Using Corpus Linguistics in Legal Research: Lessons from the Law and - - PowerPoint PPT Presentation

Using Corpus Linguistics in Legal Research: Lessons from the Law and Language at the European Court of Justice Project

Karen McAuliffe University of Birmingham k.mcauliffe@bham.ac.uk @dr_KMcA

Goldfarb, Neal, ‘Corpus LinguisFcs in Legal InterpretaFon: When Is It (In)appropriate?’ (Paper presented at the Law & Corpus LinguisFcs Conference, BYU Law School, 5 February 2019). Available at SSRN: hZps://ssrn.com/abstract=3333512 Solan, Lawrence, ‘Legal InterpretaFon by Big Data: Promises and Perils’ (Keynote Lecture at the JurilinguisFca II Conference, Seville, 24 October 2018).

Outline

• 1. The LLECJ Project: overview/methodology/

language at the CJEU

• 2. CL Study I: Pitfalls and Lessons
• 3. CL Study II: Learning from Lessons
• 4. Takeaways: a note of caution

Law and Language at the CJEU

Producing a multilingual jurisprudence Development of ‘precedent’ in ECJ judgments The changing role of the AG

Methodology

RQs

Case law analysis ObservaFonal data Interview data Corpus linguisFcs analysis SystemaFc literature reviews

www.llecj.karenmcauliffe.com

28 Member States 24 Official Languages

28 Member States 24 Official Languages

Flowchart of procedure

Case brought before CJEU Allocated to judge rapporteur (and AG where relevant) Documents translated into French Report of the judge rapporteur prepared by référendaire (in French) Where relevant, AG and référendaires prepare opinion (in pivot languages) First version of judgment drahed by référendaire (in French) Secret deliberaFons (in French) Final judgment drahed (in French) Judgment translated into language of the case (authenFc version of judgment and version signed by judges) and all other official languages

The Layers of a Judgment

The Layers of a Judgment

TRANSLATION PROCESS

TranslaFon at the ECJ

BG ES CS DA DE ET EL EN FR GA HR IT LV LT HU MT NL PL PT RO SK SL FI SV BG ES CS DA DE ET EL EN FR GA HR IT LV LT HU MT NL PL PT RO SK SL FI SV

JUDGMENT WORKING LANGUAGE: FRENCH

TranslaFon at the ECJ

BG ES CS DA DE ET

EL

EN FR GA HR IT LV LT HU MT NL PL PT RO SK SL FI SV BG ES CS DA DE ET

EL

EN FR GA HR IT LV LT HU MT NL PL PT RO SK SL FI SV

JUDGMENT WORKING LANGUAGE: FRENCH

Pivot TranslaFon

PIVOT LANGUAGES

• EN
• DE
• IT
• ES

Legal Language

• All legal language is formulaic

(MaFlla; Hiltunen; Gom; Tiersma)

• Gap in the literature

What is formulaic language?

Formulaic expressions are fixed or semi-fixed mulF-word sequences that occur frequently in a corpus.

Corpus linguisFcs approach to language analysis “The principle of idiom is that a language user has available to him or her a large number of semi-preconstructed phrases that consFtute single choices, even though they might appear to be analysable into segments” (Sinclair 1991: 110).

RepeFFveness in CJEU/REF judgments

1835 4939 1558 751 358 621 2085 1114 861 992 705

1000 2000 3000 4000 5000 6000

CJEU_fr REF_fra REF_bg CJEU_de REF_deu REF_at CJEU_en REF_uk REF_ir CJEU_it REF_it

Features of hybridity

• i)

ferFlity, newness, creaFvity;

• ii) strange and unexpected expressions that depart

from the standard use of language;

• iii) mixing of enFFes of different origin;
• iv) simplificaFon of language;
• v) a languages-in-contact situaFon;

<compeFFve NP>

• Shared: posiFon, advantage, acFvity, condiFon,

disadvantage, pressure, market, price, behaviour, process, pricing, pracFce.

• CJEU only: constraint, structure, level, interacFon,

force, balance, service, strength, neutrality, significance, acFon, conduct, purpose, factor, threat, context, effort, impact, mechanism, nature,

• pportunity
• REF only: tender, tariff, criterion, point, posiFoning,

business, evaluaFon, evoluFon, liFgaFon, offering, planning, swimmer, system, target, world.

<competent to>

• CJEU only: adopt, apply, impose, provide, rule,

declare, lay, conclude, reopen, enforce, establish, handle, proceed, raise, re-examine, use.

• REF only: issue, appeal, examine, make, try,

execute, perform, prove, bring, carry, act, conduct, exercise, have, dispose, exclude, form, invesFgate, order, reduce, speak, submit, understand, undertake.

<concentraFon>

• CJEU only: ‘concentraFon compaFble with’,

‘compaFbility of the concentraFon’, ‘implementaFon

• f the concentraFon’, ‘result of the concentraFon’,

‘compaFbility of a concentraFon’, ‘effect of the concentraFon’, ‘fact that the concentraFon’, ‘link between the concentraFon’.

• REF only: ‘concentraFon of alcohol’, ‘concentraFon of

107 milligrammes’, ‘concentraFon on the word’, ‘concentraFon on linguisFc features’, ‘concentraFon

• n an issue’.

Hybrid Language Shaping Concepts?

Language shapes how those drahing the judgments talk and think about EU law The judgments expressed in that language shape the development of EU law. New concepts are developed and expressed in that hybrid/new language The language is used/repeated by those drahing the judgments.

RepeFFveness in CJEU/REF judgments

1835 4939 1558 751 358 621 2085 1114 861 992 705

1000 2000 3000 4000 5000 6000

CJEU_fr REF_fra REF_bg CJEU_de REF_deu REF_at CJEU_en REF_uk REF_ir CJEU_it REF_it

Features of hybridity

• i)

ferFlity, newness, creaFvity;

• ii) strange and unexpected expressions that depart

from the standard use of language;

• iii) mixing of enFFes of different origin;
• iv) simplificaFon of language;
• v) a languages-in-contact situaFon;

<compeFFve NP>

• Shared: posiFon, advantage, acFvity, condiFon,

disadvantage, pressure, market, price, behaviour, process, pricing, pracFce.

• CJEU only: constraint, structure, level, interacFon,

force, balance, service, strength, neutrality, significance, acFon, conduct, purpose, factor, threat, context, effort, impact, mechanism, nature,

• pportunity
• REF only: tender, tariff, criterion, point, posiFoning,

business, evaluaFon, evoluFon, liFgaFon, offering, planning, swimmer, system, target, world.

CL Data: Misleading?

• RepeFFve language – but what is being

repeated? Does it really have any effect on the case law?

• Hybrid language – can corpus linguis,cs

actually speak to the impact of hybrid language on the case law?

• Using different language from NaFonal SCs –

is this not to be expected?

Lessons Learned

• PI must be aware of conflicFng research

agendas within the team (speaking to different audiences)

• Research design can be very difficult across

disciplinary languages (more layers of transla,on!)

Processing a case through the CJEU

Case brought before CJEU Allocated to judge rapporteur (and AG where relevant) Documents translated into French Report of the judge rapporteur prepared by référendaire (in French) Where relevant, AG and référendaires prepare opinion (in pivot languages) First version of judgment drahed by référendaire (in French) Secret deliberaFons (in French) Final judgment drahed (in French) Judgment translated into language of the case (authenFc version of judgment and version signed by judges) and all other official languages

AG Opinions

• Non-binding but Court must ‘take account’ of opinion

before delivering judgments (in relevant cases)

• Not Fed to maZers raised in a parFcular case
• Historically wriZen in the style of an academic paper
• The Court’s ‘sparring partner’ – creaFng a dialogue in

which principles of EU law are developed

• The most crea,ve point in the system of producFon of

the ECJ’s jurisprudence

CJEU Judgments

• WriZen in a new hybrid legal language
• Built up like Lego building blocks – formulaic style, mulF-

layered authorship

McAuliffe, K (2011) “Hybrid Texts and Uniform Law? The producFon of a mulFlingual jurisprudence by the Court of JusFce of the European Union” Interna,onal Journal for the Semio,cs of Law 24(1), 97-115 McAuliffe, K (2013) “The LimitaFons of a MulFlingual Legal System” Interna,onal Journal for the Semio,cs of Law 26(4) 861-882

Since 2004 (Convention)

Opinions drahed in pivot languages

EN DE ES FR IT (PL)

BG ES CS DA DE ET EL EN FR GA HR IT LV LT HU MT NL PL PT RO SK SL FI SV

Pre-2004: AGs always drahing in their mother tongue Post-2004: AGs drahing in one of the pivot languages of the Court (=EN/someFmes FR)

What implicaCons might changes in the linguis7c aspect of the AG’s role have for the construcCon and consolidaCon of ECJ jurisprudence?

• 1. To what extent has language use had an impact on the

usefulness of opinions?

• 2. Are AGs’ opinions becoming more syntheFc in construcFon

and their arguments more constrained by language as a result of the fact that they no longer drah in their mother tongue?

• 3. Do the opinions of permanent AGs differ from those drahing

in pivot languages?

LinguisFc Analysis

• 1. To what extent has language use had an impact on the

usefulness of opinions?

• 2. Are AGs’ opinions becoming more syntheFc in construcFon

and their arguments more constrained by language as a result of the fact that they no longer drah in their mother tongue?

• 3. Do the opinions of permanent AGs differ from those drahing

in pivot languages?

• 1. Judgments are stylisCcally simpler and less ‘fluent’

than opinions

• 2. Aher the 2004 language regime change opinions

become more similar (linguisFcally) to judgments

• 3. Opinions drahed aher 2004 by non-permanent AGs

(i.e. AGs drahing in a language other than their mother tongue) are stylisFcally simpler and less ‘fluent’ than the ones drahed by AGs (drahing in their mother tongues) prior to 2004.

Main objecFves

• 1. Determining the stylisFc complexity and the

fluency of the judgments and the opinions before and aher the linguisFc reform

• 2. Comparing the results

between judgments/opinions between the opinions drahed before/ aher the 2004 linguisFc reform

Research quesFons

• Which linguisFc features do represent the fluency

and the stylisFc simplicity/complexity of a text?

• How can such features be idenFfied in the

considered opinions and judgments?

• Do the judgments present less features

represenFng stylisFc complexity and fluency than the opinions?

• Do the opinions become more similar to the

judgments aher the 2004 linguisFc reform?

• Do the opinions drahed before 2004 present

more features represenFng stylisFc complexity and fluency than the ones drahed aher 2004?

DeterminaFon

• f stylisFc

complexity and fluency Comparison

• f the results

Analysed archive of corpora

EN FR Judgments……………………………………………………………………………………………….. Opinions

drahed before 2004……….……………………………….…. drahed aher 2004 from naFve AGs……….... drahed aher 2004 from non-naFve AGs…..

Judgments ………………………………………………………........................................

….

Opinions

drahed before 2004…………………………………………… drahed aher 2004 from naFve AGs…………. drahed aher 2004 from naFve AGs………....

(2208 texts, > 10 million words) (341 texts, 2 million words) (276 texts, > 2,5 million words) (367 texts, > 3 million words) (2357 texts, > 10 million words) (410 texts, > 2,5 million words) (339 texts, > 3,5 million words) (345 texts, > 3,5 million words)

Methodology

• 1. DeterminaFon of the features related to

fluency and stylisFc simplicity/complexity

• 2. Search of each one of the determined

features in each corpus

• 3. Comparison of the obtained between the

four sets of texts analyzed for each language

• 4. Extra exams suggested by the obtained

results

Results

Judgments are stylisFcally simpler and less fluent than the opinions

• Lexical variety à More repeFFons
• Lexical density
• Sentence length
• Lower presence of hypotacFc structures

In English In French

Results

Opinions drahed aher 2004 by non naFve authors are stylisFcally simpler and less fluent than the ones drahed before 2004 in a naFve language

• Lexical variety
• Lexical density
• Sentence length
• Lower presence of hypotacFc structures

Opinions become more and more similar (linguisFcally) to judgments

Results

LEXICAL VARIETY CreaFon of a WORD LIST 𝑈𝑈𝑆=​𝑢𝑧𝑞𝑓𝑡/ 𝑢𝑝𝑙𝑓𝑜𝑡 ×100

0.468 0.468 1.146 0.86 0.849 0.2 0.4 0.6 0.8 1 1.2 1.4 before 2004 aher 2004

English corpus lexical variety

Judgments NaFve opinions Non NaFve opinions 0.43 0.43 1.103 0.893 0.788 0.2 0.4 0.6 0.8 1 1.2 before 2004 aher 2004

French corpus lexical variety

Judgments NaFve opinions Non NaFve opinions

Results

Sentence length

52.01 55.07 33.63 45.01 36.22 10 20 30 40 50 60 aher 2004 before 2004

Mean sentence length of English corpus

Judgments NaFve opinions Non naFve opinions 57.32 64.13 33.92 46.24 36.09 10 20 30 40 50 60 70 aher 2004 before 2004

Mean sentence length of French corpus

Judgments NaFve opinions Non naFve opinions

Results

HYPOTACTIC STRUCTURES

• Choice of an exhausFve list of

subordinate conjuncFons

(e.g. when, than, because, etc.)

• IdenFficaFon of each conjuncFon within

the wordlist and of its frequency

• Sum of the frequency of each

subordinate conjuncFon included in the considered list

2882 + 2767 + ……. = In the French corpus:

Judgments vs. StaCsCcal significance Minimal LL value = 6,63, the greater the LL value, the greater the difference Opinions drahed before 2004 1192,59 NaFve opinions drahed aher 2004 134,28 Non-naFve opinions drahed aher 2004 0,19

Results

HYPOTACTIC STRUCTURES In the English corpus: Opinions drahed before 2004 NaFve opinions drahed aher 2004 Non naFve opinions drahed aher 2004 vs. 965.49 978.25

Results

Lexical density: no fix paZern

0.86 0.88 0.9 0.92 0.94 0.96 0.98 1 1.02 1.04 1.06 English Judgments Opinions before 2004 NaFve opinions aher 2004 Non naFve opinions aher 2004 0.88 0.9 0.92 0.94 0.96 0.98 1 1.02 French Judgments Opinions before 2004 NaFve opinions aher 2004 Non naFve opinions aher 2004

Resuming….

Opinions become gradually more similar to judgments (i.e. stylisFcally simpler and less fluent) as for:

• A decreasing lexical variety
• An increasing quanFty of repeFFons
• An increasing sentence length
• A decreasing quanFty of hypotacFc

structures

Extra exam

IdenFficaFon of idenFcal fragments between judgments and opinions

English corpora 100% matching fragments Opinions Before 2004 Aher 2004 NATIVE Aher 2004 NON NATIVE All matching fragments 10,832 100% 4,145 100% 29,218 100% ≥ 10 words 331 3% 1,972 23.17% 2,028 27.81% ≥ 30 words 85 0,7% 1,000 11.75% 897 12.30% ≥ 100 words 5 0.04% 45 0.52% 41 0.56% French corpora 100% matching fragments Opinions Before 2004 Aher 2004 NATIVE Aher 2004 NON NATIVE All matching fragments 5,407 100% 7,155 100% 6,139 100% ≥ 10 words 389 7.19% 1,438 20.09% 1,269 20.08% ≥ 30 words 115 2.12% 559 7.81% 421 6.66% ≥ 100 words 8 0.14% 24 0.33% 21 0.33%

Jumping to Conclusions?

• Results show judgments to be stylisCcally

simpler and less ‘fluent’ than Advocates Generals’ opinions

Jumping to Conclusions?

• Results show Judgments to be stylisCcally simpler and

less fluent than Advocates Generals’ opinions

• The introducCon of the linguisCc reform in 2004 did have

an influence on the style and the fluency of the opinions Opinions become more similar to judgments Opinions become stylisFcally simpler and less ‘fluent’ than previously So what?

Jumping to Conclusions?

• Results show judgments to be stylisCcally simpler and

less fluent than Advocates Generals’ opinions

• The introducCon of the linguisCc reform in 2004 did have

an influence on the style and the fluency of the opinions Opinions become more similar to judgments Opinions became stylisFcally simpler and less fluent than previously Opinions are the same as judgments so lose their value

Mixed-Method Empirical Research

Literature Reviews: Value of AGs’ opinions is in persuasiveness and language and writing style is a fundamental part

• f the ability to persuade

Interview & Observational Data: Opinions are multi-layered/multi-authored texts Tensions between law, language, and translation 2004 language changes – ‘no impact’ on language and style of opinions

Tempered Conclusions…

• Results show judgments to be stylisCcally simpler and

less fluent than Advocates Generals’ opinions

• The introducCon of the linguisCc reform in 2004 did have

an influence on the style and the fluency of the opinions Opinions become more similar to judgments Opinions became stylisFcally simpler and less fluent than previously Opinions lose their original eloquence Do they then begin to lose their value?

What is the purpose of AGs’ Opinions?

• To compensate for lack of individual judgments/

dissenting judgments?

• To persuade the ECJ in a particular case?
• To force the ECJ to (re)evaluate principles of EU law?
• To provide context for wider EU scholarship?

What might all this mean for the development of EU Law?

Takeaways

• 1. Research design is important
• 2. You need legal expertise in your research team
• 3. Resist the temptation to publish at the stage of

achieving results that are interesting from a CL point of view alone – take a deeper dive

• 4. CL can be very valuable in legal research, but
• nly as part of a wider methodology