法學資訊服務

法學資訊服務

政治大學資訊科學系劉昭麟18 September 2003

資訊檢索研討會

Outline

Knowledge Representation FormalismsSome research in legal informatics

– Case categorization– Prior case retrieval– Legal document summarization– Legal document drafting– Computer-assisted sentencing– Computer-assisted argumentation

Research at NCCU– Criminal summary judgments

Knowledge Representation

Bench-Capon and Visser (1997, UK)– Rule-based systems– Case-based systems– Statistical approach– Ontology

RDF (Ebenhoch 2001, Germany)

Some Research in Legal Informatics

Case categorizationPrior case retrievalLegal document draftingLegal document summarizationComputer-assisted sentencingComputer-assisted argumentation

Case Categorization

Thompson (2001, USA)Targets: 40 high level categoriesMethods

– kNN-like Approach• artificial cases

• tfidf

– C4.5 Rules • pruned rule sets obtained from a C4.5 decision tree

– Ripper• learn rules from training cases

Prior Case Retrieval

Al-Kofahi et al. (2001, USA)Targets: case history Features:

– title similarity and weight– docket number– etc.

Support Vector Machine

Legal Document Summarization

Moens (1997,2001, Belgium)Targets: criminal casesContents: predictable and unrestrictedText grammar for predictable contentsShallow statistical techniques for unrestricted

contents– Index terms– K-medoid clustering methods

Legal Document Drafting

Branting (1998, USA)Targets: Show-cause orders for Colorado Court of

AppealsFeatures:

– Text grammars

– Illocutionary structure

– Rhetorical structure

– Document planner

– Document drafter

Computer-Assisted Sentencing

Schild (1995, 1998, Israel)Targets: robbery and rape

Advantage: uniformity of the sentencing system

Case-based system– Source of cases: interviewing judges

– Representation: Multiple Explanation Pattern (Schank 1994)

Interaction with users

Computer-Assisted Argumentation

Argumentation theoriesStranieri et al (2001, Australia)

– Domain classification: no/bounded/narrow/unfettered discretion

– Decision trees: legal procedures– Argument trees: canonical auguments

Argumentation

Verheij (1999, The Netherlands)– Dialectical arguments

• Reasons

• Conclusions

• Exceptions

• Warrants

– Views• Line-of-argument view

• Statements view

Education

Aleven (1997, USA)– Educate students for argumentation in the legal

domain– Case-based– Issue-based

Research at NCCU

A Case-based Reasoning Approach to Classifying Criminal Summary

Judgments in Chinese

Contributors

政治大學法學院陳起行教授台灣大學國發所陳顯武教授板橋地方法院何君豪法官政治大學資科所張正宗先生

Background Information

Documents for criminal summary judgments– Indictment document– Judgment document

Approaches– Rule-based classifiers– Case-based classifiers

Sources of decision criteria– Human-provided– Machine-generated

Data for Learning and Testing

案由代號 training test

公共危險罪 C1 158 271妨害風化罪 C2 26 44賭博罪 C3 30 87傷害罪 C4 14 33竊盜罪 C5 99 243侵占罪 C6 15 46贓物罪 C7 9 15違反動產擔保交易法 C8 15 40違反毒品危害防制條例 C9 19 73違反電子遊戲場業管理條例 C10 16 23違反兒童與少年性交易防制條例 C11 19 52違反台灣地區與大陸地區人民關係條例 C12 9 23其他案件 C13 74 146

Human-Provided Rules

人工挑選關鍵詞依照下列固定的格式將 rule 儲存起來

1. 案由或法條名稱2. 門檻值3. no4. 不欲出現的詞 -15. 不欲出現的詞 -2………6. 不欲出現的詞 -n7. event8. 關鍵詞 -19. 關鍵詞 -2………10. 關鍵詞 -m

刑法第二百六十八條2event提供供給賭博場所公眾得出入之場所賭場

Human-Provided Cases

1. 案由或法條名稱2. 門檻值3. no4. 不欲出現的詞 -15. 不欲出現的詞 -2………6. 不欲出現的詞 -n7. event8. 關鍵詞 -11 、關鍵詞 -12 、…、關鍵詞 -1x ：比重 -1 9. 關鍵詞 -21 、關鍵詞 -22 、…、關鍵詞 -2y ：比重 -2 ………10. 關鍵詞 -k1 、關鍵詞 -k2 …、、關鍵詞 -kz ：比重 -m

Case Instance Examples

傷害20event爭執、口角： 10基於傷害之故意、基於傷害人身體之故意： 20毆： 10挫傷、傷害、擦傷、裂傷、死亡： 10

爭執→毆→擦傷： 30擦傷→口角→毆： 20基於傷害之故意→毆→裂傷： 40

Learning Case Instances

Segmenting Chinese character strings– Use (somewhat) customized HowNet

– Prefer longest matches

Preprocessing– 依「，；。」這三個符號，將犯罪事實欄位內的資料切成許多小片段。

– 刪除描述時間與地址的小片段。– 判斷是否為時間或地址之描述的方法

• 「年、月、日、時、分」五個出現兩個以上為時間描述之小片段。

• 「市、縣、路、村、里、段、巷、弄、號」九個出現三個以上為地點描述之小片段。

An Sample Result of Preprocessing

在某 KTV 店內服用酒類，致其反應能力降低，已不能安全駕駛動力交通工具後，撞及自對向車道駛來，欲左轉站前路，由林○○所駕駛之 Z ○○３－○○號自用小客車，嗣經警方處理，對吳○○施以酒精測試，其測定值為０‧八五 MG ／ L，始循線查知上情。

吳○○於民國九十年十月二十七日上午十時十分許，在某 KTV 店內服用酒類，致其反應能力降低，已不能安全駕駛動力交通工具後，仍駕駛車號 HY ○○○○－號自用小客車沿板橋市文化路往台北方向行駛，在行經臺北縣板橋市文化路與站前路路口時，撞及自對向車道駛來，欲左轉站前路，由林○○所駕駛之 Z ○３－○○○號自用小客車，嗣經警方處理，對吳○○施以酒精測試，其測定值為０‧八五 MG ／ L ，始循線查知上情。

A Sample Learned Case Instance

第一行儲存的是案由或法條的名稱。

第二行開始，將前處理後的小片段，做斷詞處理，並刪掉長度為 1 的詞，剩下的詞依原出現順序，以一個空白為間隔，儲存起來。

公共危險內服反應能力降低不能安全駕駛動力交通工具車道左轉駕駛自用客車警方處理施以酒精測試測定

Case Instance Applications

把欲處理的起訴書中犯罪事實欄位的資料，做與建立 case instances同樣的處理，得到一個詞的串列 X 。

Instance中第二行起的資料 Y ，與起訴書所得到的資料 X ，其相似度計算方式：

)(

),(_

)(

),(_

2

1),(1 YCounts

YXCountsOCW

XCounts

YXCountsOCWYXs

Example for Case Applications

不能安全駕駛動力交通工具程度駕駛車號自用客車途經發覺指揮交通反映遲緩盤查酒精測試含量毫克

不能安全駕駛動力交通工具駕駛自用客車酒精測試

Instance Test data

Counts = 19Counts = 21

Counts = 9

OCW

s2 = (9/19 + 9/21)/2 = 0.4511


A Sample Learned Rule Instance

第一行儲存的是案由或法條的名稱。

第二行開始，將前處理後的小片段，做斷詞處理，並刪掉長度為 1 的詞，剩下的詞以一個空白為間隔，儲存起來。詞與詞之間，原本出現順序之特徵不必保留。


Rule Instance Applications

把欲處理的起訴書中犯罪事實欄位的資料，做與建立 rule instances同樣的處理，得到一個詞的串列 X 。

Instance中第二行起的資料 Y ，與起訴書所得到的資料 X ，其相似度計算方式：

)(

),(_

)(

),(_

2

1),(2 YCounts

YXCountsUCW

XCounts

YXCountsUCWYXs

Case Refinement Strategies

Merging similar casesRemoving irrelevant keywords

Merging Similar Cases

Procedure Merge2Instances(X, Y)if ( (Size(Com(X,Y)) ≧ *Size(X)) and (Size(Com(X,Y)) ≧ *Size(Y)) ){ Remove X and Y from the instance database; Add Com(X, Y) into the instance database;}else if ( (Size(Com(X,Y)) < *Size(X)) and (Size(Com(X,Y)) ≧ *Size(Y)) ) Remove Y form the instance database;else if ( (Size(Com(X,Y)) ≧ *Size(X)) and (Size(Com(X,Y)) < *Size(Y)) ) Remove X form the instance database;

Creating Prototypical Rules

m: index of keywordsn(i): number of case instances of Cik(m,i): number of occurrences of mth

keyword in cases of CiAOF(m,i)=k(m,i)/n(i)Remove all keywords not satisfying

Recovering some rules…5.1),(5.0 imAOF

Removing Similar Keywords

m: index of keywordsn(i): number of case instances of Cik(m,i): number of occurrences of mth

keyword in cases of CiAOF(m,i)=k(m,i)/n(i)Remove gth keyword if

– AOF(g,i) – The gth keyword appears in case instances of Cj,

ji

Weighted k Nearest Neighbors

Use WkNN for classificationPrinciples

– 在相似度分數大於門檻值 (0.3) 的 instances中，選取最多 instances投票的案由或法條

– 若有兩個以上的案由或法條得票數相同，選取總分最高者

Performance Evaluation

Standard precision and recallF measure with = 1 ( ) The selection of 全部案由或法條的正確率之計算

– AP 、 AR 、 AF 、 WP 、 WR 、 WF

Correct rate and rejection rate

RP

PRF

2

2

)(

13

1

13

1

13

113

1

jj

iii

ii

cpcWP

pAP

Experimental Design

Figure 3. Structure of our experiments

Keywords Cases

+Segment HumanHumanMachine

OneRule ManyRules

+Merge -Merge+Merge -Merge E1E2

E10E11E12

Design Factors

=0.7 to 0.2 step –0.1 E3

E4, E5, E6, E7, E8, E9

Experimental Results

EXPID C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13

E1-P 100 100 94 67 98 81 100 100 99 100 98 88 76

E2-P 100 97 90 100 99 25 100 100 98 50 100 83 52

E1-R 94 93 98 91 97 83 47 85 96 78 100 91 88

E2-R 83 77 84 48 59 96 53 90 86 100 100 87 68

E1-F 99 99 95 71 98 81 82 97 98 95 98 89 78

E2-F 96 92 89 82 87 29 85 98 95 56 100 84 55

Experimental Results (2)

EXPID SIZE AP AR AF WP WR WF CR RR

E1 24 92 88 89 94 93 93 98 4

E2 12 84 79 78 88 76 79 86 10

E3 428 84 79 79 87 87 86 92 3

E4 (0.7) 282 86 80 81 90 88 88 94 5

E5 (0.6) 214 84 80 81 89 88 88 95 6

E6 (0.5) 133 83 74 76 88 83 84 96 14

E7 (0.4) 91 84 63 68 86 77 78 95 21

E8 (0.3) 47 73 53 56 80 56 60 92 44

E9 (0.2) 18 59 45 43 73 42 42 79 56

E10 428 75 68 67 80 80 76 88 1

E11 337 63 66 62 75 79 74 86 1

E12 12 62 63 56 76 61 58 64 5



E1 24 92 88 89 94 93 93 98 4

E2 12 84 79 78 88 76 79 86 10

E3 428 84 79 79 87 87 86 92 3

E4 (0.7) 282 86 80 81 90 88 88 94 5

E5 (0.6) 214 84 80 81 89 88 88 95 6

E6 (0.5) 133 83 74 76 88 83 84 96 14

E7 (0.4) 91 84 63 68 86 77 78 95 21

E8 (0.3) 47 73 53 56 80 56 60 92 44

E9 (0.2) 18 59 45 43 73 42 42 79 56

E10 428 75 68 67 80 80 76 88 1

E11 337 63 66 62 75 79 74 86 1

E12 12 62 63 56 76 61 58 64 5



E1 24 92 88 89 94 93 93 98 4

E2 12 84 79 78 88 76 79 86 10

E3 428 84 79 79 87 87 86 92 3

E4 (0.7) 282 86 80 81 90 88 88 94 5

E5 (0.6) 214 84 80 81 89 88 88 95 6

E6 (0.5) 133 83 74 76 88 83 84 96 14

E7 (0.4) 91 84 63 68 86 77 78 95 21

E8 (0.3) 47 73 53 56 80 56 60 92 44

E9 (0.2) 18 59 45 43 73 42 42 79 56

E10 428 75 68 67 80 80 76 88 1

E11 337 63 66 62 75 79 74 86 1

E12 12 62 63 56 76 61 58 64 5



E1 24 92 88 89 94 93 93 98 4

E2 12 84 79 78 88 76 79 86 10

E3 428 84 79 79 87 87 86 92 3

E4 (0.7) 282 86 80 81 90 88 88 94 5

E5 (0.6) 214 84 80 81 89 88 88 95 6

E6 (0.5) 133 83 74 76 88 83 84 96 14

E7 (0.4) 91 84 63 68 86 77 78 95 21

E8 (0.3) 47 73 53 56 80 56 60 92 44

E9 (0.2) 18 59 45 43 73 42 42 79 56

E10 428 75 68 67 80 80 76 88 1

E11 337 63 66 62 75 79 74 86 1

E12 12 62 63 56 76 61 58 64 5



E1 24 92 88 89 94 93 93 98 4

E2 12 84 79 78 88 76 79 86 10

E3 428 84 79 79 87 87 86 92 3

E4 (0.7) 282 86 80 81 90 88 88 94 5

E5 (0.6) 214 84 80 81 89 88 88 95 6

E6 (0.5) 133 83 74 76 88 83 84 96 14

E7 (0.4) 91 84 63 68 86 77 78 95 21

E8 (0.3) 47 73 53 56 80 56 60 92 44

E9 (0.2) 18 59 45 43 73 42 42 79 56

E10 428 75 68 67 80 80 76 88 1

E11 337 63 66 62 75 79 74 86 1

E12 12 62 63 56 76 61 58 64 5



E1 24 92 88 89 94 93 93 98 4

E2 12 84 79 78 88 76 79 86 10

E3 428 84 79 79 87 87 86 92 3

E4 (0.7) 282 86 80 81 90 88 88 94 5

E5 (0.6) 214 84 80 81 89 88 88 95 6

E6 (0.5) 133 83 74 76 88 83 84 96 14

E7 (0.4) 91 84 63 68 86 77 78 95 21

E8 (0.3) 47 73 53 56 80 56 60 92 44

E9 (0.2) 18 59 45 43 73 42 42 79 56

E10 428 75 68 67 80 80 76 88 1

E11 337 63 66 62 75 79 74 86 1

E12 12 62 63 56 76 61 58 64 5


Merge->Remove (before)

35%

45%

55%

65%

75%

85%

95%

0.20.40.60.81

Merge->Remove(after)

35%

45%

55%

65%

75%

85%

95%

0.20.40.60.81

00.511.52

Remove->Merge

35%

45%

55%

65%

75%

85%

95%

0.20.40.60.81


Merge>Remove(before)

0%

15%

30%

45%

60%

75%

90%

0.20.40.60.81

Merge->Remore(after)

0%

15%

30%

45%

60%

75%

90%

0.20.40.60.81

0.0 0.5 1.0 1.5 2.0

Remove->Merge

0%

15%

30%

45%

60%

75%

90%

0.20.40.60.81


Merge->Remove (before)

35%

45%

55%

65%

75%

85%

95%

0.20.40.60.81

Merge->Remove(after)

35%

45%

55%

65%

75%

85%

95%

0.20.40.60.81

00.511.52

Remove->Merge

35%

45%

55%

65%

75%

85%

95%

0.20.40.60.81

Merge>Remove(before)

0%

15%

30%

45%

60%

75%

90%

0.20.40.60.81

Merge->Remore(after)

0%

15%

30%

45%

60%

75%

90%

0.20.40.60.81

0.0 0.5 1.0 1.5 2.0

Remove->Merge

0%

15%

30%

45%

60%

75%

90%

0.20.40.60.81

Some On-Line Resources

行政院法務部 http://www.moj.gov.tw/ 立法院 http://www.ly.gov.tw/ 司法院 http://wjirs.judicial.gov.tw/jirs/ 法源 http://www.lawbank.com.tw/ 植根 http://rootlaw.lifelaw.com.tw/

http://www.ordos.nm.cn/haoxia/navigation/zhengfa.htm

References

In the following references, I use AI for Artificial Intelligence, ICAIL for International Conference on Artificial Intelligence and Law, and DEXA for International Workshop on Database and Expert Systems Applications.

1. V. Aleven, Teaching Case-based Argumentation Through a Model and Examples, Ph.D. Dissertation, University of Pittsburgh, Pittsburgh, Ohio, USA, 1997.

2. K. Al-Kofahi, A. Tyrrell, A. Vachher, P. Jackson, A machine learning approach to prior case retrieval, Proc. of the 8th ICAIL, 88–93, 2001.

3. T. J. M. Bench-Capon, P. R. S. Visser, Open texture and ontologies in legal information systems, Proc. of the 8th DEXA, 192–197, 1997.

4. K. Branting, J. Lester, C. Callaway. Automating judicial document drafting: A discourse-based approach. AI & Law, 6(2-4), 111–149, 1998.

5. M. P. Ebenhoch, Legal knowledge representation using the resource description framework (RDF), Proc. of the 12th DEXA, 369–373, 2001.

6. C.-L. Liu and C.-T. Chang. Some case-refinement strategies for case-based criminal summary judgments, Proc. of the 14th Int’l Symposium on Methodologies for Intelligent Systems, to appear, October 2003.

7. C.-L. Liu, C.-T. Chang, J.-H. Ho, Classification and clustering for case-based criminal summary judgments, Proc. of the 9th ICAIL, 252–261, 2003.

8. M.-F. Moens, C. Uyttendaele, and J. Dumortier, Abstracting of legal cases: The SALOMON experience, Proc. of the 6th ICAIL, 114–122, 1997.

9. M.-F Moens, Innovative techniques for legal text retrieval, AI and Law, 9(1), 29–57, 2001.10. U. J. Schild, Intelligent computer systems for criminal sentencing, Proc. of the 5th ICAIL, 229–238, 1995. 11. U. J. Schild, Criminal sentencing and intelligent decision support, AI and Law, 6(2-4), 151–202,1998.12. A. Stranieri, J. Yearwood, and J. Zeleznikow, Tools for world wide web based legal decision support systems,

Proc. of the 8th ICAIL, 206–214, 2001.13. P. Thompson, Automatic categorization of case law, Proc. of the 8th ICAIL, 77–77, 2001.14. B. Verheij, Automated argument assistance for lawyers, Proc. of the 7th ICAIL, 43–52, 1999.