Nowadays the technology is still not ready to give a machine the five senses with the recognition level close to the one we use every day and probably will not be ready for a few more decades. But Artificial Intelligence researchers are interested in machine learning based on these senses already today. The cognitive scientists usually develop their theories basing on limited input because gathering the data on what we feel when we touch a stone, smell a perfume, taste an apple, see a thief or hear an explosion would be a very laborious task. Not only inputting such data manually but also gathering those experiences by a robot would be laborious, costly, sometimes not possible at all. In our research we try to make a machine gather the basic knowledge only from the textual level of WWW to make learning faster and less laborious as a domain is open and retrieval of knowledge is automatic. In this paper we will propose a simple method to simulate input from the robot¡Çs sensors when only one is really used. By using this method a robot which recognizes an object with one of 5 sensors automatically expects what kind of input would be given by remaining 4 sensors even if they do not physically exist (for example, if the robot using a camera recognizes an object as a stone it automatically should know that normally it has no taste, does not make sound, does not smell and has no taste).