
Errica, F., Silvestri, F., Edizel, B., Denoyer, L., Petroni, F., Plachouras, V., & Riedel, S. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9533640 · Cited by 2
“In particular, the method proposed by [7] tackles text classification by learning the distribution of annotations given the text and that of the target class given the annotations.”
BackgroundMethodInfluentialMCMH: Learning Multi-Chain Multi-Hop Rules for Knowledge Graph Reasoning
Zhang, L., Yu, M., Gao, T., & Yu, Y. (2020) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2020.findings-emnlp.351 · Cited by 3
Towards Unifying the Explainability Evaluation Methods for NLP
Lucaci, D., & Inkpen, D. Z. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-88483-3_23 · Cited by 2
Instance-Based Neural Dependency Parsing
Ouchi, H., Suzuki, J., Kobayashi, S., Yokoi, S., Kuribayashi, T., Yoshikawa, M., & Inui, K. (2021) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00439 · Cited by 3
“While deep neural networks have improved prediction accuracy in various tasks, rationales underlying the predictions have been more challenging for humans to understand (Lei et al., 2016).”
BackgroundExplaining Deep Neural Networks in medical imaging context
Rguibi, Z., Hajami, A., & Zitouni, D. (2021) · IEEE/ACS International Conference on Computer Systems and Applications (AICCSA) · doi:10.1109/aiccsa53542.2021.9686919 · Cited by 2
Explainability as statistical inference
Senetaire, H., Garreau, D., Frellsen, J., & Mattei, P. (2022) · HAL · doi:10.48550/arxiv.2212.03131 · Cited by 4
“Lei et al. (2016) focused on natural language processing (NLP) tasks and proposed to learn jointly a predictor and a selector.”
BackgroundMethodInfluentialDon't Use a Cannon to Kill a Fly: An Efficient Cascading Pipeline for Long Documents
Li, Z., Guha, N., & Nyarko, J. (2023) · International Conference on Artificial Intelligence and Law · doi:10.1145/3594536.3595142 · Cited by 10
“In this work, we present and study a novel workflow for performing legal long document analysis that is inspired by cascading classifiers [25, 37].”
MethodInfluentialMake Your Decision Convincing! A Unified Two-Stage Framework: Self-Attribution and Decision-Making
Du, Y., Zhao, S., Wang, H., Chen, Y., Bai, R., Qiang, Z., Cai, M., & Qin, B. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.78 · Cited by 3
Moral Sparks in Social Media Narratives
Xi, R., & Singh, M. P. (2025) · ACM Conference on Hypertext and Social Media (HT) · doi:10.1145/3720553.3746669 · Cited by 3
“We employ a rationalization process [1, 13, 31] to identify the key factors influencing moral judgment verdicts.”
Towards Prediction Explainability through Sparse CommunicationPreprint
Treviso, M., & Martins, A. F. T. (2020) · arXiv (Cornell University) · Cited by 2
Gârbacea, C., Guo, M., Carton, S., & Mei, Q. (2020) · arXiv (Cornell University) · Cited by 2
A Variational Approach to Unsupervised Sentiment AnalysisPreprint
Zeng, Z., Zhou, W., Liu, X., Lin, Z., Song, Y., Kuo, M., & Chiu, W. H. K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.09394 · Cited by 2
“We evaluate our model on TripAdvisor (Wang, Lu, & Zhai, 2010) and BeerAdvocate (McAuley, Leskovec, & Jurafsky, 2012; Lei, Barzilay, & Jaakkola, 2016; Yin, Song, & Zhang, 2017) datasets, which contain seven aspects (value, room, location, cleanliness, check-in or front desk, service, and business)…”
Towards explainable artificial intelligence (XAI)
Haverinen, T. (2020) · Jyväskylä University Digital Archive (University of Jyväskylä) · Cited by 3
Sentence-Based Model Agnostic NLP Interpretability.Preprint
Rychener, Y., Renard, X., Seddah, D., Frossard, P., & Detyniecki, M. (2020) · arXiv (Cornell University) · Cited by 2
Learning Stable Classifiers by Transferring Unstable FeaturesPreprint
Bao, Y., Chang, S., & Barzilay, R. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.07847 · Cited by 2
“For the target task T , the model can access Etrain1 (T ) and E val(T ). ρ(Z, Y ) MNIST BEER REVIEW ASK2ME WATERBIRD ODD EVEN LOOK AROMA PALATE PENE.”
MethodSelf-citationWhen Stability meets Sufficiency: Informative Explanations that do not Overwhelm
Luss, R., & Dhurandhar, A. (2024) · Transactions on machine learning research · doi:10.48550/arxiv.2109.06181 · Cited by 3
“Most relevant to our paper are local sufficiency based methods [8, 10, 29, 22, 27, 12, 4, 3, 37] which highlight features that are sufficient to obtain the same classification as the original input.”
Making Document-Level Information Extraction Right for the Right ReasonsPreprint
Tang, L., Rajan, D., Mohan, S., Pradhan, A., Bryan, R. N., & Durrett, G. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.07686 · Cited by 2
“mediate latent variables for extraction (Lei et al., 2016), we focus on what can be done with offthe-shelf pre-trained models (Liu et al.”
BackgroundMethodInfluentialTowards Human-Centred Explainability Benchmarks For Text ClassificationPreprint
Schlegel, V., Guzman, E. M., & Batista-Navarro, R. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2211.05452 · Cited by 7
“…by the fact that jointly learning to both predict and explain, results in more robust classifiers that are less susceptible to weaknesses such as reliance on spurious correlations (Rajagopal et al. 2020; Dua, Singh, and Gardner 2020; Atanasova et al. 2020; Lei, Barzilay, and Jaakkola 2016).”
BackgroundBreaking Free from MMI: A New Frontier in Rationalization by Probing Input Utilization
Liu, W., Deng, Z., Niu, Z., Wang, J., Wang, H., Zeng, Z., & Li, R. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2503.06202 · Cited by 5
“The main baseline for direct comparison is the vanilla MMI-based rationalization framework RNP (Lei et al., 2016), as RNP and our N2R match in selection granularity, optimization algorithm, and model architecture, which helps us to focus on our claims rather than some potentially unknown mechanisms.”
InfluentialUnderstanding Learned Models by Identifying Important Features at the Right Resolution
Lee, K., Sood, A., & Craven, M. W. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.33014155 · Cited by 5
CAWA: An Attention-Network for Credit Attribution
Manchanda, S., & Karypis, G. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i05.6367 · Cited by 1
Interpreting Randomly Wired Graph Models for Chinese NER
Chen, J., Xu, J., Xi, X., Cui, Z., & Sheng, V. S. (2022) · Computer Modeling in Engineering & Sciences · doi:10.32604/cmes.2022.020771 · Cited by 1
MI-VFL: Feature discrepancy-aware distributed model interpretation for vertical federated learning
Xing, R., Zheng, Z., Li, Q., Wu, F., & Chen, G. (2025) · Computer Networks · doi:10.1016/j.comnet.2025.111220 · Cited by 1
Generating Natural Language Explanations for LMM-Driven Robot Behavior
Jarsania, J., Singh, M. P., & Harsh, F. (2025) · IEEE World AI IoT Congress (AIIoT) · doi:10.1109/aiiot65859.2025.11105251 · Cited by 1
“• Natural Language Explanation Generation: Research exists on generating NLEs for standard machine learning models, particularly in NLP [7] and computer vision [8].”
ruSciFact: Open Benchmark for Verifying Scientific Facts in Russian
Vatolin, A., Gerasimenko, N. A., Loukachevitch, N., Ianina, A., & Vorontsov, K. V. (2025) · Computational Linguistics and Intellectual Technologies · doi:10.28995/2075-7182-2025-23-435-459 · Cited by 3
“All the annotations in SciFact are provided with rationales (Lei et al., 2016), so each SUPPORTS / REFUTES decision has a reasoned justification.”
Angara, D., & Yellapu, J. (2025) · Journal of Information & Knowledge Management · doi:10.1142/s0219649225501163 · Cited by 1
Wang, C., Xiaoguang, G. A. O., Wang, Z., Li, B., Wan, K., Li, X., & He, C. (2025) · Knowledge-Based Systems · doi:10.1016/j.knosys.2025.114484 · Cited by 1
Reproducible synthetic clinical letters for seizure frequency information extraction
Gan, Y., Barlow, S. H., Holgate, B., Davies, J., Teo, J., Winston, J. S., & Richardson, M. P. (2026) · Journal of Biomedical Informatics · doi:10.1016/j.jbi.2026.105105 · Cited by 1
Luo, X. (2025) · Journal of Technology Informatics and Engineering · doi:10.51903/jtie.v4i2.505 · Cited by 1
IBIS-Net: An iterative bio-inspired selection network for interpretable text classification
Wang, Y., & Todo, Y. (2026) · Knowledge-Based Systems · doi:10.1016/j.knosys.2026.116047 · Cited by 1
Document Image Analysis: A Review of Existing Approaches and Future Scope
Vengalil, S. K., Rohan, M., Nagaraj, K., Sahasra, P. V. S., & Nannuri, S. L. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-19672-9_5 · Cited by 1
What You Read Is What You Classify: Highlighting Attributions to Text and Text-like Inputs
Berman, D. S., Merritt, B. B., Ta, S., Udwin, D., Ernlund, A., Ratcliff, J., & Narayan, V. (2026) · AI · doi:10.3390/ai7050168 · Cited by 1
Metadata
- Authors
- Lei, T., Barzilay, R., & Jaakkola, T.
- Affiliations
- MIT Computer Science and Artificial Intelligence Laboratory (US)
- Venue
- Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing
- Published
- 2016
- Pages
- 107-117
- OpenAlex
- W2440159969
- Cited by
- 777
- References
- 27
- Citations by year
- 2026: 27 · 2025: 33 · 2024: 39 · 2023: 88 · 2022: 92 · 2021: 138 · 2020: 126 · 2019: 101 · 2018: 78 · 2017: 39 · 2016: 4
- FWCI
- 39.87
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc-sa)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Physical Sciences › Computer Science › Artificial Intelligence › Sentiment Analysis and Opinion Mining
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- explanation generation, question retrieval, model regularization
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.