We present the design of a dialog manager that performs speech intent recognition, based on a finite state machine which enables simultaneous processing of multiple sub-modules and maintains ordered transitions of system states. The dialog manager is integrated into a spoken dialog system. The application area of this system is targeted on robots, and the core problem that we address is to recognize user's speech intents, which could be either asking questions or giving commands to a robot. Our dialog manager is a sequence classifier based on hidden Markov models, and it uses part-of-speech tags as output symbols. The classifier take sentences of variable lengths as input. It is trained on a small data set and achieves and accuracy of 83%.