Testing instance models of face repetition priming

DENNIS C. HAY ) 0 1 0 I thank Alan Collins, Peter Morris, and Peter Walker of the Lancaster Department of Psychology for their time discussing and commenting on various drafts , and Gordon Logan and John Wixted for insightful re and requests for reprints should be addressed to D.H., Department of Psychology, Fylde College, Lancaster University , Bailrigg, Lancaster, LAI 4YF, England ( 1 Lancaster University , Lancaster, England Two experiments examining repetition priming in face recognition are reported. They employed eight rather than the more usual two presentation trials so that the prediction made by Logan's (1988) instance model of power function speedup of response time (RT) distributions could be examined. In Experiment 1, we presented the same photograph on each trial; in Experiment 2, we presented photographs of varying poses. Both experiments showed repetition priming effects for familiar and unfamiliarfaces, power function speedup for both the mean and the standard deviation of RTand the power function speedup of the quantiles of the RT distributions. We argue that our findings are consistent with the predictions made by the instance model and provide an explanatory challenge for alternative theoretical approaches. - showed that tasks that involve familiarity or occupational decisions are susceptible to the Locus I priming effects whereas tasks involving face naming are susceptible to Locus 2 priming. Ellis et al. (1996) have also argued that the new data they presented were consistent with structural accounts of face repetition priming in the Burton, Bruce, and John ston's (1990) neural network implementation based on the interactive activation and competition networks of Me Clelland and Rumelhart (1981). Burton et al. use the term structural to refer to models that embrace the concept of face recognition units (FRU), which are internal represen tations directly equivalent to the logogens proposed by Morton (1979) to explain how words are recognized. In these accounts, repetition priming occurs when the first encounter with a stimulus lowers the activation threshold of the internal representation so that less stimulus activa tion is required to trigger the representation on a subse quent occasion. Ellis et al. (1996) also examined an alter native theoretical account of repetition priming-namely, the episodic or instance-based account first offered by Ja coby (1983) and Jacoby and Brooks (1984). They ques tioned the recognition unit metaphor and demonstrated how priming effects can be explained entirely in terms of instance retrieval. In addition, they suggested that repeti tion priming results from a process ofperceptual enhance ment, in which the memory of a previous encounter with a stimulus facilitates recognition. Ellis et al. (1996) fo cused on one particular instance-based account, that of Logan (1990), which draws parallels between repetition priming and the development of automaticity in task per formance that follows large amounts of practice. In an at tempt to integrate the explanations of these two phenom ena, Logan (1990) highlighted three parallels: (I) that the response time (RT) decreases that result from repetition priming and the development of automaticity are both power functions of the number of exposures; (2) that both phenomena share item specificity-that is, only prior ex periences that are similar to that which is being processed are retrieved and enhance processing speed; and, (3) that repetition priming and automaticity both share an asso ciative basis. Logan also proposed that repetition priming is dependent on associations between stimuli and re sponses, or interpretations. Ellis et a!. (1996) indicated that their data and those from a number of existing studies cre ated problems for the latter two notions. For example, the findings that the prior reading of a name primed subse quent face naming was inconsistent with Logan's (1990) definition of item specificity. If the written name and the visual appearance of a face have nothing in common, pre sentation of the face for naming should not activate the prior episode of reading the name. Similarly, the view of repetition priming as being dependent on the associations between stimuli and interpretations was contradicted by the Ellis et al. (1990) finding that repetition priming did not occur when subjects were asked to decide on the gen der of a face or to make expression judgments. When a second judgment is made, the prior episode should be acti vated, which in turn should lead to perceptual enhancement. The purpose ofthis article is to rigorously examine the first of Logan's (1990) parallels between automaticity and repetition priming. Perhaps the greatest strength of Logan's instance model is the set of strong predictions that concern the speedup in responses to repeated stimuli. In Logan's (1988) theory, speedup results from a processing shift. Initially,processing is based on a set ofgeneric, non automatic, cognitive procedures (i.e., algorithms) that be come replaced by processing that involves direct mem ory access of past instances. The mechanism by which this shift occurs is simply a race between the algorithmic processing and the direct memory mechanism. On any encounter, whichever finishes the race first generates the response. Initially the algorithm may be more reliable and/or faster, but as the number of instances increases, the race becomes uneven as the algorithm competes against an increasing number of instance competitors. Direct mem ory times speed up as the minimum retrieval time de creases as the number of instances in memory increases. This model makes a number of strong predictions that stem from mathematical simulations of the race between the algorithm and the instances. The first prediction is that performance will speed up with practice and be well fit by a power function of the form RT = a + bti, where RT is the time required to complete the task; a is a constant reflecting the asymptotic performance reached; b is a constant reflecting the difference between the ini tial and asymptotic performance; N is the index of prac tice (i.e., the number of trials); and c is a constant indi cating the rate oflearning. This function has been shown to apply to a wide range of tasks that involve both motor and cognitive learning performance (Newell & Rosen bloom, 1981). The second prediction is that the variability in perfor mance, as measured by the standard deviation of perfor mance over trials, will also decrease with repetition and that this performance is also well fit by a power function. How ever, what is most surprising is that the power functions that describe mean RT performance and the variability in the RT performance as measured by the standard devia tion ofthe RTs are predicted to have equivalent learning rate parameters. This has been proven mathematically, substantiated by simulation, and supported by empirical data (Logan, 1988). The third pre (...truncated)