Turn Taking for Artificial Conversational Agents

1
Turn Taking for Artificial Conversational Agents

• Fredrik Kronlid
• Department of Linguistics and GSLT
• Göteborg University

2
Agenda

• A turn manager
• The goals of the Synergy research group
• A description of the platform we are building
• Where my contribution fits

3
Turn taking for artificial conversational agents

• Turn management is needed when more than one
  conversational agent is present
• Implementation of Sacks, Schegloff and Jefferson
  (1974)
• Using SCXML (Harel Statecharts)
• Each (artificial agent) has its own turn manager

4
Outside Chart
5
Inside Chart
6
TRP Chart
7
Synergy goals

• Exploring similarities and differences between
• dialogue systems
• interactive agents
• chat
• games
• Building a platform (the Synergy platform)
• Creating and exploring hybrids

8
Synergy
9
Agent communities

• Task-specific agents
• one agent per task
• one agent per organisation
• Human and Artificial agents in the same
  environment
• Share information
• use information already provided to/by other
  agents

10
Synergy architecture
11
Synergy

• Oz web/agent server
• Browser code in HTML/Javascript
• Oz/SCXML/VXML Agents
• Try Synergy SCXML at http//www.ling.gu.se/lager/
  Labs/SCXML-Lab/

12
SCXML Web Lab
13
SSJ Turn Management in Synergy

• Human and artificial conversational agents
• Natural language as the agent communication
  language
• Compatible turn management protocols

14
Results

• Agents are avoiding collisions
• Agents produce false starts
• Agents take turn
• Agents are not aware of what has been said by
  others and never give up trying to answer a
  question

15
Conclusions

• SCXML supports a highly modular design
• SCXML is useful for describing concurrent
  processes
• SSJ is useful for agent turn management
• Better dialogue management is needed

16
What next

• Implementing an Information State Update Dialogue
  Manager (Larsson 2002) (based on Ginzburgs
  Dialogue Gameboard) in SCXML
• Extend DM with multilogue capabilities
  (Ginzburg Fernandez 2005)

17
Thank you for your attention!