140 likes | 294 Vues
This study explores Augmented Transition Networks (ATNs) for dialogue control through a comprehensive longitudinal analysis. We examine the architecture, applications, experiments, and results of ATNs, highlighting their versatility in training scenarios such as bank teller training, law enforcement encounters, and medical profilers. The ATN framework serves as a finite state machine incorporating various nodes and transitions to streamline dialogue management. Our findings indicate that employing variables significantly reduces network complexity while maintaining performance, paving the way for future research in efficient dialogue systems.
E N D
AUGMENTED TRANSITION NETWORKS (ATNs) FOR DIALOGUE CONTROL: A LONGITUDINAL STUDY Curry Guinn Rob Hubal (RTI)
Outline • Architecture • Applications • Experiments • Results
Applications • Question-Answer Kiosks (Tradeshow for Tradeshows, Space Congress, John Deere, ASTD, APHA) • Training Applications • Bank Teller Training • Field Interviewer Trainer • Phone Interviewer Trainer • Informed Consent Trainer • Law Enforcement Training for Encounters with the Mentally Ill • Pediatrics Trainer • Profiler • Youth risk-assessment profiler • Prison inmate risk assessment profiler
Augmented Transition Network Data Structure • Augmented Transition Network is a Finite State Machine with Variables • Nodes • Transitions • Variables If x < 2 A B If x >=2 C
Augmented Transition Network Data Structure wait_on_input /* Name */ Normal /* Type */ Grammar: "intro.gram default.gram" Response: "inform(offer_assistance)" VRString: "raise(eyebrows)" Action: "MENTOR = 0.8" Transition: proc_command TransitionAction: "INPUT = command" Conditional: "command(CONTENT)" Transition: proc_query TransitionAction: "INPUT = query" Conditional: "query(CONTENT)"
Variables Used • The linear increase is encouraging • Why? • You might expect that as the number of semantics increase the network expands exponentially • What’s holding this number down? • The use of variables
Decrease in the Number of Normal Nodes • The replication of normal (“speech”) nodes resulted in more repetitive networks • Used variables to store which grammars needed to be loaded next • Greatly decreases the complexity of the network
The number of transitions as a measure of complexity • Over time, we tended to move towards reducing the inter-node connections • Why? • Easier to trace and debug • Easier to maintain • This at the expense of large number of nodes
Future Work • How do these empirical measurements correspond to “true” costs? • Time to development • Reliability • Maintenance