Experience Is a Double‐edged Sword: a Computational Model of the Encoding/retrieval Trade‐off with Familiarity

s Photographs Words Stimulus type Midazolam Saline Fig. 8. Recognition memory measured in d 0 as a function of stimulus type (words, photographs, and abstract pictures) and drug condition (saline vs midazolam). From Reder et al., 2006a, p. 565. Copyright 2006 by the Association for Psychological Science. Reprinted with permission of the author. 302 Lynne M. Reder et al. participant, nor the nurse, nor the experimenter knew which day a particular subject was given saline or midazolam, (i.e., testing conditions were double blind). The striking result was that midazolam aVected recognition memory for words most and aVected memory for abstract pictures least (Fi g. 8 ). Our explanation for this resul t is that (a) midazolam only aV ects t he ability to c reate new bindings (Park, Q uinlan, Thornton, & Reder, 2004 ; Reder et al., 2006b) and (b) only a unitized chunk can be bound to an experimental context. The abstract pictures could not be bound to the experimental context even i n t he saline condition, and therefore the eV ect of the dr ug was minimized for t hat s timulus class. A nother finding by Dobbins and K roll (2005) can be interpreted as sup porting our hypothesis. They found t hat r ecognition m emory w as superior for s cenes and faces that were known, but that the advantage for those stimulus t ypes was eliminated when subjects w ere f orced to respond quickly or when testing was delayed for one week. Our interpretation is th at binding concepts to experimental context i s m uch m ore likely for know n f aces and s cenes; however, i f responding must be rapid, judgments are based on familiarity and so there is no advantage to having formed an episode node. With a one week delay the episode node and l ink will have decayed s ubstant ially m aking reliance on familiarity t he dominant process. Experience Is a Double‐Edged Sword 303 The notion that unitization requires prior experience is not a new idea. Hayes‐Roth (1977) and Servan‐Schreiber (1991) have hypothesized something similar; however, no one has thus far suggested that the strength of a chunk predicts the probability of encoding it and binding it to other chunks. Our explanation is that an item with no prior representation must be encoded in terms of the component features that are strongly activated. With repeated exposure, the node that binds the constituents together becomes a chunk in its own right, forming a new, higher‐level chunk involving the grouping of these features. At that point the higher‐level chunk is suYciently strong (i.e., has strong enough base‐level activation) to be bound with other co‐occuring stimuli or bound to the experimental context to make an episodic event. The abstract pictures had not been experienced before and recognition could only be based on the familiarity of the elements that were primed from exposure. Further support for the notion that chunks are constructed as their constituent elements become more familiar comes from studies with chess masters (Chase & Simon, 1973; Simon & Gilmartin, 1973) who have acquired thousands of hours of experience with various chess patterns. Although chess masters are much better than novices at reproducing a chessboard configuration when it was displayed tachistoscopically (very briefly), they are not better than a novice if the configuration of chess pieces on the board is random (de Groot, 1965). In addition, the latency between chess pieces that were put down on the board to reproduce the flashed display mirrored the chunks that one would expect. That is, subjects had shorter pauses when putting down pieces within a chunk (e.g., a Sicilian defense), but longer pauses when switching to recall of another chunk.