From legal texts to legal ontologies and question-answering systems - - PowerPoint PPT Presentation

▶
from legal texts to legal ontologies and question
SMART_READER_LITE
LIVE PREVIEW

From legal texts to legal ontologies and question-answering systems - - PowerPoint PPT Presentation

Sep 14, 2023 324 likes •855 views

From legal texts to legal ontologies and question-answering systems Paulo Quaresma pq@di.uevora.pt Spoken Language Systems Lab / Dept. of Informatics INESC-ID, Lisbon / University of vora Portugal 1 From legal texts to

slide-1
SLIDE 1

1

From legal texts to legal ontologies and question-answering systems

From legal texts to legal ontologies and question-answering systems

Paulo Quaresma pq@di.uevora.pt

Spoken Language Systems Lab / Dept. of Informatics INESC-ID, Lisbon / University of Évora Portugal

slide-2
SLIDE 2

2

From legal texts to legal ontologies and question-answering systems

Some contextualization... :-)

Paulo Quaresma pq@di.uevora.pt

Spoken Language Systems Lab / Dept. of Informatics INESC-ID, Lisbon / University of Évora Portugal

slide-3
SLIDE 3

3

From legal texts to legal ontologies and question-answering systems

Some contextualization... :-)

slide-4
SLIDE 4

4

From legal texts to legal ontologies and question-answering systems

Some contextualization... :-)

slide-5
SLIDE 5

5

From legal texts to legal ontologies and question-answering systems

Some contextualization... :-)

slide-6
SLIDE 6

6

From legal texts to legal ontologies and question-answering systems

Some contextualization... :-)

slide-7
SLIDE 7

7

From legal texts to legal ontologies and question-answering systems

Overview

– Motivation – Challenges and objectives – General architecture – Examples – Conclusions and future work

slide-8
SLIDE 8

8

From legal texts to legal ontologies and question-answering systems

Motivation

– Access to laws or Court decisions

  • FIRE track “Adhoc retrieval from legal documents”

(adapted)

– U: I'm searching for situations where someone marries to an already married person. – S: In case 123 Mary married Peter, who was already married to Susan. – OR – S: Section 5 of the Hindu Marriage Act states: “...”

slide-9
SLIDE 9

9

From legal texts to legal ontologies and question-answering systems

Motivation

– Access to laws or Court decisions

  • FIRE track “Adhoc retrieval from legal documents”

(adapted)

– U: I'm searching for situations where someone divorced because the husband/wife left home. – S: In case 999 Ann divorced John after he returned to UK and sent no news for more than 7 years. – OR – S: Section 13 of the Hindu Marriage Act states: “...”

slide-10
SLIDE 10

10

From legal texts to legal ontologies and question-answering systems

Motivation

– Access to laws or Court decisions

  • FIRE track “Adhoc retrieval from legal documents”

(adapted)

– U: I'm searching for situations where someone marries to an already married person. – – Google-like queries? » “married to already married” » “bigamy” – Not likely to obtain the relevant laws and cases!

slide-11
SLIDE 11

11

From legal texts to legal ontologies and question-answering systems

Motivation

– Access to laws or Court decisions

  • FIRE track “Adhoc retrieval from legal documents”

(adapted)

– U: I'm searching for situations where someone divorced because the husband/wife left home. – – Google-like queries? » “divorce” “leave”? “abandon”? – Not likely to obtain the relevant laws and cases!

slide-12
SLIDE 12

12

From legal texts to legal ontologies and question-answering systems

Motivation

– Access to laws or Court decisions

  • FIRE track “Adhoc retrieval from legal documents”

(adapted)

– U: I'm searching for situations where someone divorced because the husband/wife left home. – – Previous classification of laws and cases accordingly with a taxonomy of legal concepts? » Requires user knowledge of the taxonomy – Not likely to be adequate to all users/citizens!

slide-13
SLIDE 13

13

From legal texts to legal ontologies and question-answering systems

Motivation

  • Information retrieval systems based in

– Free text searches – Catalogues / thesaurus

  • Are limited and inadequate for complex queries/situations!
slide-14
SLIDE 14

14

From legal texts to legal ontologies and question-answering systems

Motivation

  • “Google-like” systems:

– Unable to fully represent user intentions and document semantic content.

slide-15
SLIDE 15

15

From legal texts to legal ontologies and question-answering systems

Motivation

  • Catalogue-based systems:

– Structured knowledge (thesaurus) – Implies that the user needs to “know” the knowledge representation structure → “unfriendly” interface – Unable to represent user intentions and document semantic content

slide-16
SLIDE 16

16

From legal texts to legal ontologies and question-answering systems

Motivation

  • (We need) Content-based systems:

– Ontologies representing concepts and relations – Instances of ontologies representing document semantic content

  • Syntactic and semantic document analysis

– Queries

  • Syntactic and semantic interpretation
  • Answer as an inference process
slide-17
SLIDE 17

17

From legal texts to legal ontologies and question-answering systems

Motivation

  • Example revisited:

– U: I'm searching for situations where someone marries to an already married person. – ...

slide-18
SLIDE 18

18

From legal texts to legal ontologies and question-answering systems

Motivation

  • Example revisited:

– Ontology

  • Concepts:

– “person” “marriage”

slide-19
SLIDE 19

19

From legal texts to legal ontologies and question-answering systems

Motivation

  • Example revisited:

– “someone marries...”

  • An event, which is an instance of a “marriage”

– “... already married person”

  • Another “marriage” event, which occurred before

the previous one

slide-20
SLIDE 20

20

From legal texts to legal ontologies and question-answering systems

Motivation

  • Example revisited:

– event(E), time(E, TE), action(E, marriage(X, Y)) – event(P), time(P, TP), action(E, marriage(Y, Z)), X != Z, TP < TE.

slide-21
SLIDE 21

21

From legal texts to legal ontologies and question-answering systems

Challenges and objectives

  • To be able:

– To automatically analyse texts and to extract the conveyed information – To represent the extracted information in adequate knowledge representation structures – To support inferences over the knowledge

slide-22
SLIDE 22

22

From legal texts to legal ontologies and question-answering systems

Challenges and objectives

  • To be able:

– To automatically analyse texts and to extract the conveyed information

  • Natural language processing techniques

– Lexical, syntactical, semantic, and pragmatic interpretation of texts – Symbolic-based and/or statistical-based approaches

slide-23
SLIDE 23

23

From legal texts to legal ontologies and question-answering systems

Challenges and objectives

  • To be able:

– To represent the extracted information in adequate knowledge representation structures

  • Ontologies

– Creation – Mapping – Merge – Population

slide-24
SLIDE 24

24

From legal texts to legal ontologies and question-answering systems

Challenges and objectives

  • To be able:

– To support inferences over the knowledge

  • Inference-engine over knowledge represented by
  • ntologies
slide-25
SLIDE 25

25

From legal texts to legal ontologies and question-answering systems

Architecture (UEvora approach)

  • Two main modules

– Information extraction and representation

  • Extracts information from documents and creates a

knowledge base; – Information retrieval

  • Processes queries, accesses the knowledge base and

generates answers.

slide-26
SLIDE 26

26

From legal texts to legal ontologies and question-answering systems

Information extraction

slide-27
SLIDE 27

27

From legal texts to legal ontologies and question-answering systems

Information Extraction

  • Syntactical analysis

– PALAVRAS [Eckhard Bick] (Portuguese) – C&C Parser [Clark and Curran] (English)

  • (Partial) semantical analysis

– DRS: entities and conditions [Kamp]

  • Boxer [Bos]
  • Semantical/pragmatical analysis

– Ontology+DRS -> new DRS -> KB

slide-28
SLIDE 28

28

From legal texts to legal ontologies and question-answering systems

Information retrieval

slide-29
SLIDE 29

29

From legal texts to legal ontologies and question-answering systems

Information Retrieval

  • Query syntactic analysis
  • Query semantic analysis
  • Semantic-pragmatic query analysis
  • Answer inference

– Knowledge base inference

  • Logic programming framework

– PROLOG + ISCO [Abreu01]

  • OWL inference engines: Pellet, F-OWL, Euler,

JENA, JESS, ...

slide-30
SLIDE 30

30

From legal texts to legal ontologies and question-answering systems

Example

  • Syntactical analysis (C&C parser)

– Veronica married Peter who was married to Susan.

slide-31
SLIDE 31

31

From legal texts to legal ontologies and question-answering systems

Example

  • Semantical analysis (Boxer)
slide-32
SLIDE 32

32

From legal texts to legal ontologies and question-answering systems

Example

  • Semantic-pragmatic analysis
slide-33
SLIDE 33

33

From legal texts to legal ontologies and question-answering systems

Example

  • Knowledge base - Facts

name(xo, veronica). name(x1, peter). name(x2, susan). event(x3). event(x4). rel('<', x3, x4). marry(x4). agent(x4, x0), patient(x4, x1). marry(x3). agent(x3, x1). patient(x3, x2).

slide-34
SLIDE 34

34

From legal texts to legal ontologies and question-answering systems

Example

  • Knowledge base - Rules

marry(E, X, Y) ← marry(E), agent(E, X), patient(E, Y). marry(E, X, Y) ← marry(E), agent(E, Y), patient(E, X). …

slide-35
SLIDE 35

35

From legal texts to legal ontologies and question-answering systems

Question-Answering

  • Situations where someone married to an already married

person? – “Who married to a married person?”

  • After query analysis:

– P1 : – event(E1). event(E2). – rel('<', E2, E1). – marry(E1, P1, P2). – marry(E2, P2, P3).

slide-36
SLIDE 36

36

From legal texts to legal ontologies and question-answering systems

Question-Answering

  • Inference engine (logic programming): gnuProlog

– ?- name(P, N), event(E1). event(E2), rel('<', E2, E1), marry(E1, P, P2), marry(E2, P2, P3). – P = x0, N = veronica

slide-37
SLIDE 37

37

From legal texts to legal ontologies and question-answering systems

Evaluation - CLEF

  • 200 queries – CLEF05 (newspaper domain)

– 25% correct and well-supported answers – 52% without answer! – 1.5% correct answers but not well-supported – 11% not correct answers– too many (or too few) words – 10.5% incorrect answers

slide-38
SLIDE 38

38

From legal texts to legal ontologies and question-answering systems

Evaluation - CLEF

  • Main problem

– 52% questions without answer!

  • Cause?

– Mainly because the ontology was not “good enough” – it didn't represent the referred concepts and their relations; it didn't allow to represent correctly the document semantic content!

slide-39
SLIDE 39

39

From legal texts to legal ontologies and question-answering systems

Ontologies

  • The proposed approach requires:

– Ontologies to represent concepts and

relations

– The inference of instances → ontology

population

slide-40
SLIDE 40

40

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Problems:

– How to choose/create the ontology?

  • Use existent ontologies
  • Automatic merge of existent ontologies
  • Automatic creation of domain specific
  • ntologies from the analysis of the

documents

slide-41
SLIDE 41

41

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Ontology mapping and mapping:

– Cássia Santos PhD [Santos2009]

  • Multi-agent system based on

argumentation and/or negotiation

– Lexical agent – Structural agent – Semantical agent

slide-42
SLIDE 42

42

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Ontologies mapping [Santos2009]

Matcher A Matcher A Matcher B Matcher B Matcher N Matcher N

Composite Framework Composite Framework

Consensus Consensus

...

slide-43
SLIDE 43

43

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Ontologies mapping
  • OAEI: Ontology Alignment Evaluation Initiative
  • Dataset: “real” ontologies
  • BibTex MIT, BibTex UMBC, BibTex Karlsruhe,

INRIA

  • Matchers: OAEI
  • ASMOV, DSSim, Falcon, Lily, Ola, OntoDNA, PriorPlus,

RiMON, Sambo, SEMA, TaxoMap, XSOM, ...

slide-44
SLIDE 44

44

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Automatic creation of domain specific
  • ntologies – [Saias2010]

– Lexical, syntactical and semantical analysis

  • f documents

– Identification of:

  • NER – Named Entities Recognition
  • Actions
  • Semantic Role Labeling

– Relations – triples subject-verb-object

slide-45
SLIDE 45

45

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Automatic creation of domain specific
  • ntologies – [Saias2010]

– Use of:

  • Top-level/upper ontologies (SUMO,

DOLCE, ...)

  • Wordnet
  • Wikipedia

– To map new concepts and to extend the

existent ontology

slide-46
SLIDE 46

46

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Automatic creation of domain specific
  • ntologies – [Saias2010]

– Example:

  • Top-level ontology has the concept

“animal”

  • Document refers “cat”
  • Wordnet states that a cat is a mammal

and mammals are animals

  • So, the to-level ontology can be

extended to have the two new concepts!

slide-47
SLIDE 47

47

From legal texts to legal ontologies and question-answering systems

Ontologies

  • How to populate ontologies?

– DRS represent document content

  • Entities and conditions can be used to

populate ontologies

– person(a), name(a, 'Veronica') – Class “person”; “name” is a property

  • f class “pessoa”

– “a” is an instance of class person

slide-48
SLIDE 48

48

From legal texts to legal ontologies and question-answering systems

Ontologies

  • Another possible approach:

– How to populate ontologies?

  • Machine learning approach

– Information Extraction techniques

  • Support-Vector Machines
  • Conditional Random Fields

– Requires a previously tagged set of

documents (learning set), which should be representative of the document collection

slide-49
SLIDE 49

49

From legal texts to legal ontologies and question-answering systems

Evaluation - CLEF

  • 200 queries – CLEF08 (newspaper domain)

– 25% → 46.5% correct and well-supported answers – 52% → 32% without answer! – 1.5% → 1% correct answers but not well-supported – 11% → 5.5% not correct answers– too many (or too few) words – 10.5% → 15% incorrect answers

slide-50
SLIDE 50

50

From legal texts to legal ontologies and question-answering systems

Conclusions

  • Proposed an ontology/knowledge-based (legal) question-

answering system – Natural language processing – Knowledge bases / Ontologies – Logic Programming framework

slide-51
SLIDE 51

51

From legal texts to legal ontologies and question-answering systems

Future Work

  • Improve “all” modules
  • Improve ontologies!

– Use of adequate top-level ontologies – Ontology merge – Extract more instances – ontology population

  • Symbolic + machine learning aproaches
  • Apply the system to the legal domain and evaluate the

results.