Storage is the task responsible for placing new concepts into the existing memory structure of a reasoner. Concepts are linked into the existing structures, based on the attributes contained in the concept. The goal of this task is to place related concepts together. However, since the function possessed by a concept may change, this measurement of relatedness is only applicable with respect to the function a concept holds at the time of storage.
The retrieval task handles the retrieval of ``proper'' concepts for the reasoner's memory. A cue is given to the memory system, consisting of a number of slots and fillers that should exist in a retrieved concept. Those nodes in the memory structure are activated. Concepts linked to those are then activated. If a concept achieves a certain level of excitation, it is returned as a potential retrieval success. Further retrievals may then be made with the same set of cues. In addition to the cue set, a context may also be specified. If this is the case, the concepts making up the context will also be activated, in the same fashion as the cue set. This allows memory retrievals to be influenced by other concepts being used by the reasoner. This use of context is also the way in which my theory and model makes use of what would be called working memory in other theories. Rather than having an area of memory which is constrained to hold a maximum set of concepts which are then always easily retrievable by the reasoner (the traditional view of working memory), the memory supertask suggests that the area of memory which is most highly activated will contain the concepts which make up the working memory of the reasoner. As there is a finite amount of activation to be allocated to the memory system, only a small portion of the memory will be primed for easy retrieval.
For a complete theory, retrieval and storage are not the only memory operations which are needed; I also need to model reminding. In my research, reminding refers to the spontaneous retrieval of concepts. It occurs as a result of two features of the memory supertask. First, the memory runs asynchronously with the rest of the supertasks. A supertask can request a memory retrieval. Once a ``close'' concept is discovered, the memory supertask returns this to the supertask which has requested it ([#!memory:ram-francis-1996!#]). If resources remain at this point (e.g., there are no other outstanding memory requests), then memory will continue to process the requests which it has already filled which have not been explicitly canceled. It is, therefore, possible for a ``closer'' concept to be discovered later. At this point, reminding can be said to have occurred. memory informs the control supertask that such a reminding has occurred; control must then decide whether to act on this new information.
The functionality which is required by the memory supertask is largely provided by the MOORE memory system ([#!memory:ram-francis-1996!#]). This model of asynchronous memory retrieval fits well into the theory I developed and was able to be used for this supertask. Currently, storage is an automatic process whenever a memory node is created. At creation time, the appropriate attribute nodes are linked into, thereby clustering similar concepts together. Retrieval is an explicit operation, implemented via spreading activation. Reminding is a natural result of the spreading activation combined with the asynchronous nature of the memory system.