SlideShare a Scribd company logo
Helping Computers
Find Meaning They
Lost in Translation
Katie Kuksenok UW Science Now!
University of Washington March 8, 2012
Hi! I’m Katie!
Computer Science and Engineering
PhD student
How people
interact with
computers
Richness of
written
language
Helping Computers Find Meaning they Lost in Translation
Posterity
Tranquility
Impeachment
liable Indictment
Concurrence
Google Searches
By Language
Google Searches
By Language
Machine Translation
Source language
Machine Translation
-----------------
-----------------
-----------------
-----------------
-----------------
---------
translation
Source language Target language
Machine Translation
+ Interaction
-----------------
-----------------
-----------------
-----------------
-----------------
---------
translation reading
Source language Target language
Machine Translation
+ Interaction
-----------------
-----------------
-----------------
-----------------
-----------------
--------- editing
translation reading
Source language Target language
Machine Translation
+ Interaction
-----------------
-----------------
-----------------
-----------------
-----------------
--------- editing
translation reading
Source language Target language
The Translating Machine
The Translating Machine
Artificial Intelligence
…where to begin?
What would a person do?
How can we describe that to a computer?
Rules & Dictionaries
?
Did it work?
Did it work?
The spirit is willing, but the flesh is weak
Did it work?
The spirit is willing, but the flesh is weak
(Russian)
Did it work?
The spirit is willing, but the flesh is weak
The vodka is good, but the meat is rotten
(Russian)
Rules & Dictionaries

Rosetta Stone
?
Translate this:
Source farok crrrok hihok yorok clok kantok yurp
Target ____ ____ _____ _____ ____ ______ ___
Translate this:
Source farok crrrok hihok yorok clok kantok yurp
Target ____ ____ _____ _____ ____ ______ ___
Source wiwok farok izok stok
Target sus clientes estan enfadados
Source lalok farok ororok lalok sprok izok enemok
Target los clientes y los asociados son enemigos
Based on 2 parallel pairs:
Source wiwok farok izok stok
Target sus clientes estan enfadados
Source lalok farok ororok lalok sprok izok enemok
Target los clientes y los asociados son enemigos
Translate this:
Source farok crrrok hihok yorok clok kantok yurp
Target ____ ____ _____ _____ ____ ______ ___
Based on 2 parallel pairs:
Source wiwok farok izok stok
Target sus clientes estan enfadados
Source lalok farok ororok lalok sprok izok enemok
Target los clientes y los asociados son enemigos
Translate this:
Source farok crrrok hihok yorok clok kantok yurp
Target ____ ____ _____ _____ ____ ______ ___
Based on 2 parallel pairs:
Source wiwok farok izok stok
Target sus clientes estan enfadados
Source lalok farok ororok lalok sprok izok enemok
Target los clientes y los asociados son enemigos
Translate this:
Source farok crrrok hihok yorok clok kantok yurp
Target ____ ____ _____ _____ ____ ______ ___
Based on 2 parallel pairs:
clientes
Machine Learning
Statistics + Data =
Surprisingly Good Results
More Data + Better Data = Better Results
DATA
Statistical Machine Translation
state of the art
1954 1966 1991 2012
Posterity
Tranquility
Impeachment
liable Indictment
Concurrence
How do people
learn different languages?
By Instruction? By Example?
By Failing
By Failing
Together
Interactive Machine Translation
?
Interactive Machine Translation
Human
Computation
Human Computation
source Vio una zorra al hambriento cuervo eternizado en la higuera, y le preguntó qué hacía.
initial Saw a vixen the hungry crow been in the fig tree, and asked you why did.
Human Computation
source Vio una zorra al hambriento cuervo eternizado en la higuera, y le preguntó qué hacía.
initial Saw a vixen the hungry crow been in the fig tree, and asked you why did.
reorder [vixen] Saw a [vixen] he hungry crow been in the fig tree, and asked you why did.
reorder [a] vixen Saw [a] the hungry crow been in the fig tree, and asked you why did.
retype {A} vixen Saw the hungry crow been in the fig tree, and asked you why did.
retype A vixen {saw} the hungry crow been in the fig tree, and asked you why did.
Human Computation
source Vio una zorra al hambriento cuervo eternizado en la higuera, y le preguntó qué hacía.
initial Saw a vixen the hungry crow been in the fig tree, and asked you why did.
reorder [vixen] Saw a [vixen] he hungry crow been in the fig tree, and asked you why did.
reorder [a] vixen Saw [a] the hungry crow been in the fig tree, and asked you why did.
retype {A} vixen Saw the hungry crow been in the fig tree, and asked you why did.
retype A vixen {saw} the hungry crow been in the fig tree, and asked you why did.
delete A vixen saw the hungry crow [been] in the fig tree, and asked you why did.
delete A vixen saw the hungry crow in the fig tree, and asked [you] why did.
replace A vixen saw the hungry crow in the fig tree, and asked [what was he doing] did.
Human Computation
source Vio una zorra al hambriento cuervo eternizado en la higuera, y le preguntó qué hacía.
initial Saw a vixen the hungry crow been in the fig tree, and asked you why did.
reorder [vixen] Saw a [vixen] he hungry crow been in the fig tree, and asked you why did.
reorder [a] vixen Saw [a] the hungry crow been in the fig tree, and asked you why did.
retype {A} vixen Saw the hungry crow been in the fig tree, and asked you why did.
retype A vixen {saw} the hungry crow been in the fig tree, and asked you why did.
delete A vixen saw the hungry crow [been] in the fig tree, and asked you why did.
delete A vixen saw the hungry crow in the fig tree, and asked [you] why did.
replace A vixen saw the hungry crow in the fig tree, and asked [what was he doing] did.
retype A vixen saw the hungry crow in the fig tree, and asked {why} was he doing did.
reorder A vixen saw the hungry crow in the fig tree, and asked why [he] was [he] doing did.
retype A vixen saw the hungry crow in the fig tree, and asked why he was {there} did.
delete A vixen saw the hungry crow in the fig tree, and asked why he was there [did].
final A vixen saw the hungry crow in the fig tree, and asked why he was there.
Human Computation
source Vio una zorra al hambriento cuervo eternizado en la higuera, y le preguntó qué hacía.
initial Saw a vixen the hungry crow been in the fig tree, and asked you why did.
reorder [vixen] Saw a [vixen] he hungry crow been in the fig tree, and asked you why did.
[a] vixen Saw [a] the hungry crow been in the fig tree, and asked you why did.
retype {A} vixen Saw the hungry crow been in the fig tree, and asked you why did.
A vixen {saw} the hungry crow been in the fig tree, and asked you why did.
delete A vixen saw the hungry crow [been] in the fig tree, and asked you why did.
A vixen saw the hungry crow in the fig tree, and asked [you] why did.
replace A vixen saw the hungry crow in the fig tree, and asked [what was he doing] did.
retype A vixen saw the hungry crow in the fig tree, and asked {why} was he doing did.
reorder A vixen saw the hungry crow in the fig tree, and asked why [he] was [he] doing did.
retype A vixen saw the hungry crow in the fig tree, and asked why he was {there} did.
delete A vixen saw the hungry crow in the fig tree, and asked why he was there [did].
final A vixen saw the hungry crow in the fig tree, and asked why he was there.
x 100
Human Computation
x 100…
= Human Insight as Data
source Vio una zorra al hambriento cuervo eternizado en la higuera, y le preguntó qué hacía.
initial Saw a vixen the hungry crow been in the fig tree, and asked you why did.
reorder [vixen] Saw a [vixen] he hungry crow been in the fig tree, and asked you why did.
[a] vixen Saw [a] the hungry crow been in the fig tree, and asked you why did.
retype {A} vixen Saw the hungry crow been in the fig tree, and asked you why did.
A vixen {saw} the hungry crow been in the fig tree, and asked you why did.
delete A vixen saw the hungry crow [been] in the fig tree, and asked you why did.
A vixen saw the hungry crow in the fig tree, and asked [you] why did.
replace A vixen saw the hungry crow in the fig tree, and asked [what was he doing] did.
retype A vixen saw the hungry crow in the fig tree, and asked {why} was he doing did.
reorder A vixen saw the hungry crow in the fig tree, and asked why [he] was [he] doing did.
retype A vixen saw the hungry crow in the fig tree, and asked why he was {there} did.
delete A vixen saw the hungry crow in the fig tree, and asked why he was there [did].
final A vixen saw the hungry crow in the fig tree, and asked why he was there.
DATA
DATA
 Human Insight as Data
Human Computation + Translation
How can we
capture the
editing process ?
source Vio una zorra al hambriento cuervo eternizado en la higuera, y le preguntó qué hacía.
initial Saw a vixen the hungry crow been in the fig tree, and asked you why did.
reorder [vixen] Saw a [vixen] he hungry crow been in the fig tree, and asked you why did.
[a] vixen Saw [a] the hungry crow been in the fig tree, and asked you why did.
retype {A} vixen Saw the hungry crow been in the fig tree, and asked you why did.
A vixen {saw} the hungry crow been in the fig tree, and asked you why did.
delete A vixen saw the hungry crow [been] in the fig tree, and asked you why did.
A vixen saw the hungry crow in the fig tree, and asked [you] why did.
replace A vixen saw the hungry crow in the fig tree, and asked [what was he doing] did.
retype A vixen saw the hungry crow in the fig tree, and asked {why} was he doing did.
reorder A vixen saw the hungry crow in the fig tree, and asked why [he] was [he] doing did.
retype A vixen saw the hungry crow in the fig tree, and asked why he was {there} did.
delete A vixen saw the hungry crow in the fig tree, and asked why he was there [did].
final A vixen saw the hungry crow in the fig tree, and asked why he was there.
Interaction
Drag and drop to reorder
Dropdowns with suggestions
Interaction
Drag and drop to reorder
vs
Dropdowns with suggestions
Free type
Interaction
Suggested Alternative Translations
Many Hypotheses
Interaction
Suggested Alternative Translations
Interaction
Suggested Alternative Translations
Interaction + Translation
Translator
expertise
NoviceExpert
NoviceExpert
Caitra (2010)
Interaction + Translation
Translator
expertise
NoviceExpert
Caitra (2010)
TransType (2000)
Interaction + Translation
Translator
expertise
NoviceExpert
Google Translate
Caitra (2010)
TransType (2000)
Interaction + Translation
Translator
expertise
Google translate
data powers
better systems
NoviceExpert
Google Translate
Caitra (2010)
Active Crowd
Translation (2010)
TransType (2000)
Interaction + Translation
Translator
expertise
data powers
better systems
collaboration +
discussion
NoviceExpert
Google Translate
MonoTrans (2011)Caitra (2010)
Active Crowd
Translation (2010)
TransType (2000)
Interaction + Translation
Translator
expertise
Beyond Human Computation
Beyond Human Computation
Munteanu, Baecker, and Penn, CHI 2008
Beyond Human Computation
Duolingo.com
Interactive Machine Translation
What if people interact with translation?
How can a computer learn from that?
How powerful is this approach?
Production: www.paperhand.org | Photos: mlight.typepad.com
A Sea of Information
Production: www.paperhand.org | Photos: mlight.typepad.com
A City Obsessed
with More and Newer
Production: www.paperhand.org | Photos: mlight.typepad.com
An Insatiable,
Hungry Monster
Production: www.paperhand.org | Photos: mlight.typepad.com
Helping Computers Find Meaning they Lost in Translation
An Insatiable,
Hungry Monster
Production: www.paperhand.org | Photos: mlight.typepad.com
An Insatiable,
Hungry Monster
Access?
Production: www.paperhand.org | Photos: mlight.typepad.com
Effective Access
Production: www.paperhand.org | Photos: mlight.typepad.com
Google Searches
By Language
Thanks to:
• Mentors
– James Fogarty (UW)
– Srinivas Bangalore (AT&T)
– Luke Zettlemoyer (UW)
– Dan Weld (UW)
– Charlotte P. Lee (UW)
– Jen Mankoff (CMU)
• And many more
– Study Participants
– Colleagues
– You!
• Supporting Organizations
– National Science Foundation
– AT&T Labs
– Microsoft Research
– Google
– University Book Store
– Engage: The Science Speaker
Series
– University of Washington
– Pacific Science Center
– Town Hall Seattle
– Microsoft
Thank You!
Questions?
Contact:
kuksenok@cs.uw.edu
Links, Citations, etc:
students.washington.edu/kuksenok/engage

More Related Content

PDF
2024 Trend Updates: What Really Works In SEO & Content Marketing
PPT
Storyboard(printscreen)
PDF
Games on Social Networks: Balancing Friendships
PDF
Games on Social Networks: Constructing Valuable Networks
PPTX
Uncertainty in Chronic Illness and Patients' Online Experience
PPTX
Nursing theory power point
PPTX
Mishel's Uncertainty in Illness Theory
PDF
IP Clustering as Exploratory Data Analysis
2024 Trend Updates: What Really Works In SEO & Content Marketing
Storyboard(printscreen)
Games on Social Networks: Balancing Friendships
Games on Social Networks: Constructing Valuable Networks
Uncertainty in Chronic Illness and Patients' Online Experience
Nursing theory power point
Mishel's Uncertainty in Illness Theory
IP Clustering as Exploratory Data Analysis

Recently uploaded (20)

PDF
Clinical guidelines as a resource for EBP(1).pdf
PPTX
oil_refinery_comprehensive_20250804084928 (1).pptx
PPTX
iec ppt-1 pptx icmr ppt on rehabilitation.pptx
PDF
168300704-gasification-ppt.pdfhghhhsjsjhsuxush
PPTX
Introduction to Basics of Ethical Hacking and Penetration Testing -Unit No. 1...
PDF
annual-report-2024-2025 original latest.
PPTX
Introduction-to-Cloud-ComputingFinal.pptx
PDF
Introduction to Data Science and Data Analysis
PDF
Introduction to the R Programming Language
PDF
BF and FI - Blockchain, fintech and Financial Innovation Lesson 2.pdf
PPTX
The THESIS FINAL-DEFENSE-PRESENTATION.pptx
PPTX
Business Ppt On Nestle.pptx huunnnhhgfvu
PDF
22.Patil - Early prediction of Alzheimer’s disease using convolutional neural...
PPTX
DISORDERS OF THE LIVER, GALLBLADDER AND PANCREASE (1).pptx
PPTX
Database Infoormation System (DBIS).pptx
PPTX
Computer network topology notes for revision
PPTX
climate analysis of Dhaka ,Banglades.pptx
PDF
Mega Projects Data Mega Projects Data
PPTX
Introduction to machine learning and Linear Models
Clinical guidelines as a resource for EBP(1).pdf
oil_refinery_comprehensive_20250804084928 (1).pptx
iec ppt-1 pptx icmr ppt on rehabilitation.pptx
168300704-gasification-ppt.pdfhghhhsjsjhsuxush
Introduction to Basics of Ethical Hacking and Penetration Testing -Unit No. 1...
annual-report-2024-2025 original latest.
Introduction-to-Cloud-ComputingFinal.pptx
Introduction to Data Science and Data Analysis
Introduction to the R Programming Language
BF and FI - Blockchain, fintech and Financial Innovation Lesson 2.pdf
The THESIS FINAL-DEFENSE-PRESENTATION.pptx
Business Ppt On Nestle.pptx huunnnhhgfvu
22.Patil - Early prediction of Alzheimer’s disease using convolutional neural...
DISORDERS OF THE LIVER, GALLBLADDER AND PANCREASE (1).pptx
Database Infoormation System (DBIS).pptx
Computer network topology notes for revision
climate analysis of Dhaka ,Banglades.pptx
Mega Projects Data Mega Projects Data
Introduction to machine learning and Linear Models
Ad
Ad

Helping Computers Find Meaning they Lost in Translation

Editor's Notes

  • #14: <when>
  • #15: AI Problem -> spam -> for translation “So how would a person figure out how to translate between languages?”
  • #22: Jean-François Champollion
  • #23: Kevin Knight
  • #24: Kevin Knight
  • #25: Kevin Knight
  • #26: Kevin Knight
  • #27: Kevin Knight
  • #45: http://stylescorner.com/wp-content/uploads/2011/11/Lemon-Skin-Care-Natural-Recipes1.jpg
  • #62: Last summer I saw a play I see technology as incredibly empowering. Providing access to this vast sea of infromation to me means providing the freedom to decide what to believe, and the
  • #64: aMALgam
  • #66: CONTROVERSIAL – transformation – continuing process we’re all experiencing
  • #70: As more people get connected from other countries and cultures, more content is produced online. There is a lot of effort right now by people in computer science and engineering to make technology more available throughout the world, to give as many people as possible access to information online and a platform for producing their own content. And as that happens, we should expect more colors on this map, not less.