ABSTRACT
This chapter describes a technical approach to the extension of knowledge representation in a semantic long-term memory (sLTM) by the use of conceptualised episodes of a short-term memory. Based on a semantic net (SN) for knowledge representation in our sLTM, a sequence of user instructions is segmented into episodes reflecting aggregates coded in the SN. When giving an instruction like “and now do the same again but with a 5-hole bar”, need arises for the modification of a conceptualized episode in the eSTM. Using the semantic knowledge about the parts still available in the episode, this task has been realized recently. As the modified episode has its roots in an episode that has been segmented on the basis of aggregate knowledge in the SN of the sLTM, the origin of the modified construction episode is available as a node in this SN. The chapter highlights a method to augment the sLTM with unspecified information found in the user's instructions.
