Date post: | 08-Jan-2017 |
Category: |
Technology |
Upload: | dataversity |
View: | 633 times |
Download: | 4 times |
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Advances in Natural Language Processing
Adrian Bowles, PhDFounder, STORM Insights, Inc.
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Advances in Natural Language Processing Context: NLU vs NLGNLU
TechnologyMarketApplications
NLGTechnologyMarketApplications
Next Steps
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
“You’re Not Just Responsible For What You Say, You Are Responsible For What People Hear”
August 9, 2016
General Michael Hayden (Retired), former Director, Central Intelligence Agency and National Security Agency
on advice to his senior staff.
In the
News
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
X
X’
Y
Under ideal conditions, people are good - but not perfect - when communicating in natural languages. We…
understand in context (environment & our own frame of reference) attempt to resolve ambiguity have to deal with competing signals, noise fill in words and meaning and may not hear/understand - what was said/meant…
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Why is it so difficult?
NL Grammar Theories Abound
Generative Grammar - system of rules that specify the complete set of valid strings/sentences in a language. regular grammars, context free grammars…
Constraint-based Grammar - system of rules that specify constraints on strings/sentences in a language. Anything not constrained is valid.
Stochastic Grammar - “correctness” based on probability, similar to fuzzy set theory.
…
NLs are inherently ambiguous
Cultural differencesSarcasm
Idioms
Metaphors
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
In English, two negatives alway make a positive.
In English, two positives never make a negative.
That is not an uncommon occurrence. That is a common occurrence.
Yeah, right. Oops…sarcasm is hard.
Inflection matters, “literally” usually isn’t…
Rules…or guidelines?
Why is it so difficult?
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Generative Grammar - system of rules that specify the complete set of valid strings/sentences in a language. regular grammars, context free grammars…
The Infinite Monkey Theorem
<postal-address> ::= <name-part> <street-address> <zip-part>
<name-part> ::= <personal-part> <last-name> <opt-suffix-part> <EOL> | <personal-part> <name-part>
<personal-part> ::= <initial> "." | <first-name>
<street-address> ::= <house-num> <street-name> <opt-apt-num> <EOL>
<zip-part> ::= <town-name> "," <state-code> <ZIP-code> <EOL>
<opt-suffix-part> ::= "Sr." | "Jr." | <roman-numeral> | "" <opt-apt-num> ::= <apt-num> | ""
Backus–Naur Form. (2016, June 27). In Wikipedia, The Free Encyclopedia. Retrieved 14:42, August 11, 2016, from https://en.wikipedia.org/w/index.php?title=Backus%E2%80%93Naur_Form&oldid=727250296
Bob Smith 22 Main Street 06880
Why is it so difficult?
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Natural Language Processing NLP
Natural Language Understanding NLU
Natural Language Generation NLG
?
Key Concept: What is Understanding?
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
SemanticsText Syntax
Voic
e
Structure
Natural Language Understanding NLU
Meaning
Modeled Understanding
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Natural Language Processing NLP
Natural Language Understanding NLU
Natural Language Generation NLG
Computational Linguistics - modeling natural language with rules or statistical modelsStatic performance based on preprogrammed logic/model
vsLearning systems that improve performance based on feedback
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Natural Language Understanding NLU
Natural Language Generation NLG
Statistical Modeling
Semantic Analysis
Syntactic Analysis
Analysis Synthesis
Models &
Representations
Human
Sensors/Systems
Input/NLU Output/NLG
Visualization
Narrative GenerationVoice/NLP
Video/Images
Reports
Gestures
Emotions
Text/NLP
Surface Structured Data
Surface Structured Data
Reports
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
NLP In Context
• Gridspace• IBM• Maluuba• MindMeld• Nuance• PopupArchive• Skymind• Viv Labs• Wit.ai• ABBYY• Altilia• Cortical.io• Digital Reasoning• Google• IBM• LoopAI• Luminoso • Maluuba• Wit.ai
• Arria • Automated Insights• Ax-Semantics• Narrative Science• Retresco• Yseop
Perception/NLP
Problem Solving Simple:
deterministic, retrieve/calculate
Complex: probabalistichypothesize, test, rank,
selectCreative:
discover, generate
OR
GA
NIZ
EDM
emor
y*
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Google: Smart Reply (email responder)
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Proximity Modeling
BoyBay
Map
Mop
Man
Nay May
MopeBuy
Hop
Mapped with vectors, proximity algorithm based on purpose.
Mapping for autocorrect/complete vs Mapping for meaning
BoyBay
Map
Mop
Man
Nay
May
Mope
BuyHop
Hope
HopeSimilar structure -> similar meaning in vision, not always in language.
NLU Technology
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Google Cloud NLPFocus: Extract meaning
NLU
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
DATAModel/Article
Structure Template
Generated Narrative
From simple chattiest* to structured responses to prose.
*I actually typed “chatbot” but without “” the app “corrected” me.
Natural Language Generation NLG
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
DATA/ Model
Generated Narrative
Vendors: Automated Insights (Wordsmith) Narrative Science (Quill) Washington Post/Arc (Heliograf)
Arria (UK) Ax-Semantics (Germany) Retresco (Germany) Yseop (France)
Natural Language Generation Market NLG
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
DATA Ticket listings
Sale data Customer history
Model/Article
Structure Template
Application: NLG emails for customer satisfaction Developer: Orlando Magic Product: Automated Insights Wordsmith Time to Deploy: <1 week
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Sentiment/Emotion/Theme/Concept Analysis
The lingering question
What Is Understanding?
Don’t let the search for perfection interfere with the path to progress.
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Next Steps
For both NLU and NLG Commercial technologies today are imperfect but useful!
Do you want to derive data/insights from NL? (NLU)
Do you want to create content from data? (NLG)
Get Started Now
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved. Source: IBM Bluemix June 9, 2016
Build it Yourself With…
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
IBM Watson Conversation
Build it Yourself With…
For more information:
Copyright (c) 2016 by STORM Insights Inc. All Rights reserved.
Twitter @ajbowles Skype ajbowles
Upcoming Webinar Dates & Topics
September 8 Autonomous Systems, from Science Fiction to Commercial Solutions October 13 Deep QA (Question/Answer) - Lessons From Watson and Jeopardy! November 10 Emerging Hardware Choices for Modern AI Data Management December 8 Leverage the IOT to Build a Smart Data Ecosystem