From Natural Language Processing to Desicion Support Systems

huntcopywriterΤεχνίτη Νοημοσύνη και Ρομποτική

24 Οκτ 2013 (πριν από 3 χρόνια και 10 μήνες)

106 εμφανίσεις

From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
From Natural Language Processing
to Desicion Support Systems
Natural Language Processing as a fundamental component
to improve communication in Decision Support Systems
Thomas LEBARB
´
E
Laboratoire LIDILEM EA609
Universit´e Stendhal - Grenoble 3
New Trends in Information Technology
Homs,Syria,18 avril 2006
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Outline
1
About the author
2
Introduction
3
Natural Language Processing
4
Integrating NLP in Decision Support Systems:examples
5
Conclusions
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Outline
1
About the author
2
Introduction
3
Natural Language Processing
4
Integrating NLP in Decision Support Systems:examples
5
Conclusions
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
About the author
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
About the author
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
About the author
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
About the author
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
About the author
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
About the author
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
About the author
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Outline
1
About the author
2
Introduction
3
Natural Language Processing
4
Integrating NLP in Decision Support Systems:examples
5
Conclusions
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Introduction
From NLP to DSS
Natural language processing ￿Computational Linguistics
Computational linguistics is to be regarded as the
processing of language matter by means of implicit or
explicit linguistic knowledge
Decision Support Systems
A decision support system (DSS) is a computer program
application that analyzes business data and presents it so
that users can make business decisions more easily

Business is “data-driven”
but data do not always sum up to numbers

Decision making is often (or should be) document based
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Outline
1
About the author
2
Introduction
3
Natural Language Processing
4
Integrating NLP in Decision Support Systems:examples
5
Conclusions
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
General principles

Language is ambiguous

it can not be processed like a programming language

NLP tools can not be built like compilers

but that is what makes language fascinating

Language obeys the “least effort principle”

general principles cover a large part of language

the effort to reach from 80 to 85% success is as expensive
as the effort to reach from 0 to 80% success

Two major kinds of analysis
statistical - automated learning
contextual - empirical,observation-based

NLP abstracts information from its materialization
words are decoys,lures...
levels of abstraction can be handled numerically

NLP tools are imperfect
due to the inherent ambiguity of language

keep this in mind but don’t reject them
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
The domains of Computational Linguistics
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
The linguistic units
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
“Low Level” NLP processes
Syntactic tagging
Attributing grammatical categories to words in context
New
adj
Trends
n(p)
in
prep
Information
n(s)
Technology
n(s)
Lemmatization
Attributing each word its lemma
New
new
Trends
trend
in
in
Information
information
Technology
technology
Syntactic parsing
Computing the grammatical structure of sentences
(New
adj
Trends
n(p)
)
NP
(in
prep
Information
n(s)
Technology
n(s)
)
PP→NP
Semantic categorisation
Attributing semantic categories to words or groups of words
task-dependent
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
Playing with word meaning
WordNet and EuroWordNet
lexical database developped by linguists (Princeton U.)
Synsets (synonym sets)
1
car,auto,automobile,machine,motorcar – (4-wheeled
motor vehicle;usually propelled by an internal combustion
engine;”he needs a car to get to work”)
2
car,railcar,railway car,railroad car – (a wheeled vehicle
adapted to the rails of railroad;”three cars had jumped
the rails”)
3
car,gondola – (car suspended from an airship and
carrying personnel and cargo and power plant)
4
car,elevator car – (where passengers ride up and down;
”the car was on the top floor”)
5
cable car,car – (a conveyance for passengers or freight on
a cable railway;”they took a cable car to the top of the
mountain”)
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
Playing with word meaning
Ontologies and semantic relations
car,auto,automobile,machine,motorcar
⇒motor vehicle,automotive vehicle
⇒vehicle
⇒conveyance,transport
⇒instrumentality,instrumentation
⇒artifact,artefact
⇒object,physical object
⇒entity,something
car,auto,automobile,machine,motorcar
HAS PART:
accelerator,accelerator pedal,gas pedal,gas,throttle,gun
HAS PART:
air bag
HAS PART:
auto accessory
HAS PART:
automobile engine
HAS PART:
automobile horn,car horn,motor horn,horn
(...)
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
Limitations of playing with word meaning
200.000 terms,150.000 synsets but still incomplete
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
From Documents to Numbers
Represent a set of documents as an n-dimension space
a dimension can be any linguistic measure or feature
occurrences of characters
occurrences of syntactic categories
occurrences of syntactic patterns
representative synsets
n-dim to 2-dim space
Sammon Projection (amongst others)
tries to respect distances between objects

TypText:text typology
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
From Documents to Numbers
Vector Space Model (VSM)
term-vectors to represent documents in a
multi-dimensional space
distance between documents in the VSM
Latent Semantic Analysis (LSA)
improved vector space model
uses matrix reduction
similar contexts paradigm
“understands” synonyms,metaphora...without resources
Document can also be a user query...
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
Q & A Systems
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
Q & A Systems
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Natural Language Processing
What’s more in NLP
Morphological analysis
Named entities extraction
Multi-lingual alignment
Anaphora resolution
Language identification
Translation memory and sentence alignment
Speech recognition
...
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Outline
1
About the author
2
Introduction
3
Natural Language Processing
4
Integrating NLP in Decision Support Systems:examples
5
Conclusions
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Preliminary
Definition
A decision support system (DSS) is a computer program
application that analyzes business data and presents it so
that users can make business decisions more easily
NLP for DSS
Natural Language Processing in Support of Decision-Making:
Phrases and Part-of-Speech tagging,
Losee R.,U.of North Carolina
NLP seen as a probabilistic process for statistical analysis
Personnal approach
Purely statistical approaches lose the richness of language
Integrating NLP competences for document analysis
is a concurrent process (not only a pre-process)
Language is a trace of human thinking and communication
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Application domain:Law and Trademarks Counseling
Law is a decision making domain (guilty or not guilty)
Law is multi-faceted - courts decision,legal advising,
legal domains...
Intellectual Property Law - Trademarks law
Trademark is an intangible property
Trademark bears the public image and financial value of a
business (product)
Trademarks are registered ￿ or not
TM
Infringement,loss,counterfeit or “passing off”
￿→unrecoverable financial and image loss
1.7 million marks currently registered in France
70 thousands registration per annum
4 to 5 thousands oppositions per annum
￿→
from characters to documents:examples of integration in
a nomminative trademarks decision support system../..
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Sub-word processing
Nominative trademarks have to be compared graphically
and phonetically
Need to identify language/manner of pronounciation
Language Identification
N-gram vector identification
NewMan →nevm˜a | niuman?
Need to transcript phonetically
Rule-based,language dependant transcription
[EN]:c+¬{ei} → k
[FR]:c+¬{ei} → k
[EN]:oo → u
[FR]:ou → u
Phonetic proximity of sounds ({t,d},{m,b},{f,s})
Phonetic encoding can be done either by proprietary
encoding or by Unicode-IPA
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Word splitting
Improving units of measures on trademark similarities
Morphemes
Zellig Harris,1909-1992
Herv´e D´ejean,1998
Morphemes of a language – atoms of the “word molecule”
– can be determined using frequency analysis
Syllables
language dependant →similar to hyphenation
related to phonetic transcription
Word concatenation →word splitting
Identification of concatenated words in ￿ for further
processing
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
“Word” processing
POS-Tagging
most queries are nouns or noun phrases
elimination or elision of non-relevant categories in queries
and documents
queries according to content words and modifiers
“contrefa¸con”:documents containing “contrefa¸con” and
not a negative tag in same sentences.
Lemmatisation
very low level approach of semantics
“foot” and “feet” both abstracted to “foot”
derivations of verbs all abstracted to the the same object
(information about tense,person...are in the POS-Tag)
Search engines and document indexation can be more
efficient
However some drawbacks:different meaning if
singular or plural (humanity,humanities),
if noun or adjective (orange),
according to gender (pound)
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Sentence processing
Sentence structure for contract writing aid
Phraseology,typical formulations,templates,integrating
“gaps” (constants relative to the contract)
Anaphora resolution
linking references to referant
→replacing anaphorae by their conceptual value
→improved search engines
Contextual structures as introductors to legal terms
Syntactic/lemmatic patterns delimitate terms
New terms can be found;their introductors used as patterns ￿
Parallel sentence processing and translation memories
Used as basis for professionnal translators
As well as for document search in other languages
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Document processing
Rhetorical analysis
Language units follow an organisational structure:
1
introductor (Jakobson’s phatic function)
2
informational content
3
closing mark (which can be next introductor)
probably acquired competence (as opposed to inate)
though multi-cultural
cf.Lucas’ work on theme/rheme identification in
document
macro-syntax
argumentative structure,detection of explanations...
identification of relevant information in document
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Corpus processing
Extracting documents according to linguistic criteria
Text extraction for (legal) language learning
Grammatical criteria
Thematic criteria (LSA)
Thematic criteria (Lemma and Wordnet)
Dicursive criteria
...and any combination...
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Human-Machine Interaction
Dialogue between user and machine can help clarify user
expectations
Can be keyboard-based or speech-based
“Co-construction du sens”
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Integrating NLP in DSS
Going further:modular conceptual representation
Current research led at Universit´e Grenoble 3 - France
Problem is:
Roman and anglosaxon caselaw obey a double hierarchy:
time and court
Finding the relevant case depends on:
decision features
hierarchy
argumentation structure
Need to build a conceptual representation
allowing for modularity – star-nested
automatically computable
aligned with text
comparable at case level
Can be a basis for document analysis in DSS
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Outline
1
About the author
2
Introduction
3
Natural Language Processing
4
Integrating NLP in Decision Support Systems:examples
5
Conclusions
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Conclusions
Call for inter-disciplinarity
NLP tools can be integrated in DSS as such
although NLP should not be seen as a simple toolbox
Generic tools have to be tailored to meet specific needs
Each decision domain has its own “special language”,
“technolecte”
Linguistics can modelize these specific language behaviours
Computational implementations can therefore improved
Better results can be achieved
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
Conclusions
Pitfalls to avoid in inter-disciplinariy
Inter-disciplinarity requires dialog and exchange
Computational linguist can not invent domain knowledge
Inter-disciplinarity is expensive and time-consuming
But as any tailor-made tool,it is usually worth the cost
￿→
Expected
Cost
using NLP
< Expected
Cost
using NLP
From NLP
to DSS
Thomas
LEBARB
´
E
About the
author
Introduction
Natural
Language
Processing
Integrating
NLP in DSS
Conclusions
From Natural Language Processing
to Desicion Support Systems
Natural Language Processing as a fundamental component
to improve communication in Decision Support Systems
Thomas LEBARB
´
E
Laboratoire LIDILEM EA609
Universit´e Stendhal - Grenoble 3
New Trends in Information Technology
Homs,Syria,18 avril 2006