A Modular Architecture for Adaptive ChatBots

We illustrate an architecture for a conversationalagent based on a modular knowledge representation. This solutionprovides intelligent conversational agents with a dynamicand flexible behavior. The modularity of the architecture allowsa concurrent and synergic use of different techniques, makingit possible to use the most adequate methodology for themanagement of a specific characteristic of the domain, of thedialogue, or of the user behavior. We show the implementation ofa proof-of-concept prototype: a set of modules exploiting differentknowledge representation techniques and capable to differentlymanage conversation features has been developed. Each moduleis automatically triggered through a component, named corpuscallosum, whose task is to choose, time by time, the most adequatechatbot knowledge section to activate.