Language Technology for Customer Relations

1
Language Technology for Customer Relations

• John Nerbonne
• Informatiekunde, Groningen
• Computer-Mediated Communication
• Consumenten Contacten in 2005
• BSC Seminar, Amsterdam
• Oct. 12, 2000

2
Language Technology

• What is language technology?
• What are applications of LT?
• Will voices replace screens?
• How to get it right
• Opportunities for consumer relations

3
Well-known LT applications

• Spell checkers
• right/wrong, nearest match, variant(s)
• Rough translation tools (AltaVista)
• Postbanks stock quotations (telephone)
• Smart search engines
• seek Kennedys daughter
• find the daughter of J.F.Kennedy, Kennedys
  children, etc.

4
Language Technology Tasks

• recognize, analyze words, phrases
• index, search, sort, retrieve, store texts
• find terminology, person/place names
• align translations, correspondences
• organize documentation for maintenance, versions,
  multilingualism

5
Potential Applications
6
Problems with LT

• Language ambiguous, volatile, sequential
• Dont stop! vs. Dont! Stop!
• Wreck a Nice Beach vs. Recognize Speech
• Spoken language quickly fades, is forgotten
• Long lists OK to scan visually, not to hear
• LT is young
• OVIS 83 of conversations successful

7
Invest in airlines, or, Why face-to-face wont
go away

• High bandwidth
• Social (vs. information) factors
• Shared experience, common space
• Inimitable presence of the body
• Bonding
• Showing commitment

8
Thinking about LT applications
Do s
Dont s

• Compare LT vs. human
• Apply where miss is disastrous
• Avoid
• one-time lookups
• general intelligence
• unrestricted language (Annual Reports,
  newspapers, patents)

• Cost/benefit analysis
• Apply w. back-up
• Seek domain
• repeated info. needs
• simple logic
• limited linguistics

9
Near-term Opportunities

• away from PC
• mobile phone, pay phone, SMS
• complex PC navigation
• users wont tolerate menu after menu,..
• hands-busy situations
• driving, examining, factory floor
• assistance to handicapped

10
Emerging Topics

• Flexible delivery
• speech, SMS, or full-screen
• via XML
• Support for human communication

11
Getting it Right

• New technologies not easy
• Bar-code readers (15 year introduction)
• Video recorders (1 competent user/family)
• Automatic tellers (banks) (90)
• Stoves, washers, dryers, dishwashers, answering
  machines,...
• Suchmans study for Xerox
• the green button

12
Myths of Interface Design

• Interfaces should allow max. functionality
  -- anything goes
• text editors that allow letters
• Wysiwyg is (always) superior
• What you see is what you get
• Problem documents, graphics for diff. media?
• Do what I mean, not what I say.
• example problem overeager spell-checkers

line
off
13
Glosser

• help with French
• endings (grammar)
• dictionary access
• other examples
• word pronunciation
• web version
• www.let.rug.nl/alfa/
• projects

14
Early Glosser Interface

• General mouse control
• Users (tried to) look up word pieces
• Solution make mouse sensitive to words
• First encouraged overuse
• Some words looked up several times
• Solution remind users
• Users took notes!
• Missed margins to write in
• Solution allow gloss between lines

15
Early NLP Interfaces (pre-OVIS)

• Competition with graphics
• Windows Excel vs. NLP
• Solution focus on other delivery (phone)
• Based on grammar-book language
• Whens the train to Zwolle, ah Meppen?
• Solution base grammar on recorded interaction

16
Relevant Developments

• Informatiekunde, RuG
• LT, Web technology
• Computer-Mediated Communication
• cooperative program IK, Communicatie- en
  Informatiewetenschappen, RuG
• 6-month work period in study

17
LT for Customer Contact

• Contact needs automation
• LT can support applications now
• modest, repetitive, frequently needed
• Repeat until right
• Design, implement evaluate in use