We examine the problem of "understanding robots' and design an F–2 emotional robot to "understand" speech and to support human-like behavior. The suggested system is an applied implementation of the theoretical concept of robotic information flow, suggested by M. Minsky ("proto-specialists") and A. Sloman (CogAff). This system works with real world input – natural texts, speech sound – and produces natural behavioral output – speech, gestures and facial expressions. Unlike other chatbots, the system relies on semantic representation and operates with a set of d-scripts (equivalents to proto-specialists), extracted from advertising and mass media texts as a classification of basic emotional patterns. The process of "understanding" is modelled as the selection of a relevant d-script for the incoming utterance.