Abstract: In this paper, we present the acquisition of synthetic dialog corpora through a dialog system that integrates a stochastic dialog manager and a rule-oriented user simulator. These modules are task-independent, and can be adapted to different semantic-restricted domains. Our stochastic dialog manager can interact with real or simulated users, storing automatically the acquired dialogs. In addition, the simulation mode allows us to acquire series of dialogs, verifying automatically their successful endings. These dialogs are used to adapt the stochastic dialog models and, therefore, to enhance the system in new acquisitions. This methodology has been applied to develop two dialog systems in different domains: a train services information system, and a sport booking system.
Index Terms: stochastic dialog management, user simulation, task independence, synthetic acquisition.