What can technology do for the legal profession?...Faculty of Informatics Technical University of Munich The Smart Workspace What can technology do for the legal profession? Bernhard

Faculty of InformaticsTechnical University of Munich

The Smart WorkspaceWhat can technology do for the legal profession?Bernhard Waltl, 2018

Introduction

• Processes of legal experts (scientists and lawyers) are…• ... time-intensive• ... knowledge-intensive• ... data-intensive.

• Legal Data Science is becoming more and more attractive, because• ... process time and memory space are cheap• ... algorithms can process data fast and accurate.

• In order to achieve highest accuracy, • algorithms (e.g., importer, segmenter, named entity recognition),• models and patterns (e.g., machine learning models, linguistic models),• training and test data sets,

• have to be adapted.

2180322 Waltl - The Smart Workspace

MotivationGartner Hype Cycle July 2017

180322 Waltl - The Smart Workspace 3

Deep Learning

Artificial General Intelligence

Smart Workspaces

Machine Learning

CognitiveComputing

The „Smart“ Workspace


The „Smart“ Workspace

1. Support of search and exploration processes§ eDiscovery, forensics, etc.§ E.g., Due diligence, technology assisted review (TAR), etc.

2. Support during creation of contracts§ Document assembling§ Consistency checks and smart recommendations

3. Structured computational contracts§ Machine-readable representation of contracts§ Clause dependent , e.g., hybrid contracts


Use Cases

Huge researchpotential

Structuring a legal contract is a journey

Recognition of names entities

POS - tagging

Stop word removal

Segmentation

Tokenization

Unstructured Information

Structured Information

Extraction andcrawling of text

Stream of tokens

Different document types

Different document formats

Normalization of names entities

Stemming

Lemmatization

Dependency parsing

Constituencyparsing

Extraction of auxiliary sentences

Recognition of subjects, objects, etc.

Semantic role labeling

Extraction of relations

Determination of functional types

Creation of controlled natural language

Normalization of text

Formal representationof concepts

Creation and reuse of ontologies

Lexical Phase

Morphological Phase

FormalizationPhase

Syntactic Phase

Semantic Phase

Raw TextPhase


Waltl, Bernhard, Semantic Analysis and Computational Modeling of Legal Documents, Dissertation, 2018 (to appear)

Technology is language agnosticEnglish works even better


Classification of Norms


Reference ArchitectureOpen-Source Software Stack


Implementation Details• Web Application• ElasticSearch• Apache UIMA • Apache Spark

Complex pattern recognizer(Apache Ruta)

Active machinelearning

componentPattern

definitions

Exporter

Dependencyparser

POS tagger

Metric calc. component

Named entityrecognizer

Splitter Stemmer

Lemmatizer Subject tagger

Tokenizer

User interface

Exploration

Text analysis engine

Navigation

Thesauri

Importer

Database and search-engine

Knowledgebases

Data access layer

Information extraction component

Classification of NormsRule-based approach

UIMA (Unstructured Information Management Architecture)• Pipes & Filters architecture

• Thread-safe

• Apache Ruta: complex pattern specification


Splitter Segmenter TokenizerReferenceExtractorDocument Sections Sentences Tokens

(Words)

Sentence Classifier

• Structured Text • Named Entities• Classified Sentences

Money Value Extractor

Date & TimeExtractor

Named Entity Extractor

Classification of NormsActive Machine Learning: A Hybrid Approach

Lexia + LexML• Machine Learning as a Service (MLaaS)• Apache Spark + Apache UIMA


Rule-based

Classifier

(e.g., naive bayes, logistic regression, perceptron, etc.)

Labeled norms

Unlabeled norms

Knowledge-engineer,Domainexpert

Labeled norms

Query strategy

Query strategies(e.g., most/least

uncertain)

Labour intensive & tedious!

Interaction with thesystem

Parameters & Options

Classification of Sentences, Phrases, etc.


Data setGerman tenancy law


Features• Bag-of-Words• Stopword removal• Tf-idf vectorization

Classes canhave very low

support!

Results on classifications of norms


Remarkableresults?

Document creation

• A lot of effort needs to be done to structure documents1. Pre-processing2. Training3. Feature extraction4. Data mining5. Post-processing

• Why are they unstructured?• What if we could structure them as we create them?

è This would completely change the situation!


Hybrid contracts

Definition• A hybrid contract is a document, which contains contractual content, combining

unstructured, structured, and computable information in a machine-readable and executable format.

What does this mean?

• Unstructured information à Text, Images, etc.

• Structured information à Parties, Metadata, Expiration Dates, etc.

• Computable information à Decision structures, code, etc.

è It is technology feasible that a document contains various kind of information.


Closing remarks and follow-ups

• Technology is ready, but there is not a “one-size-fits-all” solutions

• The ”Smart Workspace” is more than just software• Data• Methods• Workflow Integration

• Follow-Up

• What is the ”Smart Workspace” for the audience?

• How does the “Smart Workspace” for lawyers look like?

• Which kind of workflows need to be supported?

• What kind of data is relevant in the United States?


Technische Universität MünchenFaculty of InformaticsChair of Software Engineering for Business Information Systems

Boltzmannstraße 385748 Garching bei München

Tel +49.89.289.Fax +49.89.289.17136

wwwmatthes.in.tum.de

Bernhard Waltl

17124

[email protected]

Research Associate

http://wwwmatthes.in.tum.de/