An Aggregation Procedure for Building Episodic Memory

When dealing with narrative texts, a system must possess a strong domain theory, and especially knowledge about situations occurring in the world. Otherwise the system must envisage comprehension as a complex process including learning from the texts themselves to improve its capabilities. This requires managing past solutions and completing them when analoguous situations happen in other texts in order to create general situations. We propose a system, MLK (Memorization for Learning Knowledge), that organizes specific situations in an episodic memory by aggregating the similar ones in a single unit. This aggregation process leads to a progressive enrichment and generalization of the overall situations and of their specific features. MLK is a system conceived to allow the emergence of structures, their accessing being realized by a propagation process. Therefore, with MLK, we are able to address the problem of understanding and learning even when a domain theory is lacking.