Spoken Dialogue Technology

1
Spoken Dialogue Technology

• How can Jerry Springer contribute to Computer
  Science Research Projects?

2
Objectives

• Review the article, Spoken Dialogue Technology
  Enabling the Conversational User Interface by
  Michael F. McTear
• Introduce Spoken Dialogue Technology
• Discuss potential areas of research in this field

3
What is spoken dialogue technology?
Spoken dialogue technology allows users to
interact with computer-based applications such as
databases and expert systems by using natural
spoken language.
4
What problem does spoken dialogue technology
solve?

• Makes human-computer interaction more natural for
  human beings
• Why do people get carpal tunnel syndrome?
• Reaches differently-abled end users

5
How do we enable conversational user interfaces?

• McTears paper addresses this question by
• Introducing spoken dialogue systems
• Classifying these systems
• Explaining the required components
• Referencing current development trends
• Reviewing development toolkits
• Suggesting future research possibilities

6
How are spoken dialogue systems classified?

• Spoken dialogue systems are classified
  according to the systems dialogue control
  method.
• Finite State-based
• Frame-based
• Agent-based

7
Finite State-based systems

• A dialogue consists of a predetermined sequence
  of steps.
• User input is highly constrained, consisting of
  single words or phrases
• The simplicity of such systems accounts for both
  the advantages and disadvantages of this approach
• Example Nuance Demo Banking System

8
Frame-based Systems

• These systems gather all of the information
  necessary to complete a task
• The dialogue flow depends on the users input and
  the systems required data as specified in a
  template
• Example Philips Train Timetable

9
Agent-based Systems

• These systems permit complex communication
  between the system, the user, and underlying
  applications.
• Mixed-initiative systems communication is viewed
  as interaction between two agents.
• Example Circuit-Fix-It Shop System

10
Required Components

• Speech Recognition
• Language Understanding
• Dialogue Management
• Communication w/ External Systems
• Response Generation
• Speech Output

11
Current Development Trends

• Analyze Human-Human dialogues
• Design based on Simulations
• Usability Analysis
• Guidelines and Standards
• Evaluation Metrics

12
Development Toolkits

• Many toolkits exist to develop spoken dialogue
  systems
• Examples
• CSLU toolkit Center for Spoken Language
  Understanding
• SpeechMania
• More examples
• ASDA Anthropomorphic Spoken Dialogue Agent
• Psyclone

13
Future Research Possibilities

• McTear lists many possible areas for research
• Incorporation of a prosody analyzer
• Multimodal dialogue systems
• Perhaps more work on Theory
• Classification
• Architecture and Components

14
Prosody Analyzer

• Why do most ancient epic poems conform to strict
  rhyme, rhythm, and metrical forms?
• Rhyme and rhythm were used to encode and
  transmit knowledge.
• Example from Cheers Albania!

15
Prosody Analyzer

• Researchers refer to prosody as a separate
  channel of communication distinct from the
  lexical channel.
• If we could capture the information communicated
  on this channel, our spoken dialogue systems
  would be more robust.

16
Prosody Analyzer Research

• Bell Labs Tutorial http//prosodies.org/tutorial
  2002
• Prosody Analysis for Speaker Affect Determination
• Researchers concluded that the best source of
  prosody data was daytime talk shows like Jerry
  Springer!!!

17
Multimodal Dialogue Systems