Ray Kurzweil's Cybernetic Poet

geographertonguesΤεχνίτη Νοημοσύνη και Ρομποτική

30 Νοε 2013 (πριν από 3 χρόνια και 11 μήνες)

108 εμφανίσεις

An application of Markov models
(sorta)


Presented by Ehren Winterhof and
Josh Whitver

Ray Kurzweil’s Cybernetic Poet

Ray Kurzweil’s Cybernetic Poet


Poetry Analysis


Generates “Language Model”


Poetry Generation


Recursive


Goal Driven


Plagiarism Avoidance Algorithms


Thematic Consistency Algorithms



Poetry Analysis

Input


A collection of poems, usually by a
single author


Output


A “Markov model” of the
author’s style and a poet personality
file

Markov Models


First used by Andrei Markov in 1913 in a letter
-
sequence analysis of the text of
Eugene Onegin


Markov Chains


General Markov Models


Hidden Markov Models

Markov Models


Markov analysis takes a sequence of events
and models the statistical likelihood that one
event follows another.



Useful for analyzing dependent random
events (e.g., Weather, DNA Sequences,
Speech Recognition, etc)



Probability of transitioning to a given state
depends only on current state, prior states
are ignored. (the Markov Property)

Markov Model vs. Hidden
Markov Model

Regular Markov models output a sequence of
states


Each state has a unique name, so the
output uniquely determines the path
through the model


Hidden Markov Models can have the same
output appear in more than one state


Each state has a probability distribution of
possible outputs

t
1,1

t
1,2

t
2,2

t
2,end

1

2

end

p
1
(a)

p
1
(b)

p
2
(a)

p
2
(b)

1

1

2

end

a

b

a

Hidden state sequence, π

Observed symbol sequence,
x

t
1,1

t
1,2

t
2,end

p
1
(a) p
1
(b) p
2
(a)

P(
x
,π | HMM)

How a purely Markovian process
might generate poetry


Poetry analyzer reads sample poetry to
determine the likelihood that one word
follows another



From the start state take a random path to a
final state, picking up a word at each node
visited.



That’s it

How RKCP works (maybe)


Not “true” Markov Models



“Goal
-
driven” traversal



Separate algorithms
influence form and theme
of poem, while avoiding
plagiarism

Poet Personality


Defines how a poem should be generated
from the Poet Style model.



Parameters


How tightly to comply with the Poet
Personality model


Poem type : Free Verse, Haiku, Cinquain,
Structured, Thin, etc.


Theme Usage

Maintaining Thematic Consistency


Determined by Poet Personality



Keyword directs “train of thought”



Weighted Random Selection



Variable “Thematic Intensity”


Poetry
-
Specific Turing Test

Featuring Ray Kurzweil’s Cybernetic Poet

Sample 1

is a steady burning

the road the battle's fury
-

clouds and ash and waning

sending out

young people,

Sample 2

Wipe your hand across your mouth,
and laugh;

The worlds revolve like ancient women

Gathering fuel in vacant lots
.

Sample 3

0 thou,

Who moved among some fierce Maenad,
even among noise

and blue

Between the bones sang,


scattered and the silent seas.

Sample 4

Oh! did appear

A half
-
formed tear, a Tear.

By the man of the heart.


Sample 5

By action or by suffering, and whose hour

Was drained to its last sand in weal or woe,

So that the trunk survived both fruit or flower;
-

Answers

1.
RKCP after reading the poetry of


William Carlos Williams

2.
T. S. Eliot

3.
RKCP after reading T. S. Eliot and
Percy Bysshe Shelly

4.
RKCP after reading Lord Byron

5.
Percy Bysshe Shelley

Other AI Poetry Projects


Pujangga


Ruli Manurung (University of Indonesia)


Genetic Algorithm


Bahasa


Poetry CreatOR 2


Jeff Lewis and Erik Sincoff (Stanford)


Mad Libs


Racter


William Chamberlain (INRACK Corp)


First AI to “write” a book