Natural Language Processing

1
Natural Language Processing
2
Overview

• Next Week Jack Tompkins
• Midterm Paper Due (Oct 24)
• Today
• A brief intro to natural language processing
• How does one become a specialist in .?
• Some demos

3
What is Natural Language Processing?

• The study of human languages and how they can be
  represented computationally and analyzed and
  generated algorithmically

4
Natural Language Processing

• Speech recognition
• Natural language understanding
• Computational linguistics
• Psycholinguistics
• Information extraction
• Information retrieval
• Natural language generation
• Speech synthesis

5
Applications you have encountered

• Machine translation
• Spelling/grammar correction
• Information Retrieval
• Data mining
• Document classification
• Question answering, conversational agents

6
Basic Process of NLU
Spoken input
For speechunderstanding
Phonological / morphological analyser
Phonological morphological rules
Sequence of words
He loves Mary.
SYNTACTIC COMPONENT
Grammatical Knowledge
Indicating relns (e.g., mod) between words
Syntactic structure (parse tree)
He
Mary
loves
Thematic Roles
SEMANTIC INTERPRETER
Semantic rules, Lexical semantics
Selectionalrestrictions
Logical form
? x loves(x, Mary)
CONTEXTUAL REASONER
Pragmatic World Knowledge
loves(John, Mary)
Meaning Representation
7
Some Reasons Why NLP is Hard Lexical Ambiguity

• Lexical ambiguity word has more than one
  meaning.
• Hot

8
Lexical Ambiguity

• Hot
• hot can mean warm or spicy or electrified or
  radioactive or vehement or sexy or popular or
  stolen.
• back is an adverb in go back, an adjective in
  back door, a noun in the back of the room and
  a verb in back up your files.

9
Some Reasons Why NLP is Hard Syntactic Ambiguity

• Syntactic ambiguity more than one parse for a
  sentence.
• I saw a boy on the hill with a telescope.
• Leads to Semantic ambiguity.
• Each parse could have a different meaning.

10
Real News Headlines

• Hospitals are Sued by 7 Foot Doctors
• Astronaut Takes Blame for Gas in Spacecraft
• New Study of Obesity Looks for Larger Test Group
• Chef Throws His Heart into Helping Feed Needy
• Include your Children when Baking Cookies
• Kids Make Nutritious Snacks
• Teacher Strikes Idle Kids
• Killer Sentenced to Die for Second Time in 10
  Years

11
Some Reasons Why NLP is Hard Pragmatic Processing

• Context changes the meaning of a sentence
• Yes.
• He is looking at the bank.

12
How to become a specialist in . whatever
13
You want to major in Philosophy ???????

• Philosophy Computer Science?

14
Graduate Schools Often Pay You To Be a Student

• Full-time Ph.D. students are awarded tuition and
  fees as well as stipend support during their time
  of study in the form of fellowships, teaching
  assistantships, and research assistantships.
  Duke
• During the academic year, most of our students
  are supported by assistantships and fellowships.
  Applicants for assistantships are automatically
  considered for all available fellowships. The
  stipend for research and teaching assistantships
  for the current nine-month academic year
  2004-2005 is 14,500 (20 hours a week).
  UNC-Chapel Hill
• We would like to offer you a Graduate Research
  Assistantship (GRA) in the _______ of the
  University of ______ starting August 18, 2007 or
  earlier if possible. The GRA carries a stipend
  paid as follows 1425 monthly during the
  semester, and 2850 monthly during the summer,
  for a total of 21,375 per calendar year. In a
  letter to one of our undergraduates who went on
  to graduate school elsewhere

15
RTI

• 1995-2004
• http//www.rti.org/

16
My research interests

• Spoken language interfaces
• Expert systems
• Virtual reality
• Simulations
• Some recent projects
• Translation
• Spoken diary understanding
• Emotion in text and facial expression

17
Translation

• Ralph Harris

Babelfish 64 Google 56 Turbocharge 78
18
Location and Activity Diaries

• Dave Crist, Haley Werth, Dan Reeves
• Necessary data from study
• Date/Time
• Location
• Activity
• Activity and location representation CHAD
• Consolidated Human Activity Database
• Designed by EPA

19
Database Sample
20
How much does context help?
21
Most recent work

• Gulustan Dogan
• Len Lecci, Psychology
• Their study was to induce emotions.
• 97 positive examples, 97 negative
• Can the computer read and learn from examples to
  classify whether an entry is positive or
  negative?

22
Positive or Negative?

• Hearing this music reminds me of my years in high
  school. I played flute in our ensemble band. I
  enjoyed playing with that particular band full of
  musically talented individuals. It was a great
  feeling that as a whole, we were able to produce
  such an array of beautiful sounds when put
  together formed wonderful music. It warmed my
  heart and hopefully the hearts of the audience.
  We felt most proud when one of our band members
  conducted a piece of music.

23
Positive or negative?

• Before I left for college me and my girl friend
  had just broken up, and I had lost many of my
  closest friends due to the stand I was taking in
  my Christian faith. Everyone had turned their
  back on me. I felt as if I did not have a friend
  in the world. Everyone I had ever depended on to
  help me out of bad situations in the past were
  now on the other end causing them. All this took
  place because of something I took a strong stand
  for. That whole summer before college was real
  lonely and when it came time to leave I felt as
  if I really

24
How Did the Computer Do?
25
Goal Combine Text Analysis with Facial
Expression Analysis

• Matt Ratliff