QUESTION AND ANSWERING

1
QUESTION AND ANSWERING
2
Overview

• What is Question Answering?
• Why use it?
• How does it work?
• Problems
• Examples
• Future

3
What is it?

• Definition of Question Answering
• Examples
• AskJeeves is probably most well known example
• AnswerBus is an open-domain question answering
  system
• Ionaut, EasyAsk, AnswerLogic, AnswerFriend,
  Start, LCC, Quasm, Mulder, Webclopedia, etc.

4
Why use it?

• From AskJeeves Search engines do not speak your
  language. They make you speak their language a
  language that's strange, confusing, and includes
  words that no one is entirely sure of their
  meaning.
• QA engines attempt to let you ask your question
  the way you'd normally ask it .
• Inexperienced users
• DocumentAnswer?

5
How does it work?

• Natural Language Processing
• Semantic Processing
• Syntactic Processing
• Parsing
• Knowledge Base
• Answer Processing

6
Natural Language Processing (NLP)

• Engines have unique processes
• START-Natural Language System
• Parsing
• Natural Language Annotation
• Processing Component

7
Answer Processing
QA System Output
AnswerBus Sentences
AskJeeves Documents
IONAUT Passages
LCC Sentences
Mulder Extracted answers
QuASM Document blocks
START Mixture
Webclopedia Sentences
8
AskJeeves

• Has own knowledge base and uses partners to
  answer questions
• Catalogues previous questions
• Answer processing engine
• Question template response

9
AnswerBus
10
Problems

• How and Why questions
• What questions
• What happened?
• What did we do?
• Answer Quality
• Correct??
• Answer Presentation

11
Correct? (From Webclopedia)

• Question Where do lobsters like to live?Answer
  on a Canadian airline
• Question Where do hyenas live?Answer in Saudi
  ArabiaAnswer in the back of pick-up trucks
• Question Where are zebras most likely
  found?Answer near dumpsAnswer in the
  dictionary
• Question Why can't ostriches fly?Answer
  Because of American economic sanctions
• Collected by Ulf Hermjakob --November 29, 2001

12
(TREC) -- Text Retrieval Conference

• Yearly information retrieval competition
• Began in 1992 QA in 1999
• In order to encourage research into systems that
  return answers rather than document lists.
• Qs are open domain, closed class
• As are less than 50 chars and entities or noun
  phrases

13
(TREC) -- Text Retrieval Conference

• 500 Questions in 2001
• Some answers nil large difficulty
• Lots of definition questions
• QA list tasks
• Name 4 cities that have a Shubert theater.
• QA context tasks
• How many species of spiders are there?
• How many are poisonous to humans?
• What percentage of spider bites in the US are
  fatal?

14
Example Questions and Results

• What river in the US is known as the Big Muddy?
• AskJeeves
• AnswerBus
• Google

15
Example Questions and Results

• What persons head is on a dime?
• AskJeeves
• AnswerBus
• AltaVista

16
Example Questions and Results

• Show some paintings by Claude Monet
• START

17
Looking Ahead

• User Demand
• Enormous Interest in Problem
• Successes

18
Conclusion

• Question and Answering and Search Engines
• Why its used
• Future
• Moores Law for QA???

19
Sources

• AskMSR Question Answering Using the Worldwide
  Web
• Michele Banko, Eric Brill, Susan Dumais, Jimmy
  Lin
• http//www.ai.mit.edu/people/jimmylin/publications
  /Banko-etal-AAAI02.pdf
• In Proceedings of 2002 AAAI SYMPOSIUM on Mining
  Answers from Text and Knowledge Bases, March
  2002 
• Web Question Answering Is More Always Better?
• Susan Dumais, Michele Banko, Eric Brill, Jimmy
  Lin, Andrew Ng
• http//research.microsoft.com/sdumais/SIGIR2002-Q
  A-Submit-Conf.pdf

20
Sources

• AnswerBus 
• www.answerbus.com
• http//misshoover.si.umich.edu/zzheng/qa-new/
• http//www2002.org/CDROM/poster/203/
• AskJeeves
• http//www.ask.co.uk/docs/about/what_is.asp
• Webclopedia
• http//trec.nist.gov/pubs/trec9/papers/webclopedia
  .pdf
• http//www.isi.edu/natural-language/projects/webcl
  opedia/
• Start
• http//www.ai.mit.edu/projects/infolab/ailab.html
•  Text Retrieval Conference
• http//trec.nist.gov/presentations/TREC10/qa/

21
(No Transcript)