---
title: "An Information Bottleneck Approach for Controlling Conciseness"
description: "Rationale-based models make predictions from a selected part of the input text, which makes them… 35 references, cited by 51 papers here."
authors: ["Bhargavi Paranjape", "Mandar Joshi", "John Thickstun", "Hannaneh Hajishirzi", "Luke Zettlemoyer"]
year: 2020
venue: "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)"
source: "https://aclanthology.org/2020.emnlp-main.153/"
doi: "10.18653/v1/2020.emnlp-main.153"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/paranjape-2020-information-bottleneck-approach-controlling-conciseness/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/paranjape-2020-information-bottleneck-approach-controlling-conciseness/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/4/) / An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction

# An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction

**Authors:** Bhargavi Paranjape, [Mandar Joshi](https://md.research.elamur.ai/free/en/authors/mandar-joshi/), [John Thickstun](https://md.research.elamur.ai/free/en/authors/john-thickstun/), [Hannaneh Hajishirzi](https://md.research.elamur.ai/free/en/authors/hannaneh-hajishirzi/) and [Luke Zettlemoyer](https://md.research.elamur.ai/free/en/authors/luke-zettlemoyer/) · **Year:** 2020 · **Venue:** [Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)](https://md.research.elamur.ai/free/en/venues/emnlp/) ([EMNLP 2020](https://md.research.elamur.ai/free/en/venues/emnlp/2020/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2020.emnlp-main.153/) · [PDF](https://aclanthology.org/2020.emnlp-main.153.pdf) · [DOI 10.18653/v1/2020.emnlp-main.153](https://doi.org/10.18653/v1/2020.emnlp-main.153) · [OpenAlex](https://openalex.org/W3021216956)

**Citations:** 86 citations (OpenAlex, 2026-10-14) · [51](https://md.research.elamur.ai/free/en/papers/paranjape-2020-information-bottleneck-approach-controlling-conciseness/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
Rationale-based models make predictions from a selected part of the input text, which makes them easier to interpret. They usually lose accuracy compared with models that read the whole input, and norm-based sparsity penalties (L0/L1) make this trade-off hard to tune. The authors reformulate rationale extraction with the Information Bottleneck (IB) principle. They derive a variational bound in which the bottleneck is a binary mask over sentences, not a compressed vector. An explainer (a pre-trained BERT-style transformer) estimates the probability of keeping each sentence. A separate end-task predictor sees only the kept sentences, so its predictions are faithful to the rationale. Sparsity is set by a KL term that pulls the per-sentence Bernoulli mask distribution toward a fixed prior π, which sets the target fraction of retained text. On five ERASER tasks and BeerAdvocate, this "Sparse IB" objective beats norm-minimization baselines in both task scores and agreement with human rationales. It also matches the requested sparsity levels more reliably, giving finer control over the trade-off between conciseness and accuracy.

## Key points
- Compared with norm-minimization methods, Sparse IB gives up to 5% relative gain in task metrics and 6–80% relative gain in agreement with human rationale annotations.
- Without rationale supervision, the model stays within about 10% of a same-size model that uses the full input. Adding gold masks for roughly 25% of training examples closes this gap for most tasks.
- The prior π can be estimated from the share of text humans mark as rationale, or tuned directly (assumed to lie in (0, 0.5)). Unlike the ||m||₀ penalty, the KL objective allows sparsity to vary from example to example.
- The method does not depend on a particular architecture. In the word-level BERT+LSTM setting from ERASER, the Movies task score rises from 91.96 (Sparse Norm-C) to 93.46.
- Error analysis shows failures when the context contains evidence for both labels, and on shallow lexical matches and confusions requiring world knowledge. Code: github.com/bhargaviparanjape/explainable_qa.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [University of Washington](https://md.research.elamur.ai/free/en/universities/university-of-washington/) — University, Seattle, United States · 263 works · 180,581 citations
- [Allen Institute for Artificial Intelligence](https://md.research.elamur.ai/free/en/labs/allen-institute-for-artificial-intelligence/) — Research institute, Seattle, United States · 219 works · 89,712 citations
- [Allen Institute](https://md.research.elamur.ai/free/en/labs/allen-institute/) — Research institute, Seattle, United States · 5 works · 4,252 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works

## Cite

Cite as: Bhargavi Paranjape, Mandar Joshi, John Thickstun, Hannaneh Hajishirzi and Luke Zettlemoyer. 2020. An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{paranjape2020information,
  title = {{An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction}},
  author = {Bhargavi Paranjape, Mandar Joshi, John Thickstun, Hannaneh Hajishirzi and Luke Zettlemoyer},
  booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year = {2020},
  doi = {10.18653/v1/2020.emnlp-main.153},
  url = {https://doi.org/10.18653/v1/2020.emnlp-main.153}
}
```

- [← previous A Report on the 2020 VUA and TOEFL Metaphor Detection Shared Task](https://md.research.elamur.ai/free/en/papers/leong-2020-report-2020-vua-toefl-metaphor/)
- [next → ConvoKit: A Toolkit for the Analysis of Conversations](https://md.research.elamur.ai/free/en/papers/chang-2020-convokit-toolkit-analysis-conversations/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/4/)

## References

35 references, 16 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Alexander A Alemi, Ian Fischer, Joshua V Dillon, and Kevin Murphy. 2016. Deep variational information bottleneck. arXiv preprint arXiv:1612.00410. — [arXiv:1612.00410](https://arxiv.org/abs/1612.00410) [find](https://scholar.google.com/scholar?q=Alexander+A+Alemi%2C+Ian+Fischer%2C+Joshua+V+Dillon%2C+and+Kevin+Murphy%2E+2016%2E+Deep+variational+information+bottleneck%2E+arXiv+preprint+arXiv%3A1612%2E00410%2E+%E2%80%94+%5BarXiv%3A1612%2E00410%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1612%2E00410)
2. David Alvarez-Melis and Tommi Jaakkola. 2017. A causal framework for explaining the predictions of black-box sequence-to-sequence models. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 412– 421. [find](https://scholar.google.com/scholar?q=David+Alvarez%2DMelis+and+Tommi+Jaakkola%2E+2017%2E+A+causal+framework+for+explaining+the+predictions+of+black%2Dbox+sequence%2Dto%2Dsequence+models%2E+In+Proceedings+of+the+2017+Conference+on+Empirical+Methods+in+)
3. Seojin Bang, Pengtao Xie, Heewook Lee, Wei Wu, and Eric Xing. 2019. Explaining a black-box using deep variational information bottleneck approach. arXiv preprint arXiv:1902.06918. — [arXiv:1902.06918](https://arxiv.org/abs/1902.06918) [find](https://scholar.google.com/scholar?q=Seojin+Bang%2C+Pengtao+Xie%2C+Heewook+Lee%2C+Wei+Wu%2C+and+Eric+Xing%2E+2019%2E+Explaining+a+black%2Dbox+using+deep+variational+information+bottleneck+approach%2E+arXiv+preprint+arXiv%3A1902%2E06918%2E+%E2%80%94+%5BarXiv%3A1902%2E06918%5D)
4. Joost Bastings, Wilker Aziz, and Ivan Titov. 2019. [Interpretable neural predictions with differentiable binary variables](https://md.research.elamur.ai/free/en/papers/bastings-2019-interpretable-neural-predictions-differentiable-binary/). In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2963–2977.
5. Iz Beltagy, Kyle Lo, and Arman Cohan. 2019. [Scibert: A pretrained language model for scientific text](https://md.research.elamur.ai/free/en/papers/beltagy-2019-scibert-pretrained-language-model-scientific/). arXiv preprint arXiv:1903.10676.
6. Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom. 2018. [e-snli: Natural language inference with natural language explanations](https://md.research.elamur.ai/free/en/papers/camburu-2018-e-snli-natural-language-inference/). In Advances in Neural Information Processing Systems, pages 9539–9549.
7. Shiyu Chang, Yang Zhang, Mo Yu, and Tommi Jaakkola. 2019. A game theoretic approach to class-wise selective rationalization. In Advances in Neural Information Processing Systems, pages 10055– 10065. [find](https://scholar.google.com/scholar?q=Shiyu+Chang%2C+Yang+Zhang%2C+Mo+Yu%2C+and+Tommi+Jaakkola%2E+2019%2E+A+game+theoretic+approach+to+class%2Dwise+selective+rationalization%2E+In+Advances+in+Neural+Information+Processing+Systems%2C+pages+10055%E2%80%93+10065%2E)
8. Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. 2019. [Boolq: Exploring the surprising difficulty of natural yes/no questions](https://md.research.elamur.ai/free/en/papers/clark-2019-boolq-exploring-surprising-difficulty-natural/). arXiv preprint arXiv:1905.10044.
9. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. [Bert: Pre-training of deep bidirectional transformers for language understanding](https://md.research.elamur.ai/free/en/papers/devlin-2019-bert-pre-training-deep-bidirectional/). In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
10. Jay DeYoung, Sarthak Jain, Nazneen Fatema Rajani, Eric Lehman, Caiming Xiong, Richard Socher, and Byron C Wallace. 2019. [Eraser: A benchmark to evaluate rationalized nlp models](https://md.research.elamur.ai/free/en/papers/deyoung-2020-eraser-benchmark-evaluate-rationalized-nlp/). arXiv preprint arXiv:1911.03429.
11. SG Fletcher and K Ponnambalam. 1996. Estimation of reservoir yield and storage distribution using moments analysis. Journal of Hydrology(Amsterdam), 182(1):259–275. [find](https://scholar.google.com/scholar?q=SG+Fletcher+and+K+Ponnambalam%2E+1996%2E+Estimation+of+reservoir+yield+and+storage+distribution+using+moments+analysis%2E+Journal+of+Hydrology%28Amsterdam%29%2C+182%281%29%3A259%E2%80%93275%2E)
12. Emil Julius Gumbel. 1948. Statistical theory of extreme values and some practical applications: a series of lectures, volume 33. US Government Printing Office. [find](https://scholar.google.com/scholar?q=Emil+Julius+Gumbel%2E+1948%2E+Statistical+theory+of+extreme+values+and+some+practical+applications%3A+a+series+of+lectures%2C+volume+33%2E+US+Government+Printing+Office%2E)
13. Sarthak Jain, Sarah Wiegreffe, Yuval Pinter, and Byron C Wallace. 2020. [Learning to faithfully rationalize by construction](https://md.research.elamur.ai/free/en/papers/jain-2020-learning-faithfully-rationalize-construction/). arXiv preprint arXiv:2005.00115.
14. Eric Jang, Shixiang Gu, and Ben Poole. 2017. Categorical reparametrization with gumble-softmax. In International Conference on Learning Representations (ICLR 2017). OpenReview. net. [find](https://scholar.google.com/scholar?q=Eric+Jang%2C+Shixiang+Gu%2C+and+Ben+Poole%2E+2017%2E+Categorical+reparametrization+with+gumble%2Dsoftmax%2E+In+International+Conference+on+Learning+Representations+%28ICLR+2017%29%2E+OpenReview%2E+net%2E)
15. Daniel Khashabi, Snigdha Chaturvedi, Michael Roth, Shyam Upadhyay, and Dan Roth. 2018. [Looking beyond the surface: A challenge set for reading comprehension over multiple sentences](https://md.research.elamur.ai/free/en/papers/khashabi-2018-looking-beyond-surface-challenge-set/). In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics:…
16. Durk P Kingma, Tim Salimans, and Max Welling. 2015. Variational dropout and the local reparameterization trick. In Advances in neural information processing systems, pages 2575–2583. [find](https://scholar.google.com/scholar?q=Durk+P+Kingma%2C+Tim+Salimans%2C+and+Max+Welling%2E+2015%2E+Variational+dropout+and+the+local+reparameterization+trick%2E+In+Advances+in+neural+information+processing+systems%2C+pages+2575%E2%80%932583%2E)
17. Josua Krause, Aritra Dasgupta, Jordan Swartz, Yindalon Aphinyanaphongs, and Enrico Bertini. 2017. A workflow for visual diagnostics of binary classifiers using instance-level explanations. In 2017 IEEE Conference on Visual Analytics Science and Technology (VAST), pages 162–172. IEEE. [find](https://scholar.google.com/scholar?q=Josua+Krause%2C+Aritra+Dasgupta%2C+Jordan+Swartz%2C+Yindalon+Aphinyanaphongs%2C+and+Enrico+Bertini%2E+2017%2E+A+workflow+for+visual+diagnostics+of+binary+classifiers+using+instance%2Dlevel+explanations%2E+In+2017+IEE)
18. Veronica Latcinnik and Jonathan Berant. 2020. Explaining question answering models through text generation. arXiv preprint arXiv:2004.05569. — [arXiv:2004.05569](https://arxiv.org/abs/2004.05569) [find](https://scholar.google.com/scholar?q=Veronica+Latcinnik+and+Jonathan+Berant%2E+2020%2E+Explaining+question+answering+models+through+text+generation%2E+arXiv+preprint+arXiv%3A2004%2E05569%2E+%E2%80%94+%5BarXiv%3A2004%2E05569%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2004%2E05569%29)
19. Eric Lehman, Jay DeYoung, Regina Barzilay, and Byron C Wallace. 2019. [Inferring which medical treatments work from reports of clinical trials](https://md.research.elamur.ai/free/en/papers/lehman-2019-inferring-which-medical-treatments-work/). In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (…
20. Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2016. [Rationalizing neural predictions](https://md.research.elamur.ai/free/en/papers/lei-2016-rationalizing-neural-predictions/). In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 107–117.
21. Xiang Lisa Li and Jason Eisner. 2019. Specializing word embeddings (for parsing) by information bottleneck. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 2744–2… [find](https://scholar.google.com/scholar?q=Xiang+Lisa+Li+and+Jason+Eisner%2E+2019%2E+Specializing+word+embeddings+%28for+parsing%29+by+information+bottleneck%2E+In+Proceedings+of+the+2019+Conference+on+Empirical+Methods+in+Natural+Language+Processing+an)
22. Julian McAuley, Jure Leskovec, and Dan Jurafsky. 2012. Learning attitudes and attributes from multi-aspect reviews. In 2012 IEEE 12th International Conference on Data Mining, pages 1020–1025. IEEE. [find](https://scholar.google.com/scholar?q=Julian+McAuley%2C+Jure+Leskovec%2C+and+Dan+Jurafsky%2E+2012%2E+Learning+attitudes+and+attributes+from+multi%2Daspect+reviews%2E+In+2012+IEEE+12th+International+Conference+on+Data+Mining%2C+pages+1020%E2%80%931025%2E+IEEE%2E)
23. Eric Nalisnick and Padhraic Smyth. 2017. Stick-breaking variational autoencoders. In International Conference on Learning Representations (ICLR). [find](https://scholar.google.com/scholar?q=Eric+Nalisnick+and+Padhraic+Smyth%2E+2017%2E+Stick%2Dbreaking+variational+autoencoders%2E+In+International+Conference+on+Learning+Representations+%28ICLR%29%2E)
24. Bo Pang and Lillian Lee. 2004. [A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts](https://md.research.elamur.ai/free/en/papers/pang-2004-sentimental-education-sentiment-analysis-subjectivity/). In Proceedings of the 42nd annual meeting on Association for Computational Linguistics, page 271. Association for Computational Linguistics.
25. Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 4085–4094. [find](https://scholar.google.com/scholar?q=Empirical+Methods+in+Natural+Language+Processing+and+the+9th+International+Joint+Conference+on+Natural+Language+Processing+%28EMNLP%2DIJCNLP%29%2C+pages+4085%E2%80%934094%2E)
26. Andrey Zhmoginov, Ian Fischer, and Mark Sandler. 2019. Information-bottleneck approach to salient region discovery. arXiv preprint arXiv:1907.09578. — [arXiv:1907.09578](https://arxiv.org/abs/1907.09578) [find](https://scholar.google.com/scholar?q=Andrey+Zhmoginov%2C+Ian+Fischer%2C+and+Mark+Sandler%2E+2019%2E+Information%2Dbottleneck+approach+to+salient+region+discovery%2E+arXiv+preprint+arXiv%3A1907%2E09578%2E+%E2%80%94+%5BarXiv%3A1907%2E09578%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1907%2E095)
27. Nazneen Fatema Rajani, Bryan McCann, Caiming Xiong, and Richard Socher. 2019. [Explain yourself! leveraging language models for commonsense reasoning](https://md.research.elamur.ai/free/en/papers/rajani-2019-explain-yourself-leveraging-language-models/). In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4932–4942.
28. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. [” why should i trust you?” explaining the predictions of any classifier](https://md.research.elamur.ai/free/en/papers/ribeiro-2016-why-should-i-trust-explaining/). In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 1135–1144.
29. Sofia Serrano and Noah A Smith. 2019. [Is attention interpretable?](https://md.research.elamur.ai/free/en/papers/serrano-2019-is-attention-interpretable/) In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2931–2951.
30. James Thorne, Andreas Vlachos, Christos Christodoulopoulos, and Arpit Mittal. 2018. [Fever: a large-scale dataset for fact extraction and verification](https://md.research.elamur.ai/free/en/papers/thorne-2018-fever-large-scale-dataset-fact/). In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vo…
31. Naftali Tishby, Fernando C. Pereira, and William Bialek. 1999. The information bottleneck method. In Proc. of the 37-th Annual Allerton Conference on Communication, Control and Computing, pages 368–377. [find](https://scholar.google.com/scholar?q=Naftali+Tishby%2C+Fernando+C%2E+Pereira%2C+and+William+Bialek%2E+1999%2E+The+information+bottleneck+method%2E+In+Proc%2E+of+the+37%2Dth+Annual+Allerton+Conference+on+Communication%2C+Control+and+Computing%2C+pages+368%E2%80%9337)
32. Daniel S Weld and Gagan Bansal. 2019. The challenge of crafting intelligible intelligence. Communications of the ACM, 62(6):70–79. [find](https://scholar.google.com/scholar?q=Daniel+S+Weld+and+Gagan+Bansal%2E+2019%2E+The+challenge+of+crafting+intelligible+intelligence%2E+Communications+of+the+ACM%2C+62%286%29%3A70%E2%80%9379%2E)
33. Peter West, Ari Holtzman, Jan Buys, and Yejin Choi. 2019. Bottlesum: Unsupervised and self-supervised sentence summarization using the information bottleneck principle. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferenc… [find](https://scholar.google.com/scholar?q=Peter+West%2C+Ari+Holtzman%2C+Jan+Buys%2C+and+Yejin+Choi%2E+2019%2E+Bottlesum%3A+Unsupervised+and+self%2Dsupervised+sentence+summarization+using+the+information+bottleneck+principle%2E+In+Proceedings+of+the+2019+Conf)
34. Ronald J Williams. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Machine learning, 8(3-4):229–256. [find](https://scholar.google.com/scholar?q=Ronald+J+Williams%2E+1992%2E+Simple+statistical+gradient%2Dfollowing+algorithms+for+connectionist+reinforcement+learning%2E+Machine+learning%2C+8%283%2D4%29%3A229%E2%80%93256%2E)
35. Mo Yu, Shiyu Chang, Yang Zhang, and Tommi Jaakkola. 2019. [Rethinking cooperative rationalization: Introspective extraction and complement control](https://md.research.elamur.ai/free/en/papers/yu-2019-rethinking-cooperative-rationalization-introspective-extraction/). In Proceedings of the 2019 Conference on

*Updated 24 October 2026*