Retrieval and Mapping.

Visual information can be used to retrieve memories. Since visual cueing can occur, it is reasonable to think that some memories are encoded in terms of perceptual information. This is so well accepted that there is a theory that all memories are encoded in terms of perceptual information (Barsalou, 1999).

In analogical problem solving, a visual representation of the target can be used to retrieve visual representations of potential sources. An analog can be represented in terms of process, causality, uses, etc.; the visual representation is one of many. The mind will use these connected, different representations for retrieval when relevant.

Visual representations can be generated to aid in retrieval. For example, if many people are making demands of you, the sensory experience is multi-modal and complicated. However, in trying to understand the situation you might generate a visual abstraction to represent it. For example, the demands might be represented as converging red lines on a circle, representing you, perhaps triggering other memories regarding convergence.

Mapping two analogs involves aligning their elements. Since retrieval is based on matching, to some extent, it is reasonable to suppose that the processing done to retrieve an analog can be used to guide mapping (Holyoak & Thagard, 1997; Falkenhainer et al., 1990).

Since I am dealing with problem solving, and the analogical transfer of problem solving solutions, retrieval queries are based on the initial problem state and the solution constraints. See Figure 4.

Figure 4: This figure shows a target analog problem and how it relates to the potential analogs in the case memory. The items in the case memory are represented here as a series of knowledge states (represented by boxes) connected with manipulations, which are changes to the knowledge state (represented by straight arrows pointing right.) The last box in a series is the solution state, and the arrows, in order, represent the solution procedure. The target only has a single knowledge state because there is no solution procedure yet. There is also no solution state, but rather a set of criteria that must be fulfilled. The initial target knowledge state is compared to the initial states of the cases in memory for similarity (shown as wiggly arrows coming from the target knowledge state). Also, the solution criteria is compared to the solution states in memory to see if they fulfill the criteria (shown as wiggly arrows coming from the target solution criteria). Based on these measures cases are retrieved.

Mapping visual analogs may differ from mapping of non-visual analogs in two ways. First, visual analogy can use visual knowledge, such as an abstraction hierarchy (e.g. a square is a kind of rectangle). Such knowledge can be used to find matches based on similarity. Second, the agent's perceptual system can be brought to bear to inform the mapping. Seeing a truck may have something to do with matching its parts to the parts of some episode of truck experience in memory. Likewise, one might help guide mapping by using perception on generated mental images (Kosslyn 1994).