Unstructured Data and the Role of Natural Language Processing

1
Unstructured Data and the Role of Natural
Language Processing

• Philip Resnik
• Department of Linguistics and
• Institute for Advanced Computer Studies
• University of Maryland
• Wolfram Data Summit
• September 7, 2012

2
Hiya, guys. What did you think of Obamas speech
last night? I think I liked Michelles better.
? Daisy, Daisy
Beep.
I didnt watch it. I was playing a nice game of
chess.
3
(No Transcript)
4
The statistical revolution in NLP
100
80
60
40
20
0
1985
1990
1995
2000
2005
Annual Meeting of the Association for
Computational Linguistics
AI Winter
Sources graph adapted from Church, K. (2003)
Speech and Language Processing Where have we
been and where are we going, Eurospeech, Geneva,
Switzerland. Green circle data have been added
from figures in Cardie and Mooney (1999).
5
NLP is no longer about getting machines to
understand language like people do.
Its about building machines that do things with
language that people find useful.
6
Surface methods
7
Surface methods plus categories
Pennebaker, Linguistic Inquiry and Word
Count http//www.liwc.net/tryonlineresults.php
8
Surface methods plus categories
Brendan OConnor, Ramnath Balasubramanyan, Bryan
R. Routledge, Noah A. Smith, From Tweets to
Polls Linking Text Sentiment to Public Opinion
Time Series, Proceedings of the International
AAAI Conference on Weblogs and Social Media,
Washington, DC, May 2010.
9
Surface methods plus categories
Note that Noah Smith did point out this
ambiguity!
10
Surface methods plus hidden structure
http//www.statmt.org/moses/?nMoses.Background
11
One morning I shot an elephant in my
pajamas. How he got in my pajamas, I dont know.
One morning I shot an elephant in my pajamas.
One morning I shot an elephant in my pajamas.
12
Extracting structured information
HPI Atrial fibrillation. This patient is a
56-year-old white gentleman who has had a history
of atrial fib on and off since he had his bypass
surgery. Patient was originally diagnosed with
coronary artery disease as well as mitral valve
problems approximately 3 years ago. Dr. Tirona
used to take care of him at that time. He had a
bypass surgery as well as mitral valve repair
done at that time. Postop he had an episode of
A-fib which then resolved spontaneously. He
remembers somebody talking to him about
cardioversion, but then the A-fib resolved
spontaneously. So he was started on Coumadin. He
would get some occasional episodes, but usually
they are very brief, so he never bothered about
them. Of late, over the last few months, he has
been getting more frequent episodes and duration
of these episodes is also prolonged for a few
hours. So he saw Dr. Hagan who has referred him
here for further evaluation and treatment. The
patient states when he does get the A-fib, he
feels very weak, tired, and short of breath. He
denies any chest pain. Otherwise he is usually
very active physically, he works fulltime as an
electrician, and has not had any problems as far
as doing his day-to-day work. MEDICAL HISTORY
1. Coronary artery disease as mentioned above. 2.
Hypertension. 3. Hypercholesterolemia. . IMPRESSI
ON Paroxysmal atrial fibrillation in a patient
with prior mitral valve disease, currently having
more frequent breakthroughs symptoms.
13
Extracting structured information
HPI Atrial fibrillation. This patient is a
56-year-old white gentleman who has had a history
of atrial fib on and off since he had his bypass
surgery. Patient was originally diagnosed with
coronary artery disease as well as mitral valve
problems approximately 3 years ago. Dr. Tirona
used to take care of him at that time. He had a
bypass surgery as well as mitral valve repair
done at that time. Postop he had an episode of
A-fib which then resolved spontaneously. He
remembers somebody talking to him about
cardioversion, but then the A-fib resolved
spontaneously. So he was started on Coumadin. He
would get some occasional episodes, but usually
they are very brief, so he never bothered about
them. Of late, over the last few months, he has
been getting more frequent episodes and duration
of these episodes is also prolonged for a few
hours. So he saw Dr. Hagan who has referred him
here for further evaluation and treatment. The
patient states when he does get the A-fib, he
feels very weak, tired, and short of breath. He
denies any chest pain. Otherwise he is usually
very active physically, he works fulltime as an
electrician, and has not had any problems as far
as doing his day-to-day work. MEDICAL HISTORY
1. Coronary artery disease as mentioned above. 2.
Hypertension. 3. Hypercholesterolemia. . IMPRESSI
ON Paroxysmal atrial fibrillation in a patient
with prior mitral valve disease, currently having
more frequent breakthroughs symptoms.
14
Extracting structured information
ICD
15
(No Transcript)
16
(No Transcript)
17
Dr. Martin S. Kohn Clinical Decision Support
DeepQA, http//www.im.org/Meetings/Past/2012/2012A
PMWinterMeeting/Documents/Winter20Meeting20Prese
ntations/PS20II_Kohn.pdf
18
Discovering structure
Looking at just word counts often gives you a
mish-mash.
19
Discovering structure
Bayesian topic models discover the distinct
topics interwoven in documents.
Wikipedia Topic Model Blei et al. 2003
20
Model detecting topic shifts
21
Model detecting topic shifts
22
Model topic shift tendency
23
Ifill, moderator Terrible. Yes, she was
constrained by the agreed debate rules. But she
gave not the slightest sign of chafing against
them or looking for ways to follow up the many
unanswered questions or self-contradictory
answers. This was the big news of the evening.
Katie Couric, and for that matter Jim Lehrer,
have never looked so good.
24
Model topic shift tendency
25
Take-aways

• The role of NLP is not understanding. Its
  helping people do useful things with language.
• Shallow methods work extremely well except when
  they dont. Language is replete with underlying
  structure.
• The deep value to look for in NLP is in bringing
  that structure to the surface and making it
  accessible to human insight.

26
Thanks!
27
reactlabs.org
28
(No Transcript)
29
(No Transcript)
30
Four years ago, I know that many Americans felt a
fresh excitement about the possibilities of a new
president. That choice was not the choice of our
party, but Americans always come together after
elections. We're a good and generous people, and
we are united by so much more than what divides
us. When that election was over, when the yard
signs came down and the television commercials
finally came off the air, Americans were eager to
go back to work, to live our lives the way
Americans always have, optimistic and positive
and confident in the future. That very optimism
is uniquely American. It's what brought us to
America. We're a nation of immigrants, we're the
children and grandchildren and great-grandchildren
of the ones who wanted a better life. The
driven ones. The ones who woke up at night,
hearing that voice telling them that life in a
place called America could be better. They came,
not just in pursuit of the riches of this world,
but for the richness of this life.
31
Take-aways

• The role of NLP is not understanding. Its
  helping people do useful things with language.
• Shallow methods work extremely well except when
  they dont. Language is replete with underlying
  structure.
• The deep value to look for in NLP is in bringing
  that structure to the surface and making it
  accessible to human insight.

32
Thanks!
33
(No Transcript)
34
(No Transcript)
35
"In this 10-year time frame, . . . we will have
perfected speech recognition and speech output
well enough that those will become a standard
part of the interface."
Bill Gates, 1997