SlideShare una empresa de Scribd logo
Complexity Explorers Kraków #7
cekrk.com Marcin Stępień @marcinstepien Kraków 2023-02-22
Language models
and power laws
“Industrial City” 1920, by Leon Chwistek
The Queen of Laputa, Gulliver's Travels
Complexity
Explorers
Kraków
Inspired by Santa Fe Institute
Algorithmic
Information
Theory
Systems
Theory
Computational
Complexity
www.santafe.edu
discoveries
models
simulations
experiments
physical/numerical/thought
opinions
statements
influences
>
❏ There is hardly any book that contains a large number of
occurrences of both word lemma and word love.
Yaglom and Yaglom [1983]
* Information Theory Meets Power Laws: Stochastic Processes and Language Models (Wiley), Łukasz Dębowski, 2021
[Ł. Dębowski]*
Language model
Evaluate
probabilities on
sequence of
(sub)words or
characters
Read
Text
Generate
Text
... this invention had employed all his
thoughts from his youth; that he had emptied
the whole vocabulary into his frame, and
made the strictest computation of the general
proportion there is in books between the
numbers of particles, nouns, and verbs, and
other parts of speech.
[Jonathan Swift, 1726]
Gulliver’s Travels, Part III
A not so new idea
Les moniteurs de la
conversation á Laputa, 1875
The Engine
❏ Language models based on Markov Chain
❏ Andrey Markov [1906]
❏ “memory-less” - next step in the process is only dependent on the
immediate previous step, or a fixed number of previous step
❏ N-gram
❏ Mathematical Theory of Communications, Claude Shannon [1948]
❏ text generation, names generation, word suggestion, speech
recognition, OCR, protein sequencing
Before Generative Pre-trained Transformer GPT
* GPT is a kind of Markov Process (continuous-time version of Markov Chain)
src: www.adamwalanus.pl/2016/chaitin/160519-1804-19.jpg
The Zipf Mystery to 1:30 The Zipf Mystery from 6:52 to 12:30
❏ There is hardly any book that contains a large number of
occurrences of both word lemma and word love.
Yaglom and Yaglom [1983]
[Ł. Dębowski]
❏ Rejecting simple determinism from the process of natural text
generation means that we have to consider some model of
randomness.
[Ł. Dębowski]
❏ Why only a certain well-defined number of texts or sequences
of symbols of a given length gets reproduced in the process of
human culture or in the stream of consciousness?
[Ł. Dębowski]
❏ …Thus we can formally demonstrate that natural language is
NOT a hidden Markov process
[Ł. Dębowski]
❏ …natural texts, i.e., texts written in natural language, exhibit certain
statistical regularities. Some of these regularities are similar to
statistical regularities of unigram texts, i.e., random permutations
of characters, whereas some are strikingly different.
[Ł. Dębowski]
❏ For possible applications in engineering and artificial
intelligence, it is also vital to understand general constructions
of stochastic processes that satisfy these power laws as well as
to comprehend reasons why these laws are satisfied.
[Ł. Dębowski]
❏ similar words are close to each
other in the embedding space
❏ unsupervised - Word2Vec [2013] *
Word Embeddings
0.1 0.137 0.01 … 0.07 0.42
Word
❏ rare words, not in training data;
morphologically rich languages -
FastText [2016]
❏ handles homonyms, typos - ELMo
[2019]
* the morning paper introduction:
blog.acolyer.org/2016/04/21/the-amazing-power-of-word-vectors/
King - Man + Woman = Queen
WE use case: semantic change of words over time
source: https://github.com/williamleif/histwords
Łukasz Dębowski
Language Models and Power Laws
Thank you

Más contenido relacionado

Similar a CEK 7 Language Models and Power Laws. Complexity Explorers Krakow.

Digital Humanities and Computer Assisted Literary Criticism
Digital Humanities and Computer Assisted Literary CriticismDigital Humanities and Computer Assisted Literary Criticism
Digital Humanities and Computer Assisted Literary Criticism
Dilip Barad
 
Book review of language and the internet
Book review of language and the internetBook review of language and the internet
Book review of language and the internet
Hina Honey
 
essay of 1,700-1,900 words.  Select and read one of the following .docx
essay of 1,700-1,900 words.  Select and read one of the following .docxessay of 1,700-1,900 words.  Select and read one of the following .docx
essay of 1,700-1,900 words.  Select and read one of the following .docx
debishakespeare
 
Macbeth Essay Question. Macbeth essay English - Level 2 NCEA Thinkswap
Macbeth Essay Question. Macbeth essay  English - Level 2 NCEA  ThinkswapMacbeth Essay Question. Macbeth essay  English - Level 2 NCEA  Thinkswap
Macbeth Essay Question. Macbeth essay English - Level 2 NCEA Thinkswap
Ashley Champs
 
Electronic Literature - Honors Project Narrative (Final Draft)
Electronic Literature - Honors Project Narrative (Final Draft)Electronic Literature - Honors Project Narrative (Final Draft)
Electronic Literature - Honors Project Narrative (Final Draft)
Cameron Irby
 

Similar a CEK 7 Language Models and Power Laws. Complexity Explorers Krakow. (18)

Judy Malloy Short PP
Judy Malloy Short PPJudy Malloy Short PP
Judy Malloy Short PP
 
Digital Humanities and Computer Assisted Literary Criticism
Digital Humanities and Computer Assisted Literary CriticismDigital Humanities and Computer Assisted Literary Criticism
Digital Humanities and Computer Assisted Literary Criticism
 
Syntactic Structures (2nd Edition).pdf
Syntactic Structures (2nd Edition).pdfSyntactic Structures (2nd Edition).pdf
Syntactic Structures (2nd Edition).pdf
 
Code-Mixing and Code Switching
 Code-Mixing and Code Switching Code-Mixing and Code Switching
Code-Mixing and Code Switching
 
Book review of language and the internet
Book review of language and the internetBook review of language and the internet
Book review of language and the internet
 
A NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITS
A NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITSA NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITS
A NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITS
 
A Natural Logic for Artificial Intelligence, and its Risks and Benefits
A Natural Logic for Artificial Intelligence, and its Risks and Benefits A Natural Logic for Artificial Intelligence, and its Risks and Benefits
A Natural Logic for Artificial Intelligence, and its Risks and Benefits
 
A NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITS
A NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITSA NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITS
A NATURAL LOGIC FOR ARTIFICIAL INTELLIGENCE, AND ITS RISKS AND BENEFITS
 
essay of 1,700-1,900 words.  Select and read one of the following .docx
essay of 1,700-1,900 words.  Select and read one of the following .docxessay of 1,700-1,900 words.  Select and read one of the following .docx
essay of 1,700-1,900 words.  Select and read one of the following .docx
 
Natural Language Generation: Breaking the Hermeneutic Contract
Natural Language Generation: Breaking the Hermeneutic ContractNatural Language Generation: Breaking the Hermeneutic Contract
Natural Language Generation: Breaking the Hermeneutic Contract
 
Macbeth Essay Question. Macbeth essay English - Level 2 NCEA Thinkswap
Macbeth Essay Question. Macbeth essay  English - Level 2 NCEA  ThinkswapMacbeth Essay Question. Macbeth essay  English - Level 2 NCEA  Thinkswap
Macbeth Essay Question. Macbeth essay English - Level 2 NCEA Thinkswap
 
Electronic Literature - Honors Project Narrative (Final Draft)
Electronic Literature - Honors Project Narrative (Final Draft)Electronic Literature - Honors Project Narrative (Final Draft)
Electronic Literature - Honors Project Narrative (Final Draft)
 
Natural language-processing
Natural language-processingNatural language-processing
Natural language-processing
 
Linguagem, Lógica e a Natureza da Matemática
Linguagem, Lógica e a Natureza da MatemáticaLinguagem, Lógica e a Natureza da Matemática
Linguagem, Lógica e a Natureza da Matemática
 
The New Past, and a Speculative Future, of Literature: A Brief Discussion of ...
The New Past, and a Speculative Future, of Literature: A Brief Discussion of ...The New Past, and a Speculative Future, of Literature: A Brief Discussion of ...
The New Past, and a Speculative Future, of Literature: A Brief Discussion of ...
 
50 Self Evaluation Examples, Forms Questions TemplateLab
50 Self Evaluation Examples, Forms Questions TemplateLab50 Self Evaluation Examples, Forms Questions TemplateLab
50 Self Evaluation Examples, Forms Questions TemplateLab
 
Hacking Human Language (PyData London)
Hacking Human Language (PyData London)Hacking Human Language (PyData London)
Hacking Human Language (PyData London)
 
11 terms in Corpus Linguistics1 (2)
11 terms in Corpus Linguistics1 (2)11 terms in Corpus Linguistics1 (2)
11 terms in Corpus Linguistics1 (2)
 

Más de Marcin Stepien

Más de Marcin Stepien (6)

Natural born algorithms. Complexity Explorers Krakow.
Natural born algorithms. Complexity Explorers Krakow. Natural born algorithms. Complexity Explorers Krakow.
Natural born algorithms. Complexity Explorers Krakow.
 
Limits of AI. The Gödelian argument. Complexity Explorers Krakow.
Limits of AI. The Gödelian argument. Complexity Explorers Krakow. Limits of AI. The Gödelian argument. Complexity Explorers Krakow.
Limits of AI. The Gödelian argument. Complexity Explorers Krakow.
 
Randomness from Determinism. Complexity Explorers Krakow.
Randomness from Determinism. Complexity Explorers Krakow.Randomness from Determinism. Complexity Explorers Krakow.
Randomness from Determinism. Complexity Explorers Krakow.
 
Complexity Explorers Krakow - Computer Science & Philosophy
Complexity Explorers Krakow - Computer Science & PhilosophyComplexity Explorers Krakow - Computer Science & Philosophy
Complexity Explorers Krakow - Computer Science & Philosophy
 
Functional Programming in Java - Code for Maintainability
Functional Programming in Java - Code for MaintainabilityFunctional Programming in Java - Code for Maintainability
Functional Programming in Java - Code for Maintainability
 
Play Framework on Google App Engine - Productivity Stack
Play Framework on Google App Engine - Productivity StackPlay Framework on Google App Engine - Productivity Stack
Play Framework on Google App Engine - Productivity Stack
 

Último

Continuum emission from within the plunging region of black hole discs
Continuum emission from within the plunging region of black hole discsContinuum emission from within the plunging region of black hole discs
Continuum emission from within the plunging region of black hole discs
Sérgio Sacani
 
The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...
Sérgio Sacani
 
Aerodynamics. flippatterncn5tm5ttnj6nmnynyppt
Aerodynamics. flippatterncn5tm5ttnj6nmnynypptAerodynamics. flippatterncn5tm5ttnj6nmnynyppt
Aerodynamics. flippatterncn5tm5ttnj6nmnynyppt
sreddyrahul
 
Jet reorientation in central galaxies of clusters and groups: insights from V...
Jet reorientation in central galaxies of clusters and groups: insights from V...Jet reorientation in central galaxies of clusters and groups: insights from V...
Jet reorientation in central galaxies of clusters and groups: insights from V...
Sérgio Sacani
 
Quantifying Artificial Intelligence and What Comes Next!
Quantifying Artificial Intelligence and What Comes Next!Quantifying Artificial Intelligence and What Comes Next!
Quantifying Artificial Intelligence and What Comes Next!
University of Hertfordshire
 

Último (20)

Continuum emission from within the plunging region of black hole discs
Continuum emission from within the plunging region of black hole discsContinuum emission from within the plunging region of black hole discs
Continuum emission from within the plunging region of black hole discs
 
Alternative method of dissolution in-vitro in-vivo correlation and dissolutio...
Alternative method of dissolution in-vitro in-vivo correlation and dissolutio...Alternative method of dissolution in-vitro in-vivo correlation and dissolutio...
Alternative method of dissolution in-vitro in-vivo correlation and dissolutio...
 
GBSN - Microbiology Lab 2 (Compound Microscope)
GBSN - Microbiology Lab 2 (Compound Microscope)GBSN - Microbiology Lab 2 (Compound Microscope)
GBSN - Microbiology Lab 2 (Compound Microscope)
 
Microbial bio Synthesis of nanoparticles.pptx
Microbial bio Synthesis of nanoparticles.pptxMicrobial bio Synthesis of nanoparticles.pptx
Microbial bio Synthesis of nanoparticles.pptx
 
Virulence Analysis of Citrus canker caused by Xanthomonas axonopodis pv. citr...
Virulence Analysis of Citrus canker caused by Xanthomonas axonopodis pv. citr...Virulence Analysis of Citrus canker caused by Xanthomonas axonopodis pv. citr...
Virulence Analysis of Citrus canker caused by Xanthomonas axonopodis pv. citr...
 
GBSN - Microbiology Lab 1 (Microbiology Lab Safety Procedures)
GBSN -  Microbiology Lab  1 (Microbiology Lab Safety Procedures)GBSN -  Microbiology Lab  1 (Microbiology Lab Safety Procedures)
GBSN - Microbiology Lab 1 (Microbiology Lab Safety Procedures)
 
The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...
 
Aerodynamics. flippatterncn5tm5ttnj6nmnynyppt
Aerodynamics. flippatterncn5tm5ttnj6nmnynypptAerodynamics. flippatterncn5tm5ttnj6nmnynyppt
Aerodynamics. flippatterncn5tm5ttnj6nmnynyppt
 
A Giant Impact Origin for the First Subduction on Earth
A Giant Impact Origin for the First Subduction on EarthA Giant Impact Origin for the First Subduction on Earth
A Giant Impact Origin for the First Subduction on Earth
 
GBSN - Microbiology (Unit 7) Microbiology in Everyday Life
GBSN - Microbiology (Unit 7) Microbiology in Everyday LifeGBSN - Microbiology (Unit 7) Microbiology in Everyday Life
GBSN - Microbiology (Unit 7) Microbiology in Everyday Life
 
Plasmapheresis - Dr. E. Muralinath - Kalyan . C.pptx
Plasmapheresis - Dr. E. Muralinath - Kalyan . C.pptxPlasmapheresis - Dr. E. Muralinath - Kalyan . C.pptx
Plasmapheresis - Dr. E. Muralinath - Kalyan . C.pptx
 
INSIGHT Partner Profile: Tampere University
INSIGHT Partner Profile: Tampere UniversityINSIGHT Partner Profile: Tampere University
INSIGHT Partner Profile: Tampere University
 
Constraints on Neutrino Natal Kicks from Black-Hole Binary VFTS 243
Constraints on Neutrino Natal Kicks from Black-Hole Binary VFTS 243Constraints on Neutrino Natal Kicks from Black-Hole Binary VFTS 243
Constraints on Neutrino Natal Kicks from Black-Hole Binary VFTS 243
 
Cell Immobilization Methods and Applications.pptx
Cell Immobilization Methods and Applications.pptxCell Immobilization Methods and Applications.pptx
Cell Immobilization Methods and Applications.pptx
 
Jet reorientation in central galaxies of clusters and groups: insights from V...
Jet reorientation in central galaxies of clusters and groups: insights from V...Jet reorientation in central galaxies of clusters and groups: insights from V...
Jet reorientation in central galaxies of clusters and groups: insights from V...
 
Quantifying Artificial Intelligence and What Comes Next!
Quantifying Artificial Intelligence and What Comes Next!Quantifying Artificial Intelligence and What Comes Next!
Quantifying Artificial Intelligence and What Comes Next!
 
ERTHROPOIESIS: Dr. E. Muralinath & R. Gnana Lahari
ERTHROPOIESIS: Dr. E. Muralinath & R. Gnana LahariERTHROPOIESIS: Dr. E. Muralinath & R. Gnana Lahari
ERTHROPOIESIS: Dr. E. Muralinath & R. Gnana Lahari
 
METHODS OF TRANSCRIPTOME ANALYSIS....pptx
METHODS OF TRANSCRIPTOME ANALYSIS....pptxMETHODS OF TRANSCRIPTOME ANALYSIS....pptx
METHODS OF TRANSCRIPTOME ANALYSIS....pptx
 
Plasma proteins_ Dr.Muralinath_Dr.c. kalyan
Plasma proteins_ Dr.Muralinath_Dr.c. kalyanPlasma proteins_ Dr.Muralinath_Dr.c. kalyan
Plasma proteins_ Dr.Muralinath_Dr.c. kalyan
 
PLANT DISEASE MANAGEMENT PRINCIPLES AND ITS IMPORTANCE
PLANT DISEASE MANAGEMENT PRINCIPLES AND ITS IMPORTANCEPLANT DISEASE MANAGEMENT PRINCIPLES AND ITS IMPORTANCE
PLANT DISEASE MANAGEMENT PRINCIPLES AND ITS IMPORTANCE
 

CEK 7 Language Models and Power Laws. Complexity Explorers Krakow.

  • 1. Complexity Explorers Kraków #7 cekrk.com Marcin Stępień @marcinstepien Kraków 2023-02-22 Language models and power laws “Industrial City” 1920, by Leon Chwistek The Queen of Laputa, Gulliver's Travels
  • 3. Inspired by Santa Fe Institute Algorithmic Information Theory Systems Theory Computational Complexity www.santafe.edu
  • 5. ❏ There is hardly any book that contains a large number of occurrences of both word lemma and word love. Yaglom and Yaglom [1983] * Information Theory Meets Power Laws: Stochastic Processes and Language Models (Wiley), Łukasz Dębowski, 2021 [Ł. Dębowski]*
  • 6. Language model Evaluate probabilities on sequence of (sub)words or characters Read Text Generate Text
  • 7. ... this invention had employed all his thoughts from his youth; that he had emptied the whole vocabulary into his frame, and made the strictest computation of the general proportion there is in books between the numbers of particles, nouns, and verbs, and other parts of speech. [Jonathan Swift, 1726] Gulliver’s Travels, Part III A not so new idea Les moniteurs de la conversation á Laputa, 1875 The Engine
  • 8. ❏ Language models based on Markov Chain ❏ Andrey Markov [1906] ❏ “memory-less” - next step in the process is only dependent on the immediate previous step, or a fixed number of previous step ❏ N-gram ❏ Mathematical Theory of Communications, Claude Shannon [1948] ❏ text generation, names generation, word suggestion, speech recognition, OCR, protein sequencing Before Generative Pre-trained Transformer GPT * GPT is a kind of Markov Process (continuous-time version of Markov Chain)
  • 9. src: www.adamwalanus.pl/2016/chaitin/160519-1804-19.jpg The Zipf Mystery to 1:30 The Zipf Mystery from 6:52 to 12:30
  • 10. ❏ There is hardly any book that contains a large number of occurrences of both word lemma and word love. Yaglom and Yaglom [1983] [Ł. Dębowski]
  • 11. ❏ Rejecting simple determinism from the process of natural text generation means that we have to consider some model of randomness. [Ł. Dębowski]
  • 12. ❏ Why only a certain well-defined number of texts or sequences of symbols of a given length gets reproduced in the process of human culture or in the stream of consciousness? [Ł. Dębowski]
  • 13. ❏ …Thus we can formally demonstrate that natural language is NOT a hidden Markov process [Ł. Dębowski]
  • 14. ❏ …natural texts, i.e., texts written in natural language, exhibit certain statistical regularities. Some of these regularities are similar to statistical regularities of unigram texts, i.e., random permutations of characters, whereas some are strikingly different. [Ł. Dębowski]
  • 15. ❏ For possible applications in engineering and artificial intelligence, it is also vital to understand general constructions of stochastic processes that satisfy these power laws as well as to comprehend reasons why these laws are satisfied. [Ł. Dębowski]
  • 16. ❏ similar words are close to each other in the embedding space ❏ unsupervised - Word2Vec [2013] * Word Embeddings 0.1 0.137 0.01 … 0.07 0.42 Word ❏ rare words, not in training data; morphologically rich languages - FastText [2016] ❏ handles homonyms, typos - ELMo [2019] * the morning paper introduction: blog.acolyer.org/2016/04/21/the-amazing-power-of-word-vectors/ King - Man + Woman = Queen
  • 17. WE use case: semantic change of words over time source: https://github.com/williamleif/histwords