---
title: "Distinguish Confusing Law Articles for Legal — Xu 2020"
description: "Legal Judgment Prediction (LJP) means inferring a case's applicable law article, charge and term of penalty from its written fact description."
authors: ["Nuo Xu", "Pinghui Wang", "Long Chen", "Li Pan", "Xiaoyan Wang", "Junzhou Zhao"]
year: 2020
venue: "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics"
source: "https://aclanthology.org/2020.acl-main.280/"
doi: "10.18653/v1/2020.acl-main.280"
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/xu-2020-distinguish-confusing-law-articles-legal/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/xu-2020-distinguish-confusing-law-articles-legal/)

[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/2/) / Distinguish Confusing Law Articles for Legal Judgment Prediction

# Distinguish Confusing Law Articles for Legal Judgment Prediction

**Authors:** [Nuo Xu](https://md.research.elamur.ai/free/en/authors/nuo-xu/), Pinghui Wang, Long Chen, Li Pan, Xiaoyan Wang, Junzhou Zhao · **Year:** 2020 · **Venue:** Proceedings of the 58th Annual Meeting of the 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.acl-main.280/) · [PDF](https://aclanthology.org/2020.acl-main.280.pdf) · [DOI 10.18653/v1/2020.acl-main.280](https://doi.org/10.18653/v1/2020.acl-main.280) · [OpenAlex](https://openalex.org/W3034892514)

**Citations:** 172 citations (openalex-snapshot · 2026-10-14) · [86](#cited-by) in our library, of the 140884 papers whose bibliographies we read

## Summary
Legal Judgment Prediction (LJP) means inferring a case's applicable law article, charge and term of penalty from its written fact description. A recurring failure is confusion between near-identical statutes. For example, Article 385 and Article 163 both cover accepting bribes and differ mainly in whether the offender is a state employee. An earlier remedy relied on expert-defined discriminative attributes, which do not transfer easily to other legal systems.

The authors propose LADAN (Law Article Distillation based Attention Network), an end-to-end model with three stages:
- A graph construction layer groups law articles into communities of highly similar items.
- A new graph distillation operator (GDO) isolates what separates the articles inside each community, instead of averaging over neighbours.
- A macro-level representation of the case selects a community. That community's distinction vector then steers attention over the fact text to extract fine-grained cues.

LADAN was tested on the Chinese CAIL-small and CAIL-big datasets and attached to several multi-task backbones (MTL, TOPJUDGE, MPBFN). It outperforms prior systems, raising F1 by up to 5.79%. The work shows that differences between statutes can be learned automatically, without hand-crafted legal features.

## Key points
- Each case is labelled with three outputs: law article, charge and penalty term. The authors assume one applicable article per case, following earlier work.
- On CAIL-small, LADAN+MTL reaches the following scores:
  - Law articles: 81.20 accuracy, 76.47 F1.
  - Charges: 85.07 accuracy, 83.14 F1.
  - Term of penalty: 38.29 accuracy, 32.65 F1.
- On CAIL-big, the LADAN variants reach about 96.6 accuracy on law articles. All methods score higher on CAIL-big than on CAIL-small, which the authors attribute to its larger training set.
- Ablation on CAIL-small:
  - Removing both the graph construction layer and GDO (τ = 1.0) cuts law-article accuracy to 79.79 and penalty accuracy to 34.87. This is roughly the level of HARNN+MTL.
  - The graph construction layer matters more than GDO, because distillation is less useful when the communities are inaccurate.
- In the attention visualizations, the basic encoder highlights shared bribery terms such as "use position" and "accept benefit". The re-encoder instead attends to the defendant's identity, such as "company manager" or a government bureau post.

## Interesting ideas
- **Distillation instead of aggregation in a GNN.** Standard message passing tends to make neighbouring node representations look alike (over-smoothing). Explicitly extracting what differs from neighbours is a useful alternative when the goal is to tell similar classes apart, for example fine-grained taxonomies or near-duplicate categories.
- **Two-level decision: pick the group, then discriminate inside it.** Coarse features first choose a cluster of similar labels, and specialized features then separate members within it. This hierarchical approach can help any large label set that contains many confusable labels, such as tariff codes or place-name disambiguation.
- **Use label descriptions, not just label IDs.** Treating each statute's text as a document and learning from the contrasts between those texts injects label semantics automatically. This replaces hand-designed attributes and transfers to any task where categories come with textual definitions.
- **Plug-in design across backbones.** LADAN was attached to three different multi-task frameworks and improved each of them. This is a good evaluation pattern for showing that a component's benefit is not tied to one architecture.
- **Naive text-differencing as a cautionary baseline.** Simply deleting the text two articles share can leave different articles with nearly identical leftovers. Checking whether "difference" features collapse across classes is a worthwhile sanity check when designing contrastive features.

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

Fields: [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/) · [Law](https://md.research.elamur.ai/free/en/shelves/law/)

## Cited by

Cited by

86 papers of our library cite it (of the 140884 whose bibliographies we read); 3 of them have a page here, 83 not yet.

- [ALJP: An Arabic Legal Judgment Prediction in Personal Status Cases Using Machine Learning Models](https://md.research.elamur.ai/free/en/papers/abbara-2023-aljp-arabic-legal-judgment-prediction/)
- [Lawyer LLaMA Technical Report](https://md.research.elamur.ai/free/en/papers/huang-2023-lawyer-llama-technical-report/)
- [ILDC for CJPE: Indian Legal Documents Corpus for Court Judgment Prediction and Explanation](https://md.research.elamur.ai/free/en/papers/malik-2021-ildc-cjpe-indian-legal-documents/)

## Cite

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

```bibtex
@inproceedings{xu2020distinguish,
  title = {{Distinguish Confusing Law Articles for Legal Judgment Prediction}},
  author = {Xu, Nuo and Wang, Pinghui and Chen, Long and Pan, Li and Wang, Xiaoyan and Zhao, Junzhou},
  booktitle = {Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
  year = {2020},
  doi = {10.18653/v1/2020.acl-main.280},
  url = {https://doi.org/10.18653/v1/2020.acl-main.280}
}
```

- [← previous Distilling Neural Networks for Greener and Faster Dependency Parsing](https://md.research.elamur.ai/free/en/papers/anderson-2020-distilling-neural-networks-greener-faster/)
- [next → Document Ranking with a Pretrained Sequence-to-Sequence Model](https://md.research.elamur.ai/free/en/papers/nogueira-2020-document-ranking-pretrained-sequence-sequence/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/2/)

## References

21 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Stephen Bonner, Ibad Kureshi, John Brennan, Georgios Theodoropoulos, Andrew Stephen McGough, and Boguslaw Obara. 2019. Exploring the semantic content of unsupervised graph embeddings: an empirical study. Data Science and Engineering, 4(3):269– 289. [find](https://scholar.google.com/scholar?q=Stephen+Bonner%2C+Ibad+Kureshi%2C+John+Brennan%2C+Georgios+Theodoropoulos%2C+Andrew+Stephen+McGough%2C+and+Boguslaw+Obara%2E+2019%2E+Exploring+the+semantic+content+of+unsupervised+graph+embeddings%3A+an+empirical+stu)
2. Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. 2017. Neural message passing for quantum chemistry. In ICML. [find](https://scholar.google.com/scholar?q=Justin+Gilmer%2C+Samuel+S+Schoenholz%2C+Patrick+F+Riley%2C+Oriol+Vinyals%2C+and+George+E+Dahl%2E+2017%2E+Neural+message+passing+for+quantum+chemistry%2E+In+ICML%2E) [our card](https://md.research.elamur.ai/free/en/papers/gilmer-2017-neural-message-passing-quantum-chemistry/)
3. Will Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. In NeurIPS. [find](https://scholar.google.com/scholar?q=Will+Hamilton%2C+Zhitao+Ying%2C+and+Jure+Leskovec%2E+2017%2E+Inductive+representation+learning+on+large+graphs%2E+In+NeurIPS%2E) [our card](https://md.research.elamur.ai/free/en/papers/hamilton-2017-inductive-representation-learning-large-graphs/)
4. Zikun Hu, Xiang Li, Cunchao Tu, Zhiyuan Liu, and Maosong Sun. 2018. Few-shot charge prediction with discriminative legal attributes. In COLING. [find](https://scholar.google.com/scholar?q=Zikun+Hu%2C+Xiang+Li%2C+Cunchao+Tu%2C+Zhiyuan+Liu%2C+and+Maosong+Sun%2E+2018%2E+Few%2Dshot+charge+prediction+with+discriminative+legal+attributes%2E+In+COLING%2E) [our card](https://md.research.elamur.ai/free/en/papers/hu-2018-few-shot-charge-prediction-discriminative/)
5. R Keown. 1980. Mathematical models for legal prediction. Computer/lj, 2:829. [find](https://scholar.google.com/scholar?q=R+Keown%2E+1980%2E+Mathematical+models+for+legal+prediction%2E+Computer%2Flj%2C+2%3A829%2E)
6. Yoon Kim. 2014. Convolutional neural networks for sentence classification. In EMNLP. [find](https://scholar.google.com/scholar?q=Yoon+Kim%2E+2014%2E+Convolutional+neural+networks+for+sentence+classification%2E+In+EMNLP%2E) [our card](https://md.research.elamur.ai/free/en/papers/kim-2014-convolutional-neural-networks-sentence-classification/)
7. Benjamin E Lauderdale and Tom S Clark. 2012. The supreme court’s many median justices. American Political Science Review, 106(4):847–866. [find](https://scholar.google.com/scholar?q=Benjamin+E+Lauderdale+and+Tom+S+Clark%2E+2012%2E+The+supreme+court%E2%80%99s+many+median+justices%2E+American+Political+Science+Review%2C+106%284%29%3A847%E2%80%93866%2E)
8. Qimai Li, Zhichao Han, and Xiao-Ming Wu. 2018. Deeper insights into graph convolutional networks for semi-supervised learning. In AAAI. [find](https://scholar.google.com/scholar?q=Qimai+Li%2C+Zhichao+Han%2C+and+Xiao%2DMing+Wu%2E+2018%2E+Deeper+insights+into+graph+convolutional+networks+for+semi%2Dsupervised+learning%2E+In+AAAI%2E)
9. Haoxi Zhong, Guo Zhipeng, Cunchao Tu, Chaojun Xiao, Zhiyuan Liu, and Maosong Sun. 2018. Legal judgment prediction via topological learning. In EMNLP. [find](https://scholar.google.com/scholar?q=Haoxi+Zhong%2C+Guo+Zhipeng%2C+Cunchao+Tu%2C+Chaojun+Xiao%2C+Zhiyuan+Liu%2C+and+Maosong+Sun%2E+2018%2E+Legal+judgment+prediction+via+topological+learning%2E+In+EMNLP%2E) [our card](https://md.research.elamur.ai/free/en/papers/zhong-2018-legal-judgment-prediction-topological-learning/)
10. Wan-Chen Lin, Tsung-Ting Kuo, Tung-Jia Chang, Chueh-An Yen, Chao-Ju Chen, and Shou-de Lin. 2012. Exploiting machine learning models for chinese legal documents labeling, case classification, and sentencing prediction. Processdings of RO-CLING. [find](https://scholar.google.com/scholar?q=Wan%2DChen+Lin%2C+Tsung%2DTing+Kuo%2C+Tung%2DJia+Chang%2C+Chueh%2DAn+Yen%2C+Chao%2DJu+Chen%2C+and+Shou%2Dde+Lin%2E+2012%2E+Exploiting+machine+learning+models+for+chinese+legal+documents+labeling%2C+case+classification%2C+and+sente)
11. Chao-Lin Liu, Cheng-Tsung Chang, and Jim-How Ho. 2004. Case instance generation and refinement for case-based criminal summary judgments in chinese. [find](https://scholar.google.com/scholar?q=Chao%2DLin+Liu%2C+Cheng%2DTsung+Chang%2C+and+Jim%2DHow+Ho%2E+2004%2E+Case+instance+generation+and+refinement+for+case%2Dbased+criminal+summary+judgments+in+chinese%2E)
12. Bingfeng Luo, Yansong Feng, Jianbo Xu, Xiang Zhang, and Dongyan Zhao. 2017. Learning to predict charges for criminal cases with legal basis. arXiv preprint arXiv:1707.09168. [↗](https://arxiv.org/abs/1707.09168)
13. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. In NeurIPS. [find](https://scholar.google.com/scholar?q=Tomas+Mikolov%2C+Ilya+Sutskever%2C+Kai+Chen%2C+Greg+S+Corrado%2C+and+Jeff+Dean%2E+2013%2E+Distributed+representations+of+words+and+phrases+and+their+compositionality%2E+In+NeurIPS%2E)
14. Stuart S Nagel. 1963. Applying correlation analysis to case prediction. Tex. L. Rev., 42:1006. [find](https://scholar.google.com/scholar?q=Stuart+S+Nagel%2E+1963%2E+Applying+correlation+analysis+to+case+prediction%2E+Tex%2E+L%2E+Rev%2E%2C+42%3A1006%2E)
15. Octavia-Maria Sulea, Marcos Zampieri, Shervin Malmasi, Mihaela Vela, Liviu P Dinu, and Josef van Genabith. 2017. Exploring the use of text classification in the legal domain. arXiv preprint arXiv:1710.09306. [↗](https://arxiv.org/abs/1710.09306)
16. Maosong Sun, Xinxiong Chen, Kaixu Zhang, Zhipeng Guo, and Zhiyuan Liu. 2016. Thulac: An efficient lexical analyzer for chinese. [find](https://scholar.google.com/scholar?q=Maosong+Sun%2C+Xinxiong+Chen%2C+Kaixu+Zhang%2C+Zhipeng+Guo%2C+and+Zhiyuan+Liu%2E+2016%2E+Thulac%3A+An+efficient+lexical+analyzer+for+chinese%2E)
17. Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. 2017. Graph attention networks. arXiv preprint arXiv:1710.10903. [↗](https://arxiv.org/abs/1710.10903) [our card](https://md.research.elamur.ai/free/en/papers/veli-2017-vckovic-graph-attention-networks/)
18. Chaojun Xiao, Haoxi Zhong, Zhipeng Guo, Cunchao Tu, Zhiyuan Liu, Maosong Sun, Yansong Feng, Xianpei Han, Zhen Hu, Heng Wang, et al. 2018. Cail2018: A large-scale legal dataset for judgment prediction. arXiv preprint arXiv:1807.02478. [↗](https://arxiv.org/abs/1807.02478) [our card](https://md.research.elamur.ai/free/en/papers/xiao-2018-cail2018-large-scale-legal-dataset/)
19. Nuo Xu, Pinghui Wang, Long Chen, Jing Tao, and Junzhou Zhao. 2019. Mr-gnn: Multi-resolution and dual graph neural network for predicting structured entity interactions. In IJCAI. [find](https://scholar.google.com/scholar?q=Nuo+Xu%2C+Pinghui+Wang%2C+Long+Chen%2C+Jing+Tao%2C+and+Junzhou+Zhao%2E+2019%2E+Mr%2Dgnn%3A+Multi%2Dresolution+and+dual+graph+neural+network+for+predicting+structured+entity+interactions%2E+In+IJCAI%2E)
20. Wenmian Yang, Weijia Jia, XIaojie Zhou, and Yutao Luo. 2019. Legal judgment prediction via multi-perspective bi-feedback network. arXiv preprint arXiv:1905.03969. [↗](https://arxiv.org/abs/1905.03969)
21. Zichao Yang, Diyi Yang, Chris Dyer, Xiaodong He, Alex Smola, and Eduard Hovy. 2016. Hierarchical attention networks for document classification. In NAACL. [find](https://scholar.google.com/scholar?q=Zichao+Yang%2C+Diyi+Yang%2C+Chris+Dyer%2C+Xiaodong+He%2C+Alex+Smola%2C+and+Eduard+Hovy%2E+2016%2E+Hierarchical+attention+networks+for+document+classification%2E+In+NAACL%2E) [our card](https://md.research.elamur.ai/free/en/papers/yang-2016-hierarchical-attention-networks-document-classification/)