Draft of: Reingold, E. M & Toth, J. P. (1996). Process dissociations versus task dissociations: A controversy in progress. In G. Underwood (Ed.), Implicit cognition (pp. 159-202). Oxford: Oxford University Press.
Process Dissociations versus Task Dissociations: A Controversy in Progress
Eyal M. Reingold
Jeffrey P. Toth
Much of the long standing controversial status of the study of unconscious processing revolves around the lack of a general consensus as to what constitutes an adequate operational definition of conscious awareness (see Dixon, 1971, 1981; Erdelyi, 1985, 1986; Eriksen, 1960; Holender, 1986; Reingold & Merikle, 1988, 1990). An attempt to review definitional issues relevant to the measurement of awareness is quick to reveal a very curious discrepancy between the prominence of the debate of such issues in the context of the study of perception without awareness, versus the absence of such discussions in the study of unconscious, or implicit memory. Referring to debates concerning criteria for establishing perception without awareness, Schacter (1987) suggested that "memory researchers would do well to attempt to incorporate some of the lessons from these investigations into research on implicit memory" (p. 511). Until recently, discussions of definitional criteria relevant to the measurement of awareness remained scarce despite numerous studies exploring dissociations between implicit/indirect and explicit/direct measures of memory.
Following the introduction of the process dissociation approach (Jacoby, 1991; Jacoby, Toth & Yonelinas, 1993), there has been a surge of interest in issues pertaining to the relationship between memory and awareness. Much of this newly found interest centers around criticisms of various aspects of the process dissociation paradigm. This scrutiny is perhaps to be expected given that the process dissociation approach represents a novel approach both to the measurement of conscious control, and to the study of conscious and unconscious influences on behavior. However, whereas some authors attempt to provide a thorough and balanced review of the assumptions made within the framework of the process dissociation approach (henceforth PDA) and the assumptions made by the task dissociation approach (henceforth TDA) (e.g., Roediger & McDermott, 1993), other critiques of the PDA tend to be rather selective and biased (e.g., Graf & Komatsu, 1994). The reader of such critiques may be led to believe that while the PDA involves particularly problematic assumptions, the implicit memory TDA is somehow assumption free. We believe that this distorted impression is largely attributable to the fact that, while the PDA includes explicitly stated assumptions regarding the relationship between memory and awareness, the assumptions of the TDA are often tacit, unacknowledged and unexamined.
As a case in point, in a recent critique by Graf and Komatsu (1994) readers are forewarned in the title "Process Dissociation Procedure: Handle with Caution!". These authors argue that results obtained by applying the PDA cannot be validly compared to results obtained within the TDA. We dispute this claim and argue that, as with any radically new paradigm, the PDA emerged because of issues which were not adequately handled by the dominant paradigm, in this case the TDA. That is, the PDA has taken on problems that the TDA did not solve. Many of these problems are a function, not of the PDA or TDA, but of assuming a two-process/system, rather than a one-process/system, view of memory. Indeed, some issues (e.g., involuntary conscious memory) which have been raised as problems unique to the PDA, were originally introduced as problems for the TDA (see Schacter, 1987). In addition, the PDA included an explicit assumption regarding the relationship between conscious and unconscious memory processes. The particular assumption adopted in the original implementation of the PDA was one of independence between these processes. This assumption has generated considerable controversy (see Curran & Hintzman, in press; Gardiner & Java, 1993; Jacoby & Begg, submitted; Jacoby, Toth, Yonelinas & Debner, 1994; Jacoby, Yonelinas & Jennings, in press; Joordens & Merikle, 1993). However, what is easily forgotten is that the need for exploring what is the appropriate relational assumption is not the unique responsibility of the PDA, but should be addressed by any dual process/system model of memory, including the TDA. Thus, the fact that there was no explicit treatment of this issue within the TDA does not exempt this framework from dealing with it. Hence, many criticisms of the PDA are not intrinsic to the procedure, but are indicative of the current state of memory research. Discarding the PDA will not eliminate these problems, unless the distinction between conscious and unconscious memory is also discarded. Unlike Graf and Komatsu (1994), we emphasis the need for convergent evidence across paradigms. We conclude that the PDA and TDA can, and should co-exist, and that the comparison of results obtained by applying these paradigms may be very informative and productive.
Accordingly, rather than attempting to provide a comprehensive review of research performed within the framework of the PDA (for such reviews see Jacoby & Kelley, 1991; Jacoby, Ste-Marie, & Toth, 1993; Jacoby, Toth, Lindsay & Debner, 1992; Jacoby et al. in press; Toth, Lindsay & Jacoby, 1992), the present chapter explores three issues as they relate to the PDA and TDA. These issues are response bias, process purity, and involuntary conscious memory. Following the suggestions of Schacter (1987), we begin by providing a brief overview of the response bias and process purity problems as they have been raised in the context of the perception without awareness debate. The relevance of these problems to implicit memory research is then illustrated by considering studies by Graf and Mandler (1984), and Graf, Squire, and Mandler (1984)1. These studies are widely cited as providing some of the most powerful demonstrations of functional dissociations between explicit and implicit measures of memory in amnesic patients, as well as in normal subjects. In addition, these studies illustrate what has come to be known as the retrieval intentionality criterion, and consequently, we consider the relevance of the process purity and response bias issues to this criterion. Next, we describe the PDA and outline the solutions it provides to the response bias and process purity problems. We then consider and respond to recent critiques that have singled out involuntary conscious memory as particularly problematic for the PDA (Graf & Komatsu, 1994; Richardson-Klavehn, Gardiner, & Java, 1994, this volume; Roediger & McDermott, 1993). Finally, we focus on the claim that an empirical comparison across the PDA and TDA is invalid (Graf and Komatsu, 1994). We examine the implications of this claim, and reject it on both scientific grounds, and within the larger context of the sociology of science as an exemplar of defining a phenomenon in terms of a particular paradigm.
The Perception Without Awareness Controversy
One of the most curious discrepancies between the literatures on perception without awareness and implicit memory concerns the employment of the subjective report, or "claimed awareness" measure. This measure requires the observer or the rememberer to comment upon their subjective phenomenal awareness while performing the experimental task. Although this measure has been used for over a century in studies of unconscious perception (see Adams, 1957, for a review of early studies), a subjective report measure of awareness has only recently been introduced in the context of the study of implicit memory (see Bowers & Schacter, 1990; Gardiner, 1988; Tulving, 1985). It is therefore not surprising that the most formidable alternative explanation raised against subjective report measures of awareness, namely the response bias problem, has been virtually ignored in the study of implicit memory. Eriksen's (1959, 1960) classic critique of the perception without awareness literature may represent the first effective articulation of the response bias problem. Consider the following caricature by Eriksen (1959), which illustrates how response bias can affect the measurement of awareness:
"A psychophysicist in setting about to determine the absolute threshold for light would not use an experimental situation where the subject was to press a button directly in front of him to indicate that he did not see the light and to get up and walk across the room to press a button on the wall if he did. Few would expect such an arrangement would yield as low an absolute threshold as would be obtained if the subject had the two buttons directly in front of him." (p. 205)
Obviously, there are much more subtle ways in which the demand characteristics of an experimental situation may bias subjects against reporting that they perceived a stimulus. However, Eriksen's caricature poignantly highlights a major weakness in any approach to the measurement of awareness which is solely based upon subjective report, or claimed awareness. The basic difficulty with this approach is that it places "on the individual subject the responsibility for establishing the criterion of awareness" (Eriksen, 1959, p. 203). Consequently, factors unrelated to awareness such as demand characteristics and pre-conceived biases may lead subjects to adopt a conservative response criterion and report null perceptual awareness even under conditions in which conscious perceptual information is available. Response bias represents a threat not only to the validity of the subjective report measure of awareness, but also to its reliability. In particular, variability in response criteria makes it difficult to compare reports of null subjective confidence across subjects, or within subjects across conditions.
Similar considerations led Eriksen (1959, 1960) to reject subjective report as an adequate measure of awareness, and to suggest instead that awareness be operationally defined in terms of performance on tasks that measure perceptual discriminations. Eriksen advocated employing forced choice discrimination as a measure of awareness. Examples of such tasks include forced choice presence-absence decisions, and forced choice discriminations among several stimulus alternatives. Measures based on discriminative responding have two important advantages over measures based on subjective reports. First, they may allow for obtaining a bias free measure of perceptual sensitivity (Green & Swets, 1966; Swets, 1964), and second, they represent a much more reliable index of perception relative to subjective report. For these reasons, following Eriksen, most researchers have preferred discriminative responding as an index of awareness (but see, Dixon, 1971, 1981; Henley, 1984).
Yet, whether evidence of perception without awareness is obtained crucially depends upon how awareness is operationally defined. While perception in the absence of subjective confidence is a very robust phenomenon (see Adams, 1957, for a review), evidence of perception under conditions which establish chance discrimination has not been forthcoming. Consequently, Eriksen (1960) concluded that there was no evidence for perception without awareness. Thus, the change in the operational definition of awareness introduced by Eriksen appeared to have completely eliminated the perception without awareness phenomenon.
Eriksen's conclusions were challenged when Marcel (1974, 1983) as well as others (e.g., Balota, 1983; Fowler, Wolford, Slade & Tassinary, 1981) reported findings which appeared to demonstrate perception in the absence of detection in a masked priming paradigm. The results of these studies suggested that a stimulus can influence responding even when subjects cannot discriminate between its presence or absence. Unfortunately, it subsequently became clear that while in principle the present-absent discrimination allows for obtaining a bias free measure of perceptual sensitivity, the implementations of this task in the above studies were inadequate to establish chance discrimination, and these studies more likely demonstrated perception in the absence of subjective confidence (e.g., Cheesman & Merikle, 1985; Holender, 1986; Merikle, 1982; Nolan & Caramazza, 1982).
For researchers who assume that measures of discriminative responding represent valid indicators of awareness (e.g., Holender, 1986), the failure to demonstrate perception in the absence of discriminative responding provides evidence against the existence of perception without awareness. However, as pointed out by Reingold and Merikle (1988, 1990), the validity of discriminative responding as a measure of awareness crucially depends upon its underlying exclusiveness assumption. Specifically, discriminative responding may constitute a valid measure of awareness if and only if it is influenced exclusively by conscious perceptual experience. If on the other hand a measure of discriminative responding is sensitive to both conscious and unconscious information, then equating awareness with discriminative responding may result in defining unconscious perception out of existence (see Bowers, 1984). The exclusiveness assumption has also been referred to as the process purity assumption (Jacoby, 1991) and the selective influence assumption (Dunn & Kirsner, 1988). We shall use the term process purity to describe any assumption that a task is exclusively influenced by a single process. Thus, while response bias represents a difficult problem for the subjective report approach to the measurement of awareness, process purity represents an equally devastating problem for the discriminative responding approach to the measurement of awareness. Given that demonstration of perception without awareness crucially depends upon which operational definition of awareness is used (i.e., subjective report vs. discriminative responding), converging on a valid indicator of awareness is of utmost importance.
Graf and Mandler (1984): A Case Study of Implicit Memory Research
With a few notable exceptions (e.g., Bowers & Schacter, 1990; see also the Remember/Know paradigm -- Gardiner & Java, 1993; Tulving, 1985), the subjective report measure of awareness has rarely been used in the study of the relationship between consciousness and memory. That is, asking subjects to provide commentary on their subjective phenomenal awareness during retrieval is seldom considered an adequate method for distinguishing conscious from unconscious retention. Instead, just as perceptual discriminations such as detection and identification have been used to assess perceptual awareness, memory discriminations such as recognition and recall have been linked to conscious recollection. Conversely, indirect or implicit memory tasks have often been considered as exclusively tapping unconscious memory. Thus, most memory researchers have tacitly adopted some form of the process purity assumption, equating tasks and processes. In addition, most studies of implicit memory have completely ignored the problem of response bias. To further clarify these points, we next turn to a closer examination of implicit memory research, and in particular, one of the most influential studies in this area, namely Graf and Mandler (1984).
During the past decade, numerous studies have compared the effects of independent variables on explicit versus implicit measures of memory in both amnesic patients and normal subjects (see Moscovitch, Vriezen, & Goshen-Gottstein, 1993; Richardson-Klavehn & Bjork, 1988; Roediger & McDermott, 1993; Schacter, 1987; Shimamura, 1986, for reviews). The basic aim of these studies was to demonstrate functional dissociations between explicit and implicit measures of memory. Briefly, a functional dissociation is observed if the nature of the effect of an experimental manipulation differs across the explicit and implicit tasks (see Dunn & Kirsner, 1988, for a critique of the functional dissociation paradigm). Although many such dissociations have been reported, there is no general consensus as to their interpretation (see Dunn & Kirsner, 1989). One problem in interpreting dissociations stems from the fact that the explicit and implicit tasks employed are often quite dissimilar (see Merikle & Reingold, 1991; Reingold & Merikle, 1988, 1990). Thus, the observed dissociation may simply be an artifact of such task differences. An example of one task dimension which very often differs across the explicit and implicit tasks is the nature of the test cues available during retrieval. For instance, if a word (e.g., TRAVEL) is presented during study, the same word would be presented as a test cue in a recognition task, in a degraded form (e.g., brief presentation followed by a mask) in a perceptual identification task, with only the first three letters in a stem completion task (e.g., TRA ____ ), with several letters deleted in a fragment completion task (e.g., T __ A __ __ L ), and not at all in a free recall task.
It is precisely in this context that Graf and Mandler (1984) made a truly unique contribution. They demonstrated a dissociation between an explicit cued recall task and an implicit stem completion task under conditions in which the test cues were identical across the two tasks. That is, during retrieval subjects were always presented with the first three letters of a word (e.g., DEFEND during study, DEF _____ during test), and were instructed either to complete each stem with a word they had seen during the study phase (cued recall instructions), or to respond using the first word that came to mind (stem completion instructions). In addition to the retrieval instructions, Graf and Mandler (1984) also manipulated the encoding task. They examined the effect of level of processing (i.e., semantic versus non-semantic; Craik & Lockhart, 1972) upon both the cued recall and stem completion tasks.
Graf and Mandler's (1984) findings are summarized in the top of Table 1. Two aspects of the results are particularly important. First, level of processing had an effect on the explicit cued recall task, but not on the implicit stem completion task (i.e., a functional dissociation). Second, there was a very impressive cross-over interaction. For words encoded semantically, the cued recall task was a more sensitive measure of memory relative to the stem completion task, whereas the reverse was true for non-semantically encoded words. This pattern of findings is all the more impressive given that the test cues were identical across both retrieval tasks; consequently, Graf and Mandler's (1984) findings have often been cited as a very powerful demonstration of functional independence between explicit and implicit memory tasks.
In draft, Table 1 appears here.
Graf and Mandler's (1984) finding that performance on the implicit stem completion task did not vary as a function of level of processing, whereas the closely matched cued recall task demonstrated such an effect, had tremendous impact on the field of implicit memory. Indeed, this finding was instrumental in the emergence of the tacit consensus that implicit measures exclusively reflect unintentional, automatic, unconscious retrieval. So ingrained is this belief in the process purity of implicit measures that it has endured despite recent reviews which clearly document that in the vast majority of implicit tests small, if not always statistically significant, level of processing effects are found (Brown & Mitchell, in press; Challis & Brodbeck, 1992). Rather than re-examine the process purity assumption, some researchers have preferred to account for level of processing effects on implicit tests by assuming another type of automatic priming which is sensitive to lexical and/or semantic processing (e.g., Challis & Brodbeck, 1992; see Toth & Reingold, this volume, for further discussion). The alternative of course, is that such effects reflect a consciously controlled influence on implicit performance, also referred to as conscious contamination (e.g., Toth, Reingold, & Jacoby, 1994). The potential danger in explaining incidents of conscious contamination as new types of automatic priming is the unwarranted proliferation of memory systems/processes (see Roediger, 1990).
However, subsequent findings of level of processing effects on implicit tests do not invalidate the impressive cross over interaction reported by Graf and Mandler (1984). Unfortunately, the interpretation of this effect is also unclear because, aside from instructions, there is an important difference between the cued recall task and the stem completion task used by Graf and Mandler (1984). That is, the two tasks differed in terms of their susceptibility to the influence of response bias: Subjects were required to give a response to every stem in the stem completion task, but not in the cued recall task. Thus, the dissociation may simply reflect a difference in response bias across tasks. More specifically, failures to provide a response in the cued recall task are ambiguous. The absence of a response to a stem may reflect the lack of conscious memory. Alternatively, subjects may have completed the stems only when they were very confident that they had seen the words during the study phase; but not when they had a poorer recollection of the prior occurrence of the word. That is, subjects may have simply adopted a conservative response bias.
One possible solution which may allow for ruling out the alternative explanation of response bias in the interpretation of Graf and Mandler (1984) would be to use a forced cued recall procedure rather than an unforced cued recall procedure. In the forced cued recall task subjects would be required to give a response to every stem even if they subjectively feel that they are only guessing. The forced cued recall procedure guarantees the same number of responses, and consequently the same response criterion, across subjects and tasks. Therefore, if the same pattern of dissociation obtained by Graf and Mandler (1984) occurs using forced cued recall, then the alternative explanation of response bias would have been successfully eliminated. A series of experiments by Graf, Squire, and Mandler (1984) is relevant in this context. In these experiments, amnesic and non-amnesic patients were tested using a procedure similar to Graf and Mandler (1984) with one of the differences being that forced cued recall was employed rather than unforced cued recall. The data from the non-amnesic patient control group in Graf et al. (1984) are presented in the bottom of Table 1. An inspection of the results reveals a very different pattern of findings relative to Graf and Mandler (1984). More specifically, in contrast to Graf and Mandler (1984), and consistent with the general pattern in the literature, level of processing affected both the cued recall and stem completion tasks. As an interesting aside, the results from the amnesic group (not shown in Table 1) demonstrated a sizable level of processing effect on the implicit stem completion task. Furthermore, the magnitude of the effect was not significantly different across the amnesic and patient control groups. Most importantly, the cross-over interaction obtained by Graf and Mandler (1984) was not obtained using the forced cued recall procedure. In particular, while cued recall was more sensitive than stem completion for semantically encoded words, for non-semantically encoded words stem completion was not more sensitive than cued recall. To verify that forced versus unforced cued recall is responsible for the difference in results between Graf et al. (1984) and Graf and Mandler (1984), Reingold and Merikle (1991) manipulated cued recall instructions by requiring subjects either to complete every stem (forced cued recall) or to complete stems only when they were sure that the completion represented an old word (unforced cued recall). As shown in Table 2, forced cued recall instructions (Experiment 1) replicated Graf et al. (1984) (i.e., a levels effect on the implicit test and no cross over interaction), while unforced cued recall instructions (Experiment 2) substantially replicated Graf and Mandler (1984) (i.e., a significant levels effect, but importantly, a cross over interaction).
Insert Table 2 about here.
What is the meaning of the difference in results across the forced versus unforced cued recall conditions? The interpretation of these findings is ambiguous. It may mean that the cross over interaction demonstrated in Graf and Mandler (1984) is an artifact of the response bias difference between the unforced cued recall task and the stem completion task. Alternatively, it is possible that the employment of the forced cued recall procedure, despite having the advantage of equating the influence of response bias across tasks, changes the nature of the cued recall task in another, more fundamental way. That is, when forced to provide a response to a stem in the absence of subjective phenomenal awareness, subjects' memory performance may be unconsciously influenced. This unconscious influence may operate similarly in both stem completion and forced cued recall, increasing the similarity across tasks and eliminating the cross over interaction.
The advantages and pitfalls of the forced cued recall procedure relative to the unforced cued recall procedure are very similar to the advantages and pitfalls of a discriminative responding measure of awareness relative to a subjective report measure of awareness. Recall that when subjects claim that they are purely guessing as to whether a stimulus was presented (i.e., claimed null awareness), this may reflect either a true lack of awareness, or the adoption of a conservative response criterion. Similarly, the absence of a response to a stem in the cued recall task is ambiguous. It may reflect the lack of a subjective experience of remembering on the part of subjects. Alternatively, subjects may have completed the stems only when they fully recollected, but not when they somewhat recollected, the prior occurrence of the word (i.e., a conservative response bias). Thus, the unforced cued recall procedure is more similar to the subjective report measure of awareness than is immediately apparent. In response to every stem subjects are required to use their own definition of what constitutes remembering. Given that subjective phenomenal awareness is the essence of the subjective experience of remembering (see Tulving, 1989), asking subjects to judge whether or not they remember is very similar to asking them to comment on their subjective phenomenal awareness during retrieval. In contrast, the forced cued recall procedure is very similar to the discriminative responding measure of awareness. Specifically, by eliminating variability in response criteria across both subjects and tasks, the forced cued recall procedure provides a more reliable and objective assessment of retention, and has the potential of ruling out the response bias interpretation. However, this increased reliability and objectivity may be achieved by sacrificing validity. In particular, forced cued recall may not be a valid measure of conscious remembering because it may in fact be influenced by both conscious and unconscious uses of memory.
Process Purity, Response Bias, and the Retrieval Intentionality Criterion
The difficulties posed by the process purity and response bias problems for the interpretation of Graf and Mandler's (1984) findings are especially worthy of close examination because these findings represent a prototypical example of a successful application of the retrieval intentionality criterion (henceforth, RIC). As proposed by Schacter, Bowers, and Booker (1989) this "criterion is comprised of two key components. First, the nominal or external cues provided to subjects on implicit and explicit tests should be the same, . . . Second, an experimental manipulation should be identified that selectively affects performance on one of these tasks and not the other." (p. 53). Graf and Mandler (1984) demonstrated precisely such a functional dissociation. With identical retrieval cues, level of processing affected the explicit cued recall test but did not affect the implicit stem completion test.
The RIC represents a refinement of the TDA to the study of unconscious or implicit memory. In part, it acknowledges the danger of equating between tasks and processes. As clearly expressed by Schacter et al. (1989):
"just because a test does not require a subject to think back to the study episode does not prevent the subject from doing so anyway. Once we acknowledge this possibility, the basis for drawing an implicit vs. explicit distinction becomes hazy indeed; we have no way of determining a priori whether we are dealing with an implicit or explicit form of memory on an allegedly "implicit test" unless we can convincingly distinguish between intentional and unintentional retrieval of information acquired during the study episode." (pp. 52-53).
This problem is commonly referred to as conscious contamination of indirect or implicit measures (e.g., Toth et al. 1994). A similar problem not mentioned by Schacter et al. (1989) is that explicit or direct measures of memory such as recognition and recall may also reflect both conscious and unconscious uses of memory. That is, direct or explicit measures may be contaminated by unconscious processes (Jacoby, Toth, & Yonelinas, 1993; Reingold & Merikle, 1988, 1990). This is just another way of stating the process purity problem (Jacoby, 1991). Both direct and indirect measures should not be considered process pure, or exclusive indicators of conscious and unconscious memory respectively. Given that the RIC approach acknowledges the process purity problem, what solution does it offer? According to Schacter et al. (1989):
"The logic underlying this retrieval intentionality criterion is straightforward: If the external cues are held constant on two tasks and only the retrieval instructions are varied, then differential effects of an experimental manipulation on performance of the two tasks can be attributed to differences in the intentional vs. unintentional retrieval processes that are used in task performance." (p. 53).
Despite the authors' claim that the logic underlying the RIC is straightforward, we do not believe it provides any clear solution to the response bias and process purity problems. The RIC acknowledges that a prerequisite for attributing any dissociation between explicit and implicit tests to retrieval intentionality is that the tests be matched on all other dimensions. In so doing it recognizes that the vast majority of reported dissociations in the implicit memory literature are open to criticism in that the tests used are not closely matched on dimensions such as retrieval cues. Unfortunately, by the same logic, the RIC does not go far enough because, as illustrated earlier, direct and indirect tests are rarely matched in terms of their susceptibility to response bias, a task dimension which was left out by the RIC.
A more rigorous approach to matching the direct and indirect tests was proposed by Reingold and Merikle (1988, 1990). This approach requires matching the tasks on all dimensions including retrieval cues and response bias. However, as illustrated by the results of the forced cued recall procedure discussed earlier, once such careful matching is adhered to, dissociations are very difficult to obtain (but see Merikle & Reingold, 1991). Furthermore, a thorough review of the literature suggests that dissociations such as the one reported by Graf and Mandler (1984), and required by the RIC, are rarely cleanly obtained. For example, the vast majority of studies using implicit tasks demonstrate a small, if not significant, level of processing effect (Brown & Mitchell, in press; Challis & Brodbeck, 1992). Thus, a meta-analysis would not support the conclusion of a functional dissociation as a result of level of processing. In addition, a functional dissociation which is demonstrated when an independent variable (e.g., levels of processing), effects one of the tests (e.g., explicit ), but not the other (e.g., implicit) (i.e., a single functional dissociation), by itself provides little evidence for functional independence between intentional versus unintentional retrieval processes because such dissociations may be easily explained by a single process model (Dunn & Kirsner, 1988).
Finally, and perhaps most importantly, Schacter et al. (1989) seem to suggest that attaining an experimental dissociation once under conditions of identical external cues may forever after be taken as evidence that the particular implicit measure is process pure. Specifically, they argue that if a variable produces parallel effects on previously dissociated implicit and explicit tests, it should not be concluded that this is a result of conscious contamination of the implicit measure. They justify this conclusion in the following manner:
"if we have already established that performance on these two tasks can be dissociated by experimental variable Q, then we can argue strongly against the idea that subjects treated the implicit test like an explicit test; if they had, variable Q could not have produced the dissociation that it did." (Schacter et al. 1989, p. 53).
Even if one accepts that a specific task dissociation constitutes evidence for unintentional retrieval in the performance on the implicit test, there is no guarantee that the implicit measure will always reflect unintentional retrieval under all experimental conditions. Thus, the RIC seems to make the questionable assumption that once uncontaminated is equal to never contaminated, and consequently an implicit measure should be considered as process pure with the only requirement being that it be dissociated once from its explicit counterpart.
Another attempt to solve the process purity problem is based on the performance of amnesics on implicit tests. As pointed out by Roediger and McDermott (1993), some researchers seem to make the assumption that if a particular implicit test shows preserved priming in amnesia, then such a test may be considered a process pure measure of implicit, unintentional retrieval. However, even if preserved priming in amnesia as demonstrated for a particular measure reflects unintentional retrieval, it does not at all follow that under other conditions (e.g., when performed by normal subjects with a specific encoding manipulation), the same measure may not be contaminated by consciously controlled retrieval. The real problem with using the RIC, or amnesia, as empirical benchmarks is that such a practice still tries to preserve the association between a particular task and a particular process. Consequently, conscious contamination is regarded as a red herring or a nuisance factor that can be examined once and then discarded forever.
This attitude is best illustrated by Endel Tulving's critical comment on Roediger and McDermott's (1993) treatment of the process purity issue. Tulving writes:
"your hand-wringing over how to tell a 'pure' implicit task, or implicit memory performance is a bit overdone...The issue is not that important at this early stage of the game...Putting it into the sharp focus of researchers' attention may overly cramp their style. Instead of finding out about the real brain/mind, they-- and editors and referees-- may get fixated on the problem of how 'pure' are implicit and explicit tasks." (Roediger & McDermott, 1993, p. 79, Footnote 5.)
Thus, Tulving (1985), who had once lamented that "nowhere is the benign neglect of consciousness more conspicuous than in the study of human memory" (p. 1) is now advocating the neglect of the relationship between consciousness and implicit/explicit task performance. It is important to note that in addition to downplaying the importance of the process purity problem as it applies to implicit measures, the implicit memory literature virtually ignores the possibility that performance on explicit measures is influenced by unconscious, unintentional retrieval (Jacoby et al. 1993; Reingold & Merikle, 1988, 1990). As shown by our analysis of Graf and Mandler (1984) and Graf et al. (1984), both the response bias and process purity problems have not been adequately dealt with by the TDA. We next examine the PDA and the solutions it provides to the process purity and response bias problems.
The Process Dissociation Approach: Assumptions and Contributions
The starting point for the PDA is the rejection of the process purity assumption. The PDA acknowledges that all tasks, whether explicit or implicit, can potentially be sensitive to both conscious and unconscious influences. This contrast represents a fundamental difference between the PDA and TDA philosophies. The PDA is based on the contention that the joint contributions of conscious and unconscious influences on task performance are the rule rather than the exception. It argues that manifestations of process purity are the true exceptions because they may only be obtainable under very extreme circumstances such as highly impoverished encoding (e.g., brief exposure, unattended input) in normal subjects, or in the case of a pure amnesic patient. From this perspective, if one is indeed a serious student of "the real brain/mind", one cannot afford to avoid studying conscious and unconscious processes as they interact in the codetermination of task performance. In the remainder of this section we describe how the PDA can yield evidence of unconscious, automatic influences on memory under conditions which rule out a response bias explanation. We then describe how the PDA can be used to derive quantitative estimates of conscious control and unconscious, automatic influences on memory. Finally, we consider the implications of involuntary conscious memory for the PDA and the TDA.
Unconscious Influences: Exclusion versus Implicit Tests
How then can one unambiguously demonstrate unconscious influence under conditions that rule out the response bias criticism? The solution offered to this question within the framework of the PDA is in the form of the exclusion condition which was originally developed by Jacoby and his colleagues under the term the opposition paradigm (see Jacoby, Kelley & Dywan, 1989; Jacoby & Whitehouse, 1989; Jacoby, Woloshyn & Kelley, 1989; Kelley & Jacoby, 1990), and was subsequently incorporated as one of the two conditions which constitute the Process Dissociation Procedure. To illustrate this condition, consider how it was applied in three different investigations (Reingold & Merikle, 1991; Richardson-Klavehn et al. 1994; Toth et al. 1994). Each of these was very similar to Graf and Mandler (1984) in that level of processing was manipulated at study, and words stems served as retrieval cues at test. However, in contrast to both stem completion and cued recall instructions, subjects were now asked to avoid completing the stems with old studied words. Instead, in the exclusion condition subjects were required to complete as many stems as possible with new, unstudied words. For example, if DEFEND was the studied word, the stem
DEF ____ should be completed with other words such as DEFECT, DEFINE, and DEFEAT. In this situation, the pattern of results which will support an interpretation of unconscious influence while ruling out the plausibility of a response bias explanation requires two important components. First, it should be demonstrated that for non-semantically encoded words, even when subjects are instructed to avoid old completions (exclusion instructions), subjects nevertheless complete the stems with old words above baseline. Second, it should be established that this inability to avoid old completions is not due to a general difficulty in generating new completions. That is, in the exclusion condition, for semantically encoded words, subjects should be better able to avoid old completions relative to their ability to do so for nonsemantically encoded words.As shown in Table 3, the results from the three studies which manipulated level of processing in the exclusion condition reveal exactly such a pattern of findings.
In draft, Table 3 appears here.
Above chance old completions in the exclusion condition are a compelling counterexample to the response bias interpretation in much the same way as Stroop interference is a compelling demonstration of automaticity (Stroop, 1935). In both cases, despite a strong incentive to consciously oppose automatic influences, such automatic influences are nevertheless manifested. Thus, the power of the exclusion condition to reject a response bias explanation is a direct result of it being an interference paradigm. This similarity between the exclusion condition and Stroop interference is perhaps not surprising given that theories of attention and automaticity inspired the development of the PDA (Jacoby, 1991; Jacoby, Ste-Marie, & Toth, 1993). Theories of attention and automaticity have long recognized that interference such as that in the Stroop paradigm is a much more compelling demonstration of automaticity relative to facilitation paradigms. Implicit tests constitute facilitation paradigms, and consequently, it is unclear to what degree performance reflects automatic facilitation or priming of old completions because such priming is at least sometimes joined by consciously controlled retrieval of old completions. In contrast, in the exclusion condition, automatic influences of memory are in the form of interference to the instructional set to produce new, unstudied completions. Thus, the exclusion instruction, which is an interference condition, has the potential to produce an unambiguous example of automaticity (i.e., above chance old completions), while implicit performance inevitably confounds conscious remembering with unconscious influences because both will result in old completions.
It is important to note that although above chance completions in an exclusion condition exclusively reflect unconscious influences on memory, the rate of completion with old words in an exclusion condition cannot by itself be used to estimate the magnitude of such unconscious, automatic influences. This is the case because performance in the exclusion task represents the combined effect of conscious control and unconscious influences. An estimate of unconscious, automatic influences can be derived but, as elaborated later, the relationship between consciously controlled and automatic influences must first be clearly defined (i.e., one must assume either independence, exclusivity, or redundancy).
The exclusion condition is more than just a clever methodological tool intended to rule out the alternative explanation of response bias. Underlying the exclusion condition is the more general view that a major function of consciousness is to oppose unwanted contextually primed or habitual responses. Perseveration is often given as an example of a failure to exercise conscious control in opposing the tendency for repeating an action which is automatically generated in response to contextual cues. Frontal lobe functioning is considered to be crucially involved in conscious control, and consequently, frontal damage due to brain injury or normal aging often results in perseveration. Other examples of failures of conscious control to oppose unconscious influences include phenomena such as action slips and slips of the tongue (see Erdelyi, 1985). In the exclusion condition, consciously controlled and automatic influences are said to result in opposing or qualitatively different behavioral consequences. More specifically, if subjects consciously retrieve an old completion, they are required to replace it with another word. Thus, any above baseline completion with old words can only result from unconscious influence coupled with a failure to consciously oppose it. As proposed by many investigators, the exploration of qualitative differences between conscious and unconscious processes is vital for the emergence of a general theory of the mind (e.g., Cheesman & Merikle, 1985, 1986; Dixon, 1971, 1981; Jacoby et al. 1989; Kelley & Jacoby, 1990; Merikle & Reingold, 1990; Reingold, 1992; Reingold & Merikle, 1990; Shevrin & Dickman, 1980). The exclusion condition is ideally suited for demonstrating strong qualitatively different (i.e., opposing) consequences of conscious control versus unconscious influences.
PDA and the Measurement of Conscious Control and Unconscious Influence
Within the PDA framework the exclusion condition is contrasted with the inclusion condition. In this condition, as implemented with word stems as cues, subjects are instructed to complete the stems with previously studied words. However, the instructions are similar to a forced cued recall procedure rather than an unforced cued recall procedure. That is, subjects are encouraged to complete as many stems as possible. To verify that there are no differences in response criteria across the inclusion and exclusion conditions, both conditions include baseline items (i.e., stems which do not correspond to any of the studied words). Provided that baseline completion rates are equivalent across conditions, proportion of old completions in the inclusion and exclusion conditions can then be used to derive quantitative estimates of the proportion of consciously controlled trials, and the proportion of unconsciously influenced trials. Proportion of trials completed with old words in the inclusion condition will include consciously controlled and/or unconsciously influenced trials, while in the exclusion condition proportion of old completions reflects unconsciously influenced trials which were unopposed by conscious control. Note that, in the exclusion condition, if a trial is both consciously controlled and unconsciously influenced, conscious control overrides the unconscious influence and consequently the stem is completed with an unstudied word. It is therefore essential to specify the proportion of consciously controlled trials in which an unconscious influence also occurred; that is, "overlap trials". Given that there currently exists no empirical method for estimating the proportion of overlap trials, three theoretical models have been used to specify this proportion: exclusivity, independence and redundancy (see Jones, 1987).
To facilitate understanding of the implications of the different models for the estimates of the proportion of consciously controlled trials (C), and the proportion of unconsciously influenced trials (U), they are graphically depicted in Figure 1. In the original implementation of the PDA (Jacoby, 1991; Jacoby et al. 1993) the independence model was adopted. According to this model, C and U are independent, and thus the proportion of overlap trials equals C times U. Several authors have suggested alternative models. Gardiner and Java (1993) proposed an exclusivity model in which C and U are mutually exclusive, and the proportion of overlap trials equals zero. At the other extreme Joordens and Merikle (1993) argued for a redundancy model in which C is always a subset of U, and, consequently, the proportion of overlap trials equals C.
In draft, Figure 1 appears here.
An inspection of Figure 1 reveals the computation of C does not depend upon which model is assumed. Regardless of the model assumed, C equals the proportion of old completions in the inclusion condition (I), minus the proportion of old completions in the exclusion condition (E):
This is because consciously controlled trials contribute to the proportion of old completions only under inclusion, but not under exclusion, and consequently, they represent the difference between these conditions (but see the later discussion of the differences in the meaning of the estimate of conscious control as a function of the various models). The value of C varies between 0 and 1. Zero represents the value expected in the case of a pure, densely amnesic patient who, regardless of instructional set, produces equal rates of completions with old words in both the inclusion and exclusion conditions. Such patients clearly do not intentionally control the influences of past experiences on their task performance. In contrast, a value of 1 is expected in the case of perfect conscious memory, such as when a single study word is clearly presented to a normal subject just before its corresponding stem with either inclusion or exclusion instructions. In this case one would expect subjects to always complete the stem with the study word when instructed to do so, and to always be able to avoid such a completion when so instructed. Obviously both of these extreme situations can be taken as representing process pure performance. These examples help to explain the PDA's approach to the measurement of conscious control, but, it should be emphasized that the PDA is especially important under conditions when task performance is codetermined by conscious and unconscious influences.
The basic idea behind the PDA measure of conscious control, is that if memory expression is under intentional conscious control, then subjects can modify and regulate the manner in which they demonstrate their remembering to fit task requirements. To the extent that subjects have conscious control over the act of memory expression, they should be equally capable of completing stems with old words under inclusion, and avoiding such completions under exclusion. Conversely, given that subjects have no control over how unconscious influences of memory are expressed, such influences will affect subjects' responses in an identical manner under both inclusion and exclusion instructions by increasing the probability of old completions.
To illustrate the importance of the concept of conscious control and its relationship to the initiation of an intentional and deliberate action, consider the following compelling example by Marcel (1986):
"Cortically blind patients who have no phenomenal experience of an object in the blind field will nonetheless preadjust their hands appropriately to the size, shape, orientation, and 3-D location of that object in the blind field when forced to attempt to grasp it...Yet such patients will make no spontaneous attempt to grasp a glass of water in their blind field even when thirsty. Voluntary actions often depend upon conscious perception." (p. 41).
The reason that the blindsight example is compelling is very similar to the reason that preserved learning in amnesia is impressive and compelling. That is, both cases represent an example of unconscious influences coupled with a complete inability to act intentionally upon perceived or remembered information, i.e., a total lack of conscious control. Indeed, one of the most tragic aspects of neurological disorders such as blindsight and amnesia is that despite the availability of relevant perceptual and memory information, patients are unable to act upon and make conscious use of this information. As illustrated by Marcel's (1986) blindsight example, the idea that conscious awareness is a prerequisite for initiating an intentional action is not new. However, the PDA offers a very elegant operationalization of conscious control as the contrast between trying to respond with target words (Inclusion) versus trying to avoid responding with target words (Exclusion).
Estimating the proportion of unconsciously influenced trials (U), is more complicated however. This is the case because such an estimate crucially depends upon the extent of overlap between consciously controlled and unconsciously influenced trials. As can be seen in Figure 1, the three models vary dramatically with respect to the hypothesized overlap between C and U. If such overlap exists, then the proportion of old completions in the exclusion condition (E) cannot be used as an estimate of U. As mentioned earlier, because conscious control overrides unconscious influence, such trials will not lead to old completions in the exclusion condition, even though an unconscious influence to do so was present. Thus, we need to calculate the proportion of such trials and add them to E to obtain an estimate of U. More precisely, each one of the three models -- exclusivity, independence, and redundancy -- specifies the conditional probability of U given C [i.e., P(U|C)]. The proportion of overlap trials between C and U is equal to the product of P(U|C) and C. Finally, the estimate of U can be obtained using the following equation:
 U = E+P(U|C)*C
According to the independence model P(U|C) = U. When applied to equation  we get:
 U = E+U*C
 U = E/(1-C)
The independence model is bracketed by two extreme models of overlap, the exclusivity model, which claims no overlap between C and U, i.e., P(U|C) = 0, and the redundancy model, which claims complete overlap between C and U,
i.e., P(U|C) = 1. Consequently, for the exclusivity model the estimate for U is:
 U = E
and for the redundancy model:
 U = E+C
Elsewhere, Jacoby, Yonelinas, and Jennings (in press) provided a very compelling case for the independence model. Indeed, the independence model has been tacitly assumed by implicit memory researchers in their quest to demonstrate functional and stochastic dissociations as evidence for independent memory systems or processes. A detailed discussion of the relative merits of the different models is beyond the scope of the present paper (see Curran & Hintzman, in press; Gardiner & Java, 1993; Jacoby & Begg, submitted; Jacoby, Toth, Yonelinas & Debner, 1994; Jacoby et al. in press; Joordens & Merikle, 1993).
Nevertheless, on the basis of the preceding analysis we would like to argue that, quite separate from the precise relational model adopted (i.e., independence, exclusivity, and redundancy), the PDA constitutes a major contribution to the study of conscious and unconscious influences on behavior by virtue of its handling of thorny problems such as process purity and response bias. Furthermore, the explicit relational assumption made in the original exposition of the PDA, namely the independence assumption, served as the impetus for the rapidly growing number of alternative proposals. The mere existence of this debate holds the promise that future theories will no longer be able to avoid specifying the relationship between consciously controlled uses of memory and unconscious influences of memory. Indeed, it is argued here that perhaps one of the most important long-term contributions of the PDA is in highlighting the importance of specifying the relationship between conscious and unconscious influences on task performance as a prerequisite for the study of cognitive processing with and without awareness.
In the next section we examine the implications of the different models as they relate to retrieval processes in general, and to involuntary conscious memory in particular. We argue that it is essential to specify any of the three relational assumptions in terms of a model of retrieval if the goal is to progress beyond a priori speculation to empirical verification.
The Problem of Involuntary Conscious Memory
The concept of involuntary explicit memory was first introduced by Schacter (1987) who defined it as "cases in which a test cue leads to an unintentional but fully conscious and explicit "reminding" of the occurrence of a prior episode. " (p. 510). He further stated that "At present, we know little about the relation between implicit memory and involuntary explicit memory, but future research and theorizing should be directed toward this issue." (p. 510). It is important to note that contrary to Schacter's (1987) suggestion, this concept has received little theoretical and empirical attention (but see Richardson-Klavehn et al. 1994, this volume). Ironically, despite its unresolved status within the TDA, this concept has been recently reintroduced as the cornerstone of several critiques of the PDA (Graf & Komatsu, 1994; Richardson-Klavehn et al. 1994). These critiques have received a tremendous amount of exposure (e.g., Roediger & McDermott, 1993), and in our opinion, have created substantial confusion. Accordingly, we dedicate this section to evaluating the implications of involuntary explicit memory for the PDA framework. We begin by summarizing these critiques. We then argue that in its current form the concept of involuntary explicit memory is too underspecified to serve as a useful theoretical constraint, or as an
a priori objection to either the PDA or the TDA. To clarify this concept, we argue that it is important to differentiate between subjective experience versus retrieval processes. We also highlight the need to specify a model of retrieval, and to operationalize retrieval intentionality. Next, the three relational assumptions (exclusivity, redundancy, and independence) are examined in terms of their implications for modeling retrieval in general, and involuntary conscious memory in particular. Finally, we consider the implications of involuntary conscious memory for the derivation of PDA estimates.
A grim assessment of the PDA in light of the involuntary conscious memory issue was put forward by Graf and Komatsu (1994):
"We maintain that the PDP is not suitable for learning about implicit versus explicit memory test performance because it does not distinguish between remembering that is initiated and guided by conscious intention versus remembering that is accompanied by "consciousness. . " . .The failure to distinguish between the different forms of remembering is built into the exclusion condition . . . where subjects are instructed to say NO to two kinds of critical items: those that are remembered as a result of a conscious intention, as well as those whose retrieval is followed by conscious awareness of their occurrence in a previously studied list." (p. 116)
A remarkably similar argument has been independently advanced by Richardson-Klavehn et al. (1994) who argued "that it is necessary to make independent distinctions based on retrieval intentionality (intentional vs involuntary) and awareness (conscious memory vs unconscious memory), and in particular, that it is necessary to take account of influences of memory that are involuntary, but accompanied by an awareness of the past (involuntary conscious memory)." (p.22). They further argued that "when items associated with conscious memory come to mind involuntarily in an exclusion test, those items will be excluded from subjects' overt responses, reducing observed priming." (p.22). As a consequence they argued that, the estimate of unconscious, automatic influence within the PDA "captures only a subset of involuntary influences -- those that are unaccompanied by an awareness that an item has been previously encountered." (p. 22).
As an indication of the impact of these critiques, consider how they are echoed in a thorough and scholarly review by Roediger and McDermott (1993):
"[S]ubjects may exclude responses on several distinguishable bases: (a) they may intentionally recollect the item and exclude it, or (b) they may unintentionally retrieve the word and then recognize it and exclude it . . . If subjects covertly produce the response through an automatic process (or through incidental retrieval) and then later exclude it on the basis of recognizing it as being from the list, then the process dissociation procedure will overestimate how contaminated implicit tests are by intentional retrieval processes." (p. 77)
The basic thrust of the critiques is that in principle involuntary conscious memory cannot be handled by the PDA. This flaw is said to be related to the exclusion procedure, and to result in an underestimation of unintentional, automatic retrieval. On the basis of this a priori assessment, a comparison of PDA estimates with implicit/explicit test performance is argued to be invalid and misleading.
To date the concept of involuntary conscious memory is based on an appeal to the introspective experience of subjects in a stem completion-like task. In such a task it occasionally appears to subjects that the completion popped to mind effortlessly, and only later did they become aware that this completion was a studied word. However, one should be very cautious in formulating theories which are solely based on such introspective experience. For example, it is widely accepted that subjects are not typically aware of the processes underlying retrieval. Thus, phenomenologically speaking, experiences of involuntary explicit memory may occur in a variety of memory tasks such as explicit cued recall. Given that retrieval intentionality cannot be equated with retrieval instructions, and cannot be directly observed, how is it possible to empirically distinguish involuntary from intentional retrieval? In the absence of an independent measure of retrieval intentionality, such appeals to subjective phenomenal experience can only remain in the speculative realm. Furthermore, if the concept is to be used as a focal point for assessing measures of memory, it is imperative that the concept be considered in the context of specific models of retrieval pertaining to specific experimental tasks. Failing this, introducing this underspecified concept does nothing more than muddy the theoretical waters. In the next section we consider involuntary conscious memory in the context of the direct retrieval and generate/recognize models of retrieval.
Figure 2 illustrates the direct retrieval and the generate/recognize accounts of involuntary conscious memory. An inspection of Figure 2 points out some of the key differences between these two interpretations. First, whereas the generate/recognize model assumes sequential dependence between automatic generation and recognition, the direct retrieval model assumes the existence of independent, consciously controlled and automatic retrieval processes. The stem or fragment cues serve as part of the input for both of the retrieval processes assumed by direct retrieval. In contrast, whereas these cues also feed into automatic generation, the input to recognition is a word completion generated automatically. Because part of the input to recognition is the output of generation, recognized items are always a subset of generated items, i.e., a redundancy assumption. Figure 2 is not meant to illustrate all test trials, or even a typical test trial as explained by these retrieval models. Rather, this Figure illustrates that both models can account for the subjective phenomenal experience which has been termed involuntary conscious memory. This subjective experience is composed of two separate parts, the first being an experience of a completion popping to mind effortlessly without any episodic detail about its prior occurrence. Subsequently, the subject becomes aware of the prior occurrence of the word, and this awareness is accompanied by some amount of episodic detail. Within the generate/recognize model these two aspects of the subjective experience correspond very accurately to the output of generation and the subsequent output of recognition. Note that recognition is not assumed to occur for every generated item, but is a necessary part of involuntary conscious memory. More importantly, in the generate/recognize model the temporal aspect of subjective experience is deemed an accurate reflection of the sequential dependence between processes. This is not the case for direct retrieval. In the latter, the two aspects of the subjective experience are linked to the output of automatic retrieval and consciously controlled retrieval. However, the crucial difference is that these outputs are assumed to be asynchronous. As depicted in Figure 2, in the case of involuntary conscious memory, the output of automatic retrieval is subjectively accessed prior to the output of the consciously controlled retrieval. Such a scenario is quite compatible with notions of the fast acting nature of automatic influences, and the typically slower speed of consciously controlled influences. However, while it is assumed that this is the case for "involuntary conscious memory" trials, the direct retrieval model does not assume that this is the case in every trial. In fact, the independence of the two processes is likely to result in all kind of patterns of asynchrony between processes, both at input and output. In other words, according to the direct retrieval model, the temporal aspect of the subjective experience of involuntary conscious memory is misleading because it suggests a sequential dependence when independence exists. We thus conclude that the subjective experience of involuntary conscious memory can be accommodated by the PDA and the independence assumption within the context of direct retrieval, but is problematic for the independence version of the PDA when cast in the context of a generate/recognize model.
In draft, Figure 2 appears here.
Involuntary Conscious Memory and Exclusivity, Independence and Redundancy
By definition, a trial in which involuntary conscious memory occurred represents an automatic, unintentional, unconscious influence coupled with conscious remembering. Such trials have been referred to above as "overlap trials". As explained earlier, the three relational assumptions, exclusivity, independence and redundancy, differ in terms of the hypothesized probability of occurrence of such overlap trials. It is therefore necessary to examine the implications of these assumptions for models of retrieval, and for the issue of involuntary conscious memory.
The exclusivity model assumes that there is no overlap between unconscious and conscious influences, i.e., the probability of overlap trials equals zero. Consequently, we do not see how the concept of involuntary conscious memory can be accommodated to within this framework. Accordingly, Figure 2 only includes the generate/recognize and direct retrieval accounts of involuntary conscious memory which correspond to the redundancy and the independence assumptions respectively. What is ironic is that the same individuals who advocate the exclusivity model (Gardiner & Java, 1993; see also Richardson-Klavehn, Gardiner, & Java, this volume), also promote involuntary conscious memory as a difficulty for the PDA (Richardson-Klavehn et al. 1994, this volume). Perhaps the reason for this is that Gardiner and Java (1993) did not really propose a model of retrieval based on the exclusivity assumption. Indeed, any attempt to do so yields the highly counterintuitive notion that on any given trial the occurrence of consciously controlled retrieval somehow precludes unconscious retrieval from occurring, and vice versa. Not surprisingly, as pointed out by Jones (1987), exclusivity models have not been forthcoming in the context of memory theorizing.
Rather than truly arguing for an exclusivity model of retrieval, Gardiner and Java (1993) state that the PDA independence model "merits serious consideration but it does raise the problem of specifying how independent processes can be associated with states of awareness that are exclusive" (p. 181). In other words, Gardiner and Java (1993) point out that according to the independence model conscious and unconscious influences can and do co-occur. However, conscious and unconscious states of awareness are mutually exclusive; that is, a subject cannot be simultaneously aware and unaware of the prior occurrence of a given word. The key to this confusion is of course that the PDA assumes the co-occurrence of conscious and unconscious influences, but does not assume that they co-occur in time (i.e., simultaneity of influence). It appears that Gardiner and Java (1993) tacitly interpret the PDA independence model to mean that consciously controlled and unconscious influences should operate simultaneously. This assumption is clearly not made by the independence model adopted by the PDA. In fact, some asynchrony between conscious and unconscious influences is entirely to be expected. One can further suggest that in line with theories of automaticity, unconscious influences are typically faster than consciously controlled influences. Consequently, it is entirely possible that a fast acting unconscious influence occurs, and is subjectively experienced as the effortless popping to mind of the target word. Subsequently, when the same target word is independently retrieved in a consciously controlled manner, the result is a full blown phenomenal awareness of the word's prior occurrence (see Figure 2).
To further clarify how the independence model can accommodate for the subjective experience of involuntary conscious memory, it is important to realize that Jacoby et al. (1993) argued for a specific model of retrieval which they referred to as direct retrieval. This model assumes that there are no separate generation and recognition stages involved in the retrieval of target words as a response to stem cues. Rather, the direct retrieval model proposes independent, consciously controlled and automatic retrieval processes that may or may not access and output studied words as completions (see Figure 2). Thus, there is a certain probability that both processes will retrieve the studied word. On any given trial the probability of such overlap will equal U*C. On some proportion of such overlap trials subjects may subjectively experience the unconscious influence first followed by the consciously controlled influence. However, this temporal aspect of phenomenal experience does not imply sequentiality, or dependence of the underlying processes.
Turning to the redundancy assumption, it is important to note that Joordens and Merikle's (1993) focused on reinterpreting the results of Jacoby et al. (1993) in light of the redundancy rather than the independence assumption. These authors did not propose a model of retrieval as a rationale for their preference for redundancy over independence. In fact, a generate/recognize retrieval model which is consistent with the redundancy assumption does exist, and was considered by Jacoby et al. (1993). We argue that it is vital to specify relational models at the retrieval level if they are to be meaningfully and empirically contrasted. Accordingly, we take a closer look at current formulations of generate/recognize models of retrieval as they apply to the PDA and TDA.
Jacoby and Hollingshead (1990) proposed a generate/recognize model to explain performance on stem completion and cued recall tasks. Jacoby et al. (1993) considered the implications of this model in the context of performance in exclusion and inclusion tasks. Recently, Challis and Tulving (1993) proposed a generate/recognize model which is very similar to the one proposed by Jacoby and Hollingshead (1990). Challis and Tulving (1993) argued that the "generation/recognition account of performance on word-fragment cued tests holds that implicit processes of a perceptual and semantic nature underlie generation of items, and depending on the test instructions, recognition is used to select studied items from those generated." (p.1 in the program of the 34th Annual Meeting of The Psychonomic Society). Similarly, Jacoby and Hollingshead (1990) suggested that "a manipulation of cued-recall vs. stem-completion instructions has little, if any, effect on generation processes. Rather, the effect of giving cued-recall instructions is to add recognition-memory processes to the generation processes required for stem-completion performance." (p.435).
For the present discussion, the important aspects of these generate/recognize models are as follows. First, performance on memory tasks employing stem or fragment retrieval cues is thought to involve two distinct stages -- automatic, unintentional generation of completions, sometimes followed by recognition of their prior occurrence. Recognition processes are dependent on generation processes in that unintentional, automatic generation precedes, and is a prerequisite for recognition. In other words, an item can be recognized if and only if it was first automatically generated, i.e., a relationship of redundancy. Recognized items are a subset of generated items because recognition is dependent on prior automatic retrieval. Most importantly, manipulations of intentional retrieval instructions (e.g., cued recall) versus incidental retrieval instructions (e.g., stem completion) are predicted to have no influence on automatic generation of completions; rather, their impact is thought to be restricted to the recognition process. Thus, cued recall is hypothesized to be equivalent to stem completion in terms of generation. However, cued recall differs from stem completion in that a covert recognition stage is added subsequent to generation.
Jacoby and Hollingshead (1990, Experiment 2) tested this prediction by contrasting a generate/recognize test condition with a cued recall test condition. During study, words were either generated from an anagram, or read. The generate/recognize test condition was designed to simulate the two stage sequence hypothesized to underlie cued recall test performance. In the generate/recognize test condition subjects were instructed to complete stems with the first word that came to mind, and then to make an old/new recognition decision. Cued recall instructions required subjects to complete the stem with studied words. Subjects were asked to avoid completions if the stem did not correspond to a studied word. Both single completion and multiple completion stems were used during test.
An important feature of this experiment was an attempt to equate recognition processes across the cued recall and generate/recognize conditions. Note that a generate/recognize model assumes that performance in cued recall reflects a two stages process whereby completions are automatically generated and then subjected to a covert recognition decision. If the decision criterion of this covert recognition stage in cued recall is equated to the decision criterion used in the overt recognition stage in the generate/recognize condition, then, the probability of cued recall should equal the probability of completion plus recognition in the generate/recognize condition. To verify that decision criteria were equated across the overt and covert recognition decisions, baselines stems were included which did not correspond to any of the studied words. If the probability of falsely recognizing a completion of a baseline stem as old did not differ significantly from the probability of false recall (i.e., completion of baseline stems), then decision criteria for overt and covert recognition would be assumed to be equivalent (but see Jacoby & Hollingshead, 1990).
As shown in Table 4, despite no difference in baseline (i.e., false recognition versus false recall), the probability of cued recall was higher than the probability of completion plus recognition in the generate/recognize test condition. Importantly, this finding was obtained for both single and multiple completion stems. With multiple completions it can be argued that in the cued recall test subjects internally performed multiple generate/recognize cycles which increased the probability of studied words being retrieved relative to the single cycle imposed in the generate/recognize condition. However, this explanation cannot account for the single completion stem results which are therefore all the more impressive. This is the case because the generate/recognize condition with single completion stems represents a very close simulation of the hypothesized two stage process underlying cued recall performance with these stems. Consequently, the difference in performance across these conditions is very difficult to interpret within the framework of current generate/recognize models. This is the case because the instruction to retrieve studied words in the cued recall test condition resulted in the retrieval of more words relative to the incidental retrieval instructions in the generate/recognize test condition. Given equivalent baseline performance, this result cannot be easily explained as a difference in response criteria across the generate/recognize and the cued recall conditions.
In draft, Table 4 appears here.
We see two potential interpretations for these results, both of which are problematic for current generate/recognize models. The first interpretation would suggest that at least in some trials, the instruction to intentionally retrieve in the cued recall condition resulted in consciously controlled retrieval of studied words, as would be predicted by a direct retrieval model. The second interpretation would require generate/recognize models to be modified so as to acknowledge the possibility that retrieval instructions can affect both generation and recognition processes, and consequently, an instruction to intentionally retrieve studied words in the cued recall condition may increase the automatic generation of studied items.
What are the implications of the preceding analysis? Although current implementations of generate/recognize models require modification, it appears that the redundancy relational assumption could be implemented to form a viable, specific model of retrieval. Furthermore, as demonstrated by Jacoby and Hollingshead (1990), the predictions of such models could be empirically tested and contrasted with the predictions of a direct retrieval model. This is obviously a very important direction for future research and theorizing.
The Jacoby and Hollingshead (1990) results may have another important implication. Specifically, several researchers, for example, Richardson-Klavehn et al. (1994), modified the instructions in an exclusion task relative to the original instructions used in the PDA. These researchers used generate/recognize type instructions in the exclusion condition asking subjects to complete stems with the first word that came to mind, and if this word was a studied word, they were to write down another completion instead. In contrast to these generate/recognize retrieval instructions, PDA exclusion instructions provide subjects with an intentional set vis a vis the retrieval of studied words. In the PDA, subjects are equally oriented toward the study episode in both inclusion and exclusion. However, whereas in the inclusion condition they are required to complete with old words, in the exclusion condition they are to avoid responding with these words. Given the Jacoby and Hollingshead (1990) result, this procedural departure may be more meaningful than immediately apparent. Indeed, in a recent study, Jacoby and Begg (submitted) manipulated retrieval instructions in inclusion and exclusion tasks. They contrasted generate/recognize instructions with the original PDA instructions, and reported that the former yield results consistent with a redundancy model, while the latter yield results consistent with an independence model. This is further support for our conclusion that both redundancy and independence represent viable models of retrieval in specific test conditions. This highlights the need to empirically investigate conditions which produce performance consistent with redundancy versus independence models. Thus, it is futile to look for a single model that applies to all retrieval tasks. However, in contrast to Joordens and Merikle (1993), we argue that current empirical evidence derived from studies implementing the PDA using the original instructions, produced results which are more coherently interpreted by estimates based upon the independence assumption. (For a detailed empirical evaluation see Jacoby & Begg; submitted; Jacoby et al. in press; Jacoby, Yonelinas, & Jennings, in press.)
Finally, it is important to consider how a generate/recognize model can account for the subjective experience of involuntary conscious memory. As seen in Figure 2, in the case of generate/recognize models, the subjective experience of involuntary conscious memory represents a very accurate description of the two stage sequence of processing underlying performance. Specifically, the unintentional, automatic generation would be subjectively experienced as the completion popping to mind effortlessly unaccompanied by episodic detail. Subsequently, this completion could be recognized, resulting in the subjective experience of remembering the prior occurrence of the completion (i.e., episodic detail). Given this correspondence between phenomenology and retrieval processes, we suspect that the critiques introducing involuntary conscious memory as a problem for the PDA tacitly assume a generate/recognize model. However if this is the position of Graf and Komatsu (1994), it is not apparent in their critique. In fact, Graf and Komatsu (1994) state their commitment to the independence assumption as part of their modified "general model for inclusion and exclusion test performance" (p. 123). Similarly, Richardson-Klavehn et al. (1994, this volume), do not acknowledge adopting a generate/recognize retrieval model as part of their critique of the PDA. If the authors of these critiques wish to advance involuntary conscious memory within the context of a generate/recognize model, they should be prepared to explicitly state their commitment to a redundancy assumption. Contrary to the argument advanced in the critiques, it is clear that involuntary conscious memory is not in principle incompatible with the PDA. While it will represent a problem if interpreted within the generate/recognize model, it can be handled in a straightforward manner by the direct retrieval model. Empirical evidence should be obtained to decide between the direct retrieval and the generate/recognize interpretations of the subjective experience of involuntary conscious memory.
Involuntary Conscious Memory, the Exclusion Condition, and Estimates of the PDA
As mentioned earlier, the critiques of the PDA claimed that a basic flaw in the exclusion procedure makes the PDA ineffective in distinguishing between involuntary conscious memory versus consciously controlled influences on memory, resulting in an underestimation of unconscious automatic influence. In this section we consider the implications of involuntary conscious memory for computing PDA estimates using the independence equations. We do this in turn for the direct retrieval and generate/recognize interpretations of this phenomenon.
Within the context of direct retrieval, a subjective phenomenal experience of involuntary conscious memory may or may not accompany an overlap trial, i.e., a trial in which both conscious and unconscious influences co-occurred. The total proportion of overlap trials is U*C. As pointed out in the critiques, in the exclusion condition overlap trials, and therefore "involuntary conscious memory" trials, should not result in completions with studied words. Consequently, it is true that the proportion of stems completed with studied words in the exclusion condition, P(Exclusion), may underestimate unconscious, automatic influences (U). The only relational assumption for which this is not true is exclusivity (U=P(Exclusion), see Equations  and ). Again this is ironic given that Richardson-Klavehn et al. (this volume) and Gardiner and Java (1993) have both raised involuntary conscious memory as a problem unique to the PDA, and at the same time favoured the exclusivity assumption. However, the leap in logic which the critics appear to make is to conclude that because performance in exclusion underestimates unconscious, automatic influence, the PDA underestimates unconscious, automatic influence. This is very puzzling because as explained earlier, the proportion of overlap trials is added to the proportion of stems completed with studied words under exclusion to provide an estimate of unconscious, automatic influence: (U=P(Exclusion)+U*C; see Equations  and , and Figure 1 above). To summarize, contrary to the criticisms of the PDA, under the direct retrieval interpretation of involuntary conscious memory, the PDA estimates using the independence model are valid. This is not surprising given that within this framework involuntary conscious memory is viewed as an epiphenomenon that is misleading vis a vis the underlying retrieval processes.
Quite a different picture emerges if a generate/recognize interpretation of involuntary conscious memory is adopted. According to this interpretation "involuntary conscious memory" trials would inflate the proportion of overlap trials in excess of the U*C value. In the most extreme condition, if all trials in which consciously controlled influences occurred conformed to generate/recognize retrieval, the true proportion of overlap trials would equal C, rather than U*C. This is of course the prediction of a redundancy model which holds that consciously controlled influences are a subset of unconscious influences. Under such circumstances computing PDA estimates by applying the independence formulas is obviously inappropriate. If independence formulas are nevertheless applied, the result will be an underestimation of unconscious, automatic influences. Such an inappropriate application of the independence formulas will produce varying degrees of underestimation depending on the difference between the true proportion of overlap trials minus the independence estimate of overlap trials:
 P(U|C)*C - U*C = Equation  -minus Equation 
where P(U|C) is the conditional probability of U given C. The magnitude of the underestimation will vary between zero when independence formulas are employed appropriately in the case of true independence, and:
 [C-U*C] = Equation  -minus Equation 
with the latter reflecting the underestimation expected by an inappropriate application of independence formulas in a situation where redundancy holds.
It is also worth contrasting the implications of a generate/recognize implementation of the redundancy assumption versus a direct retrieval implementation of the independence assumption in terms of their impact upon the estimate of conscious control. Whereas in both conditions conscious control equals P(Inclusion) minus P(Exclusion) (see Equation  and Figure 1), the meaning of conscious control is different between these two cases in an important way. Specifically, in both models conscious control over the act of memory expression is accurately assessed. However, only in the context of the direct retrieval independence implementation does it also meaningfully reflect conscious control over the act of memory retrieval. For the generate/recognize model, the story is much more complex, and is yet to be specified. Within current implementations of the generate/recognize model, applying the term conscious control to generation is an oxymoron because generation of completions to stems or fragments is assumed to always be unconscious, unintentional, and automatic (but see our discussion of the Jacoby & Hollingshead, 1990, result which may conflict with this assumption). In contrast, conscious control of retrieval may be meaningful for the recognition stage in this model. The recognition stage receives the generated completion as part of its input, and it is likely that the generate/recognize model would have to assume that recognition is at least in part under conscious control. It is the responsibility of proponents of generate/recognize retrieval to develop much more explicit models than currently exist. In our discussion of Jacoby and Hollingshead (1990) we have tried to point to potential possibilities. Nevertheless, conscious control of the act of memory expression can still be meaningfully assessed within this model. For generated words which were subsequently recognized, conscious control can be exercised in including or excluding them as completions in a manner consistent with task instructions. In contrast, unrecognized generated studied words result in memory expression over which subjects have no conscious control. This is the case because in both inclusion and exclusion such unrecognized generations will result in an increase in the probability of completing stems with studied words.
Thus, if involuntary conscious memory is equated with a generate/recognize redundancy model, the result will be an underestimation of unconscious, automatic influences. This is the case because the use of independence formulas in this context would be inappropriate. In addition, under these circumstances the meaning of the conscious control estimate is substantially different relative to its meaning within the context of a direct retrieval independence model.
To summarize, involuntary conscious memory cannot be used to invalidate the PDA for the following reasons. First, the critiques introduced involuntary conscious memory in a manner which was vague and underspecified, and which did not distinguish between subjective experience and the underlying retrieval processes. Second, the critiques become meaningful only in the context of a specific generate/recognize implementation of the redundancy model which we provided. In addition, on the basis of evidence obtained by Jacoby and Hollingshead (1990) we argued that current generate/recognize models require modification. Third, it is not at all clear that a generate/recognize model is assumed in the critiques. In fact, there are indications to the contrary. Fourth, we have demonstrated that the direct retrieval model can accommodate the subjective experience of involuntary conscious memory in a straightforward manner, therefore, soundly rejecting the argument that involuntary conscious memory is inherently incompatible with the PDA. Fifth, our analysis clearly indicated that deciding between various interpretations of involuntary conscious memory requires an empirical evaluation, and cannot be done solely on the basis of
a priori considerations such as the ones advanced in the critiques. Finally, we find it ironic that proponents of the TDA (e.g., Graf & Komatsu, 1994) wish to disqualify the PDA on the basis of an issue which was originally introduced as a problem for the TDA (Schacter, 1987), and which is yet to be resolved, or even seriously considered, within this framework.
Is an Empirical Comparison Between TDA and PDA Valid?
We would like to use this final section of the chapter to respond to Graf and Komatsu's (1994) assertion that the PDA is invalid for learning about implicit versus explicit test performance (for a broad response to Graf & Komatsu (1994) see Toth, Reingold & Jacoby, in press). We view their statement and its implications as very questionable on both scientific grounds, as well as in terms of the sociology of science. On scientific grounds this assertion ignores the value of convergent evidence across methodologies for studying the relationship between consciousness and memory performance. One strategy employed by PDA researchers has been to compare PDA estimates with implicit performance results using some of the primary experimental variables employed in TDA research (e.g., Jacoby et al. 1993; Reingold & Goshen-Gottstein, submitted; Toth et al. 1994). We believe that such a strategy provides vital input without which an evaluation of the validity of either paradigm is impossible. The value of a priori analyses such as the PDA critiques and the current response not withstanding, the final test of any theoretical/methodological paradigm is empirical. Such paradigms should be evaluated in terms of their ability to parsimoniously accommodate a large body of empirical evidence, as well as their ability to generate and stimulate empirical research. By discouraging empirical comparisons between paradigms, the possibility of arriving at a unified theoretical explanation is sabotaged, and consequently, fragmentation of the literature prevails, and empirical progress is stifled.
Most importantly, the PDA makes specific predictions about the relationship between PDA estimates and performance on implicit tests, and these predictions should be empirically explored. For example, the PDA predicts that under conditions where conscious control is minimal, the estimate of unconscious, automatic influence should approximate performance on comparable implicit tests. This is the case because under such conditions the indirect measure could be expected to reflect a relatively uncontaminated measure of automatic, unconscious influence. Such a pattern was obtained by Toth et al. (1994, Experiment 1). In this experiment a level of processing manipulation was used in an indirect stem completion task, as well as in inclusion and exclusion tasks. In the nonsemantic encoding condition, the estimate of conscious control approximated zero (.03), and the estimates of automaticity and indirect performance were in close agreement (.44 versus .45 respectively). In the semantic encoding condition where the estimate of conscious control was substantial (.27), indirect performance exceeded the estimate of unconscious, automatic influence (.51 versus .45). The interpretation proposed by Toth et al. (1994) was that in the semantic encoding condition, indirect stem completion performance was not a process pure measure of unconscious, automatic influence. Rather, it was contaminated by consciously controlled influences. Thus, Toth et al.'s (1994) comparison of performance across the TDA and PDA highlights the potential interpretation of level of processing effects in implicit tests as representing the conscious contamination of these tests. The alternative interpretation of these effects based on an assumption of process purity is one which invokes additional types of priming effects which are sensitive to lexical and/or semantic processing (e.g., Challis & Brodbeck, 1992). If for no other reason, parsimony dictates that the conscious contamination explanation should be given serious consideration (for a detailed discussion see Jacoby et al. in press; Toth & Reingold, this volume; Toth et al., 1994; for additional examples of convergence between implicit performance and PDA estimates see Jacoby et al. (1993); Reingold & Goshen-Gottstein, submitted).
By not acknowledging the process purity problem within the context of TDA, Graf and Komatsu (1994) fail to appreciate the solution to this problem which is embodied in the PDA. Indeed, once the potential pitfalls of equating processes with tasks are acknowledged, it becomes clear that there are at least two potential strategies that can be pursued in the study of implicit memory. The first involves designing or identifying retrieval tasks which may be less sensitive to the influence of conscious control (e.g., Rajaram & Roediger, 1993; Roediger, Weldon, Stadler, & Riegler, 1992). The second strategy is to use a method like the PDA for the purpose of quantifying the separate contributions of consciously controlled and unconscious, automatic processes. Whereas the former strategy focuses on the search for process pure tasks, the latter strategy searches for an appropriate technique for separating processes within a single task. We have opted for the latter strategy because we believe that in most situations conscious and unconscious influences codetermine task performance. Nonetheless, a PDA-like approach may also prove useful in the search for process pure tasks. This is the case because the inclusion/exclusion contrast provides a novel approach to the measurement of conscious control. Thus, if inclusion, exclusion, and indirect test conditions are used with the same encoding manipulation (e.g., Toth et al. 1994; Reingold & Goshen-Gottstein, submitted), it may be possible to correlate the magnitude of conscious control with implicit task performance. If the involvement of conscious control in a task is strongly manipulated without an effect on implicit performance this would provide evidence that for the specific manipulations tested, this task was not consciously contaminated. Even in the case of amnesic patients estimating conscious control may prove valuable in determining to what extent a given patient exemplifies "pure amnesia". Apart from the potential diagnostic implications of such an application, this possibility may prove important in the context cognitive neuropsychology research. This is the case because it may be used to evaluate the homogeneity of patient groups for the purpose of between group comparisons. One of the classic problems facing researchers studying patients with neurological deficits, is that each patient has unique pattern of deficits making the interpretation of group comparisons difficult. While the possibility of employing the PDA in the search for "pure tasks" and "pure amnesics" is intriguing, it is important to remember that both tasks and patients may not be pure across all situations. Thus, the two research strategies should not be viewed as mutually exclusive, rather, as complimentary.
As part of their critique of the PDA, Graf and Komatsu (1994) proposed "a more general model for inclusion and exclusion test performance" (p. 123). The questions then become, is the PDA the appropriate technique for separating consciously controlled versus unconscious influences in a specific task? and which relational assumption should be used? Resolution of these questions must await further empirical evaluation. However, we believe that the PDA takes a meaningful step in the right direction, and that the recent interest in modeling conscious and unconscious influences is a direct result of this change in emphasis brought about by the PDA. One of the motivations underlying the original formulation of the PDA (Jacoby, 1991; Jacoby et al. 1993) was to start with a minimal number of parameters, as well as with the independence assumption, which as pointed out by Jones (1987), is the most parsimonious of the relational assumptions. The idea was to attempt to model a substantial body of evidence with as simple an approach as possible, and if so indicated, to expand it later on the basis of empirical feedback. In contrast, it is important to point out that the model proposed by Graf and Komatsu (1994) has employed a substantial number of additional parameters without providing a clear operationalization. As such, it is at best a description, not a model (see Toth et al., in press). In addition to the emphasis on quantifying the separate contributions of processes, the exclusion condition included in the PDA provides a powerful methodology for conclusive demonstrations of unconscious influences, and for the exploration of qualitative differences between conscious versus unconscious processes.
In order to provide a more complete treatment of the TDA versus PDA controversy, we feel compelled to address the sociological and political context surrounding this controversy. One important reason why we so strongly object to Graf and Komatsu's (1994) statement that the TDA and PDA cannot be validly contrasted empirically, is that we perceive this statement as attempting to establish a territorial claim, and a monopoly on the study of conscious versus unconscious memory. The dominance of the TDA paradigm during the last decade may make it easy to forget that this methodology was initially devised to study phenomena which were previously investigated using a multitude of paradigms, and a variety of different terms such as consciousness, automaticity, attention, intentionality, and awareness. Indeed, the terms implicit and explicit memory were initially defined with respect to conscious recollection at the time of retrieval (e.g., Schacter, 1987). As often happens, the initial goal which motivated the development of the TDA methodology has sometimes been replaced with the goal of studying the methodology and tasks for their own sake. That is, redefining the phenomena in terms of the particular methodology which was originally devised with an external research goal distinguishable from this methodology. Thus, some researchers investigate memory tasks such as stem completion, fragment completion, and perceptual identification, while losing sight of the larger goal of studying conscious versus unconscious memory. Similarly, Erdelyi (1985) noted another example of the confusion between theoretical goals and the tasks designed to achieve them. The general consensus in the late 1950's was that "the failure of experimental methodology to corroborate the existence of unconscious processes was taken, as a matter of course, to reflect a failure of the concept rather than a failure of the extant methodology" (pp. 58-59)(see Reingold & Merikle, 1990 for a detailed discussion). To Graf and Komatsu's (1994) claim that the PDA cannot be used to study performance in implicit and explicit tests we reply that the PDA is not intended to study implicit and explicit performance, rather it is intended to study the phenomena which such tests were originally designed to investigate.
In the present chapter we argued that presenting issues such as the relational assumption problem, and the involuntary conscious memory problem, as unique to the PDA, while ignoring their obvious relevance to the TDA, represents a double standard. As an additional example, consider the argument by Graf and Komatsu (1994) that the PDA "does not take into account false alarm rates" (p. 120). The irony of this argument should be glaringly apparent in the context of the present paper. As we clearly documented, the TDA completely ignored the issue of response bias, and differences in base rates across contrasted implicit and explicit tests (e.g., Graf & Mandler, 1984). In contrast, PDA researchers are extremely careful in comparing base rates across inclusion and exclusion conditions. Indeed, demonstrating no difference in response bias (i.e., base rates) across inclusion versus exclusion has been routinely used as a prerequisite for computing the PDA estimates (see Yonelinas, 1994). Thus, a demonstration that base rates differ significantly across inclusion and exclusion, such as the one provided by Komatsu, Graf, and Uttl (1994), cannot be argued to invalidate the PDA, any more than an example of the violation of the assumptions underlying ANOVA can be used to invalidate that procedure (see Jacoby & Begg, submitted). The explicit treatment of the issue of response bias within the PDA, is the best guarantee that when there is a difference across conditions, it will be detected and acknowledged. This cannot be said for the treatment of this issue within the TDA. Similarly, Richardson-Klavehn et al. (1994, this volume) and Gardiner and Java (1993) have criticized the independence assumption of the PDA while at the same time failing to acknowledge that the modified PDA-like procedure they proposed is based on an assumption of independence between involuntary conscious memory and involuntary unconscious memory. This tacit assumption is clearly evident when inspecting the computational formulas they proposed (see pp. 24-26, Richardson-Klavehn et al. 1994).
The intensive criticism of the assumptions attributed to the PDA, coupled with the failure to acknowledge their relevance to other paradigms, creates the impression that while the PDA involves especially problematic assumptions, other methodologies, such as the TDA, are much more straightforward and sound on a priori grounds. The illusion that the TDA is assumptionless is further aided by the fact that its underlying assumptions, (e.g., process purity, relational assumption) often remain tacit and unacknowledged. The use of the implicit/explicit terms to refer to both tasks and processes/systems contributed in a subtle way to the tacit assumption that implicit and explicit tests exclusively measure implicit and explicit memory respectively (see Reingold & Merikle, 1988, 1990; Richardson-Klavehn & Bjork, 1988). In contrast, PDA assumptions are stated explicitly, and thus are more open to criticism. In the sociology of science the perception that a particular paradigm is problematic often discourages the use of the paradigm, and impedes publication of research carried out within that paradigm. This may occur regardless of whether or not this perception is justified. When authority figures in a field promote such perceptions, less informed researchers may tend to take such criticisms at face value. This is especially true when the area in question is as complex as the study of consciousness and memory.
The relevance of considering the sociological and political context of the present controversy is vividly illustrated in the previously cited quote in which Tulving cautions that Roediger and McDermott's (1993) treatment of the process purity issue may result in reviewers becoming overly concerned about this issue, and consequently impeding the ability to publish TDA research. We want to strongly emphasize that by discussing issues which may be problematic for the TDA, it is not our intention to impede empirical progress within this framework. Despite our clearly stated theoretical bias, we attempted to provide a serious examination of issues as they apply to both the PDA and TDA. We share Tulving's concern, and we think that it is very unfortunate when empirical data is held hostage awaiting the resolution of theoretical debates. Such an unfortunate, inadvertent effect occurred following Eriksen's (1959. 1960) classic critique of the perception without awareness literature, and more recently, following Holender's (1986) critique of the same field. Currently, this area of research has dwindled to a mere trickle. With the recent successful demonstrations of unconscious perception by applying the PDA, and opposition paradigms (i.e., exclusion) (Debner & Jacoby, 1994; Jacoby & Whitehouse, 1989; Joordens & Merikle, 1992; Merikle & Joordens, in press) we hope that this trend will be reversed. However, we would add that if this is a concern for an established paradigm such as the TDA, it is definitely a concern for an emerging paradigm such as the PDA. Consequently, we object to the use of a priori critiques as a basis for rejecting empirical PDA research. At the same time, we do not think the solution to the potential danger that a preoccupation with assumptions will stifle empirical research, is to ignore important theoretical issues. Obviously, paradigms differ in their ability to handle certain theoretical issues, as well as some empirical findings. But it is precisely because all paradigms have some weaknesses that convergence across paradigms is imperative. The solution therefore is to promote an atmosphere in which the broader implications of issues are recognized as challenges for the entire field. Failing this, the TDA versus PDA controversy may become more polarized, and the field may become more fragmented, to the detriment of scientific progress.
Preparation of this chapter was supported by a Natural Science and Engineering Research Council (NSERC) operating grant to Eyal M. Reingold and by a fellowship awarded to Jeffrey P. Toth by the Rotman Research Institute of Baycrest Centre and the Clarke Institute of Psychiatry. We wish to thank Elizabeth Bosman, Bob Lockhart, Colin MacLeod, Phil Merikle, Morris Moscovitch, Stephan Kohler, and especially Larry Jacoby, for their helpful comments on earlier versions of this paper.
As the chapter neared completion, it became obvious that our focus on studies by Graf and colleagues (Graf & Mandler, 1984; Graf et al., 1984) in the first part of the chapter, and on the Graf and Komatsu (1994) critique in the final part of the chapter, may be perceived as unduly personalizing the debate. We apologize for this impression, and would like to explain that the first part of the chapter was written before we became aware of the Graf and Komatsu (1994) critique. Our focus on Graf and Mandler (1984), and Graf et al. (1984) was motivated by the significance of these studies within the field of implicit memory. Similarly, our focus on Graf and Komatsu (1994) in the final part of the chapter was dictated by our strong opposition to the extreme position it advanced.
Adams, J. K. (1957). Laboratory studies of behavior without awareness. Psychological Bulletin, 54, 383-405.
Balota, D. A. (1983). Automatic semantic activation and episodic memory. Journal of Verbal Learning and Verbal Behavior, 22, 88-104.
Bowers, K. S. (1984). On being unconsciously influenced and informed. In K. S. Bowers & D. Meichenbaum (Eds.), The unconscious reconsidered (pp. 227-272). New York: Wiley.
Bowers, J. S., & Schacter, D. L. (1990). Implicit memory and test awareness. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 404-416.
Brown, A. S., & Mitchell, D. B. (in press). Levels of processing in implicit memory: A reevaluation. Memory and Cognition.
Challis, B. H., & Brodbeck, D. R. (1992). Level of processing affects priming in word-fragment completion. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 595-607.
Challis, B. H., & Tulving, E. (1993). Further support for a generation/recognition account of performance on word-fragment cued tests. Paper presented at the 34th annual meeting of The Psychonomic Society, Washington, D.C..
Cheesman, J., & Merikle, P. M. (1985). Word recognition and consciousness. In D. Besner, T. G. Waller, & G. E. MacKinnon (Eds.), Reading research: Advances in theory and practice (Vol. 5, pp. 311-352). New York: Academic Press.
Cheesman, J., & Merikle, P. M. (1986). Distinguishing conscious from unconscious perceptual processes. Canadian Journal of Psychology, 40, 343-367.
Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684.
Curran, T. & Hintzmann, D. L. (in press). Violations of the independence assumption in process dissociation. Journal of Experimental Psychology: Learning, Memory, and Cognition
Debner, J. A., & Jacoby, L. L. (1994). Unconscious perception: Attention, awareness, and control. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 304-317.
Dixon, N. F. (1971). Subliminal perception: The nature of a controversy. New York: McGraw-Hill.
Dixon, N. F. (1981). Preconscious processing. Chichester: Wiley.
Dunn, J. C., & Kirsner, K. (1988). Discovering functionally independent mental processes: The principle of reversed association. Psychological Review, 95, 91-101.
Dunn, J. C., & Kirsner, K. (1989). Implicit memory: Task or process? In S. Lewandowsky, J. C. Dunn, & K. Kirsner (Eds.), Implicit memory: Theoretical issues (pp. 17-31). Hillsdale, NJ: Erlbaum.
Erdelyi, M. H. (1985). Psychoanalysis: Freud's cognitive psychology. New York: Freeman.
Erdelyi, M. H. (1986). Experimental indeterminacies in the dissociation paradigm. Behavioral and Brain Sciences, 9, 30-31.
Eriksen, C. W. (1959). Unconscious processes. In M. R. Jones (Ed.), Nebraska symposium on motivation, 1958 (pp. 169-227). Lincoln: University of Nebraska Press.
Eriksen, C. W. (1960). Discrimination and learning without awareness: A methodological survey and evaluation. Psychological Review, 67, 279-300.
Fowler, C. A., Wolford, G., Slade, R., & Tassinary, L. (1981). Lexical access with and without awareness. Journal of Experimental Psychology: General, 110, 341-362.
Gardiner, J. M. (1988). Functional aspects of recollective experience. Memory and Cognition, 16, 309-313.
Gardiner, J. M., & Java, R. I. (1993). Recognizing and remembering. In A. Collins, M. A. Conway, S. E. Gathercole, & P. E. Morris (Eds.), Theories of Mind (pp. 163-188). Hillsdale, NJ.: Erlbaum.
Graf, P., & Komatsu, S. (1994). Process dissociation procedure: Handle with caution! European Journal of Cognitive Psychology, 6, 113-129.
Graf, P., & Mandler, G. (1984). Activation makes words more accessible, but not necessarily more retrievable. Journal of Verbal Learning and Verbal Behavior, 23, 553-568.
Graf, P., Squire, L. R., & Mandler, G. (1984). The information that amnesic patients do not forget. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 164-178.
Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. New York: Wiley.
Henley, S. H. A. (1984). Unconscious perception re-revisited: A comment on Merikle's (1982) paper. Bulletin of the Psychonomic Society, 22, 121-124.
Holender, D. (1986). Semantic activation without conscious identification in dichotic listening, parafoveal vision, and visual masking: A survey and appraisal. Behavioral and Brain Sciences, 9, 1-23.
Jacoby, L. L. (1991). A process dissociation framework: Separating automatic from intentional uses of memory. Journal of Memory and Language, 30, 513-541.
Jacoby, L. L., & Begg, I. M. (submitted). In defense of functional independence: Violations of assumptions underlying the process-dissociation procedure?. Journal of Experimental Psychology: Learning, Memory, and Cognition.
Jacoby, L. L., & Hollingshead, A. (1990). Toward a generate/recognize model of performance on direct and indirect tests of memory. Journal of Memory and Language, 29, 433-454.
Jacoby, L. L. & Kelley, C. M. (1991). Unconscious influences of memory: Dissociations and automaticity. In D. Milner & M. Rugg (Eds.), The neuropsychology of consciousness (pp. 201-233). San Diego, CA: Academic Press.
Jacoby, L. L., Kelley, C. M., & Dywan, J. (1989). Memory attributions. In H. L. Roediger & F. I. . M. Craik (Eds.), Varieties of memory and consciousness: Essays in honour of Endel Tulving (pp. 391-422). Hillsdale, NJ: Erlbaum.
Jacoby, L. L., Ste-Marie, D., & Toth, J. T. (1993). Redefining automaticity: Unconscious influences, awareness and control. In A. D. Baddeley & L. Weiskrantz (Eds.), Attention, selection, awareness and control. A tribute to Donald Broadbent (pp. 261-282). Oxford, England: Oxford University Press.
Jacoby, L. L., Toth, J. P., Lindsay, D. S., & Debner, J. A. (1992). Lectures for a layperson: Methods for revealing unconscious processes. In R. F. Bornstein & T. S. Pittman (Eds.), Perception without awareness (pp. 81-120). New York, NY: Guilford Press.
Jacoby, L. L., Toth, J. P., & Yonelinas, A. P. (1993). Separating conscious and unconscious influences of memory: Measuring recollection. Journal of Experimental Psychology: General, 122, 139-154.
Jacoby, L. L., Toth, J. P., Yonelinas, A. P., & Debner, J. A. (1994). The relationship between conscious and unconscious influences: Independence or redundancy? Journal of Experimental Psychology: General, 123, 216-219.
Jacoby, L. L., & Whitehouse, K. (1989). An illusion of memory: False recognition influenced by unconscious perception. Journal of Experimental Psychology: General, 118, 126-135.
Jacoby, L. L., Woloshyn, V., & Kelley, C. (1989). Becoming famous without being recognized: Unconscious influences of memory produced by dividing attention. Journal of Experimental Psychology: General, 118, 115-125.
Jacoby, L. L., Yonelinas, A. P., & Jennings, J. (in press). The relation between conscious and unconscious (automatic) influences: A declaration of independence. To appear in J. D. Cohen & J. W. Schooler (Eds.) Scientific approaches to the question of consciousness . Hillsdale, New Jersey: Erlbaum.
Joordens, S., & Merikle, P. M. (1992). False recognition and perception without awareness. Memory and Cognition, 20, 151-159.
Joordens, S., & Merikle, P. M. (1993). Independence or redundancy? Two models of conscious and unconscious influences. Journal of Experimental Psychology: General, 4, 462-467.
Jones, G. V. (1987). Independence and exclusivity among psychological processes: Implications for the structure of recall. Psychological Review, 94, 229-235.
Kelley, C. M., & Jacoby, L. L. (1990). The construction of subjective experience: Memory attributions. Mind & Language, 5, 49-68.
Komatsu, S., Graf, P., & Uttl, B. (in press). Process dissociation procedure: Core assumptions fail, sometimes. European Journal of Cognitive Psychology.
Marcel, A. J. (1974, July). Perception with and without awareness. Paper presented at the meeting of the Experimental Psychology Society, Stirling, Scotland.
Marcel, A. J. (1983). Conscious and unconscious perception: Experiments on visual masking and word recognition. Cognitive Psychology, 15, 197-237.
Marcel, A. J. (1986). Consciousness and processing: Choosing and testing a null hypothesis. Behavioral and Brain Sciences, 9, 40-41.
Merikle, P. M. (1982). Unconscious perception revisited. Perception & Psychophysics, 31, 298-301.
Merikle, P. M. & Joordens, S. (in press). Measuring unconscious influences. To appear in J. D. Cohen & J. W. Schooler (Eds.) Scientific approaches to the question of consciousness . Hillsdale, New Jersey: Erlbaum.
Merikle, P. M., & Reingold, E. M. (1990). Recognition and lexical decision without detection: Unconscious perception? Journal of Experimental Psychology: Human Perception and Performance, 16, 574-583.
Merikle, P. M., & Reingold, E. M. (1991). Comparing direct (explicit) and indirect (implicit) measures to study unconscious memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 224-233.
Moscovitch, M., Vriezen, E., & Goshen-Gottstein, Y. (1993). Implicit tests of memory in patients with focal lesions or degenerative brain disorders. In H. Spinnler & F. Boller (Eds.) Handbook of neuropsychology, Vol. 8 (pp. 133-173) (F. Boller & J. Grafman, Volume Editors). Amsterdam: Elsevier.
Nolan, K. A., & Caramazza, A. (1982). Unconscious perception of meaning: A failure to replicate. Bulletin of the Psychonomic Society, 20, 23-26.
Rajaram, S. & Roediger, H. L. (1993). Direct comparison of four implicit memory tests. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 765-776.
Reingold, E. M. (1992). Conscious versus unconscious processes: Are they qualitatively different? Behavioral and Brain Sciences 15, 218-219.
Reingold, E. M. & Goshen-Gottstein, Y. (submitted). Separating consciously controlled and automatic retrieval processes in memory for new associations. Journal of Experimental Psychology: Learning, Memory, and Cognition.
Reingold, E. M., & Merikle, P. M. (1988). Using direct and indirect measures to study perception without awareness. Perception & Psychophysics, 44, 563-575.
Reingold, E. M., & Merikle, P. M. (1990). On the inter-relatedness of theory and measurement in the study of unconscious processes. Mind & Language, 5, 9-28.
Reingold E. M., & Merikle, P. M. (1991). Stem completion and cued recall: The role of response bias. Paper presented at the 32nd annual meeting of the Psychonomic Society, San Francisco, CA.
Richardson-Klavehn, A., & Bjork, R. A. (1988). Measures of memory. Annual Review of Psychology, 39, 475-543.
Richardson-Klavehn, A, Gardiner, J. M., & Java, R. I. (1994). Involuntary conscious memory and the method of opposition. Memory, 2, 1-29.
Roediger, H. L. (1990). Implicit memory: A commentary. Bulletin of the Psychonomic Society, 28, 373-380.
Roediger, H. L., & McDermott, K. B. (1993). Implicit memory in normal human subjects. In H. Spinnler & F. Boller (Eds.), Handbook of Neuropsychology, Vol. 8 (pp. 63-131). Amsterdam: Elsevier.
Roediger, H. L., Weldon, M. S., Stadler, M. L., & Riegler, G. L. (1992). Direct comparison of two implicit memory tests: Word fragment and word stem completion. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 1251-1269.
Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518.
Schacter, D. L., Bowers, J., & Booker, J. (1989). Intention, awareness, and implicit memory: The retrieval intentionality criterion. In S. Lewandowsky, J. C. Dunn, & K. Kirsner (Eds.), Implicit memory: Theoretical issues (pp. 47-65). Hillsdale, NJ: Erlbaum.
Shevrin, H., & Dickman, S., (1980). The psychological unconscious: A necessary assumption for all psychological theory? American Psychologist, 35, 421-434.
Shimamura, A. P. (1986). Priming effects in amnesia: Evidence for a dissociable memory function. Quarterly Journal of Experimental Psychology, 38A, 619-644.
Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643-661.
Swets, J. A. (1964). Signal detection and recognition by human observers. New York: Wiley.
Toth, J. P., Lindsay, D. S., & Jacoby, L. L. (1992). Awareness, automaticity, and memory dissociations. In L. R. Squire & N. Butters (Eds.), Neuropsychology of memory, 2nd Ed. (pp. 46-57).
Toth, J. P., Reingold, E. M., & Jacoby, L. L. (1994). Toward a redefinition of implicit memory: Process dissociations following elaborative processing and self-generation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 290-303.
Toth, J. P., Reingold, E. M., & Jacoby, L. L. (in press). A response to Graf and Komatsu's (1994) critique of the process-dissociation procedure: When is caution necessary? European Journal of Cognitive Psychology.
Tulving, E. (1985). Memory and consciousness. Canadian Psychology, 26, 1-12.
Tulving, E. (1989). Memory: Performance, knowledge, and experience. European Journal of Cognitive Psychology, 1, 3-26.
Yonelinas, A. P. (1994). Receiver-operating characteristics in recognition memory: Evidence for a dual-process model. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 1341-1354.