Recognition and Lexical Decision Without Detection: Unconscious Perception?

Philip M. Merikle and Eyal M. Reingold

Abstract

Stimulus detection is often assumed to index conscious perceptual experience. The adequacy of this assumption was evaluated in a series of experiments in which stimulus detection was compared to concurrent forced-choice measures of either stimulus recognition or lexical status. In each experiment, the target stimuli were drawn from large sets of materials containing either words or nonwords, and on each trial, either a target stimulus or a blank field was presented. Following each presentation, subjects first decided whether a target stimulus had been presented and then made a recognition decision. When the target stimuli were familiar words, correct recognition judgements and correct lexical decisions were made even when subjects failed to detect the presence of a stimulus. However, when the target stimuli were nonwords, neither the recognition nor the lexical decision tasks indicated any sensitivity for stimuli that subjects failed to detect. These qualitatively different patterns of results for word and nonword stimuli provide preliminary converging evidence consistent with the assumption that stimulus detection is an adequate measure of conscious awareness.

Recognition and Lexical Decision Without Detection: Unconscious Perception?

The most widely followed approach for studying perception without awareness involves the dissociation paradigm. Although various versions of this paradigm have been used (see Adams, 1957 for a review of many early studies), the basic assumption underlying all versions is that perception without awareness can be demonstrated via a dissociation between two indices of perception. For example, in one frequently employed version of the paradigm (see Erdelyi, 1986; Holender, 1986), it is assumed that perception without awareness is demonstrated whenever a measure of conscious perception indicates null sensitivity to a stimulus and a second measure of perception indicates that the stimulus was nevertheless perceived. In spite of this relatively straightforward logic, a number of unresolved methodological and conceptual issues have made applications of the dissociation paradigm to the study of perception without awareness a source of continuing controversy (see Eriksen, 1960; Holender, 1986; Reingold & Merikle, 1988).

One particularly problematic aspect of many applications of the dissociation paradigm is the lack of any agreement as to what constitutes an adequate measure of conscious awareness. In the absence of any explicit theory or data to guide the selection of measures, the general approach has been to select a direct measure of perception that is "intuitively" reasonable. Typical measures used in recent studies include forced-choice, presence-absence decisions (e.g., Balota, 1983; Fowler, Wolford, Slade & Tassinary, 1981; Groeger, 1984, 1986; Kemp-Wheeler & Hill, 1988; Marcel, 1980, 1983a), forced-choice discriminations among a small number of stimulus alternatives (e.g., Avant & Thieman, 1985; Cheesman & Merikle, 1984, 1986; Purcell, Stewart & Stanovich, 1983), stimulus identification (e.g., Hines, Czerwinski, Sawyer & Dwyer, 1986; McCauley, Parmelee, Sperber & Carr, 1980), and forced-choice spatial or temporal localization decisions (e.g., Greenwald, Klinger & Liu, 1989; Groeger, 1988). What these measures have in common is that they all provide a direct index of perceptual processing in that subjects are explicitly instructed to attend to a certain aspect of the stimulus and to respond accordingly. However, it is very improbable that all of these measures provide either adequate or even comparable indices of conscious perceptual experience.

Of all the measures that have been used, the measure often assumed to provide the most satisfactory index of conscious perception is stimulus detection. The appeal of stimulus detection is the minimal stimulus discrimination required for a correct response; even the slightest perceivable stimulus change should lead to a correct detection decision. Thus, if insufficient information is perceived to even indicate the presence of a stimulus, one reasonable conclusion is that the observer experienced no subjective phenomenal awareness attributable to the presentation of the stimulus (see Marcel, 1983a, 1983b).

If stimulus detection provides an adequate measure of conscious awareness, then any dissociation between stimulus detection and another measure of perception is sufficient to demonstrate perception without awareness. Although stimulus detection is frequently compared to indirect measures of perception (e.g., priming), the logic of the dissociation paradigm implies that comparisons between stimulus detection and other direct measures of perception may also provide a basis for establishing perception without awareness. In fact, Marcel (1983a, Experiment 1) made such comparisons and found that observers correctly recognized response alternatives that were either orthographically or semantically similar to target words even when their detection responses indicated no sensitivity to the presence of the words. If it is assumed that an absence of stimulus detection reflects an absence of conscious awareness, then these results suggest that the perceptual processes responsible for both orthographic and semantic decisions can occur without awareness.

Two methodological considerations, however, raise serious questions concerning Marcel's experiment. First, Fowler et al. (1981) demonstrated that task differences may account for these apparent dissociations. These investigators conducted a pseudoexperiment in which only the response alternatives were presented, and each subject's responses were evaluated in terms of the actual stimulus word that should have been presented prior to each set of response alternatives. The results indicated that both semantic and orthographic decisions were better than chance. These results suggest that the dissociations between detection and recognition reported by Marcel may simply reflect the influence of the different retrieval environments across tasks; recognition decisions were made in the context of stimulus alternatives, whereas detection decisions were made in the absence of any stimulus alternatives.

An even more serious problem with Marcel's experiment concerns whether his methodology was adequate to establish that the measure of stimulus detection indicated a true null sensitivity to the presence of the stimuli. Null sensitivity is extremely difficult to demonstrate for any measure of perception (see Macmillan, 1986), and critiques of Marcel's experiments suggest that his methodology was probably inadequate for establishing null stimulus detection (e.g., Cheesman & Merikle, 1985, Merikle, 1982; Nolan & Caramazza, 1982). Given the logic of his approach, null stimulus detection must be demonstrated convincingly before any dissociation between stimulus detection and another measure of perception can provide any compelling evidence for perception without awareness. Otherwise, it is always possible to argue that the subjects were aware of the stimuli (i.e., could detect the stimuli), at least on some proportion of the trials.

Although Marcel's results are inconclusive, other studies involving concurrent measures of stimulus detection and recognition suggest that these measures are in fact dissociable indices of perception. In a classic psychophysical study, Lindner (1968) measured both detection and recognition for auditory stimuli. There were three different stimulus states: two different tones and a null stimulus. On each trial, either one of the tones or the null stimulus was presented, and the subjects were required first to decide whether or not a tone had been presented (i.e., stimulus detection) and second to indicate which tone had been presented (i.e., stimulus recognition). The important empirical result was that stimulus recognition was better than chance even on those trials when the subjects failed to detect (i.e., "missed") the presence of one of the tones. Thus, Lindner's results provide a clear indication that stimulus recognition can occur in the absence of stimulus detection.

Similar results have been reported by other investigators. For example, Rollman and Nachmias (1972) found that stimulus recognition also occurs in the absence of stimulus detection when visual stimuli are used. In their study, the tasks for the subjects were to decide whether one of two chromatic flashes (red or green) or a null stimulus had been presented and then to indicate which color had been presented. The results were very similar to Lindner's (1968) findings in that subjects were able to recognize the correct color at a better than chance level of accuracy even when they failed to detect its presence (i.e., said "no"). Taken together, the results reported by both Lindner (1968) and Rollman and Nachmias (1972), as well as other similar findings (Diener, 1981; Parasuraman, Richer & Beatty, 1982), suggest that, at least under certain conditions, stimulus recognition can occur even when observers fail to detect the presence of a stimulus.

If one accepts the assumption that a failure to detect a stimulus indicates an absence of awareness, then the results of these psychophysical investigations provide strong evidence for perception without awareness. Given the intuitive appeal of stimulus detection as a measure of awareness, an interesting question which then arises is whether similar dissociations can be found using the large sets of verbal materials which characterize many studies investigating perception without awareness (e.g., Marcel, 1983a). With the small stimulus sets used in the psychophysical studies (e.g., Lindner, 1968; Rollman & Nachmias, 1972), familiarity with the stimulus set developed during the experimental sessions may have made it possible to base both detection and recognition decisions on minimal, low level, stimulus information. With large stimulus sets, on the other hand, it is much more probable that stimulus recognition would require a more extensive perceptual analysis than stimulus detection. Dissociations between stimulus detection and recognition with large stimulus sets would thus provide very compelling demonstrations of perception in the absence of detection.

Even though stimulus detection has considerable intuitive appeal as a measure of awareness, demonstrations that recognition occurs in the absence of detection do not necessarily indicate that this recognition was mediated by unconscious perceptual processes. Such findings only support perception without awareness, if it is assumed that stimulus detection provides an adequate measure of awareness (see Reingold & Merikle, 1988). If this assumption is not made, then any dissociation between a measure of stimulus detection and a measure of stimulus recognition may only indicate that each measure is maximally sensitive to a different aspect of the perceptual process, which may be conscious, unconscious, or both conscious and unconscious. What is needed to support the assumption that stimulus detection provides an adequate index of conscious perceptual experience is some converging evidence indicating that the "detect" and the "nondetect" states are qualitatively different. Without converging evidence, any dissociation between detection and another measure of perception can never provide strong evidence for or against perception without awareness.

Given these considerations, the purpose of the present experiments was to explore the possible adequacy of stimulus detection as an measure of awareness. To meet this objective, concurrent measures of stimulus detection and recognition were used under conditions involving large stimulus sets of verbal material. Concurrent measurement of stimulus detection and recognition bypasses many of the problems associated with establishing a threshold for null stimulus detection, which may in fact be an almost impossible objective to obtain (see Macmillan, 1986). With concurrent measures, the critical observations involve a straightforward analysis of stimulus recognition conditional on stimulus detection. Dissociations are demonstrated if correct stimulus recognition exceeds a chance level of performance following both hits and misses. Alternatively, if correct recognition only exceeds a chance level of performance following hits, then the measures do not provide dissociable indices of perception.

In the present experiments, the target stimuli were drawn from large sets of materials containing either words (Experiments 1 and 2), nonwords (Experiment 3), or both words and nonwords (Experiment 4), and no stimulus was ever presented more than once to any subject. In each experiment, subjects initially made a detection decision and then either a forced-choice recognition judgement (Experiments 1, 2 and 3) or a lexical decision (Experiment 4). The detection decision was always made first to ensure that stimulus detection was measured under optimal conditions. If forgetting occurred following stimulus presentation, this task order ensured that any bias in sensitivity across measures favored stimulus detection rather than stimulus recognition or lexical decision.

Two critical questions were addressed by these experiments. First, are stimulus detection and recognition, as well as stimulus detection and lexical decision dissociable measures of perception when large sets of verbal materials are used? If these measures are dissociable under the conditions of the present experiments, it would indicate that dissociations between stimulus detection and other direct measures of perception reflect a very general phenomenon. The second question concerned possible boundary conditions for observing dissociations between measures. It is for this reason that both words and nonwords were used as stimuli. Perception without awareness should only occur for familiar stimuli which have a preexisting memory representations. Therefore, words should be more likely to lead to perception without awareness than the unfamiliar letter sequences of nonwords. If dissociations between stimulus detection and recognition are only observed when the target stimuli are words, then such findings would provide important converging evidence that stimulus detection provides an adequate measure of perceptual awareness.

Experiment 1

In this experiment, stimulus detection and two-alternative forced-choice recognition were compared. The general approach was similar to that followed in previous studies involving the concurrent measurement of detection and recognition (e.g., Lindner, 1968; Rollman & Nachmias, 1972). On each trial, either a word or a blank visual field was presented. Following each presentation, the subjects first indicated whether or not a stimulus had been presented. Once this detection decision had been made, the subjects were then shown two words and were required to select the word that "may" have been presented on that trial.

Apparatus. The stimulus materials were displayed on an Electrohome color monitor that was interfaced to an Apple II Plus microcomputer via an Electrohome Supercolor board. The monitor was viewed through a hood that physically divided the screen into separate left-eye and right-eye fields; fusion of these fields was aided by viewing the display through a set of rotating prisms. The viewing distance was 65 cm and the luminance of each field measured 32 cd/m2 when the light beige background color (Color No. 91) was displayed.

Two button boxes were placed in front of the subjects. The button box for the right hand contained a start key to initiate display presentation and two buttons to indicate whether or not a target word was detected. The button box for the left hand was used for the recognition task; it contained two keys to indicate whether the upper or lower word was the most likely one to have been presented.

Stimulus Materials. A pool of words containing 1200 nouns was compiled for the experimental trials. These words were selected from the Kucera and Francis (1967) norms, and they consisted of the 1200 most frequent nouns between 3 and 11 letters in length. The frequency of the selected nouns ranged from 847/million (i.e., "people") to 25/million (e.g., "adult").

All nouns of each length were stored in a separate computer file. Prior to testing each subject, the words in each file were paired randomly, and one randomly-selected member of each pair was designated as the "target." In addition, half the word pairs for each word length were assigned on a random basis to "present" trials, and half the "target" words in the word pairs for each word length were assigned randomly to be the "upper" word when the entire word pair was presented on the recognition trials. Finally, the presentation order for the resulting 600 word pairs was randomly determined for each subject prior to the beginning of the first experimental session.

An additional 200 words were selected from the Kucera and Francis norms for use on the practice trials which preceded the experimental trials. These words were also 3 to 11 letter nouns and they ranged in frequency from 21/million to 26/million. For each subject, the 200 practice words were presented in a different, randomly-determined order.

Each letter in each word when presented on the monitor was approximately 6 mm (0.6o)1 high and 5 mm (0.5o) wide. Considering the spacing between the letters in the words, the length of the words varied from 17 mm (1.5o) for the three-letter words to 62 mm (5.5o) for the eleven-letter words.

Procedure. Each subject was tested individually in two separate sessions administered on two consecutive days. The first session lasted approximately 1.5 hours, and the second session took approximately 45 minutes to complete.

At the beginning of the first session, 200 practice trials were given to familiarize each subject with the detection task. There were five blocks of 40 practice trials, and in each block of trials, a word was presented on 20 randomly-selected trials while a blank field was presented on the remaining trials. In addition, the stimulus onset asynchrony (SOA) separating the target field from the subsequently presented backward mask was decreased systematically across successive blocks of trials according to the following sequence: 267, 167, 100, 67, and 50 msec. The task for the subjects on each trial was simply to decide whether a word or a blank field had been presented.

The sequence of events on each practice trial began with the subjects viewing a fixation pattern which consisted of two dotted, horizontal lines centered within the visual field. This fixation pattern was presented to both eyes, and it remained in view during the entire trial sequence. Each line in the fixation pattern was 74 mm (6.6o) in length, and the vertical separation between the lines was 18 mm (1.6o). To initiate each trial, subjects pressed the start key located on the right-hand button box. Immediately following trial initiation, a forward mask was presented. The forward mask consisted of a different random ordering of the letters A, B, E, G, H, M, N, Q, R, S, and W on each trial. It was presented to the right eye for 50 msec, and it was centered within the area defined by the fixation lines. At the offset of the forward mask, there was a 16.7-msec delay prior to the presentation of the target field to the left eye for 16.7 msec. On half of the trials, the target field was blank, while on the remaining trials a randomly selected word was presented. Following presentation of the target field, a backward mask, which was constructed in the same manner as the forward mask, was presented to the right eye for 50 msec. At the end of each trial sequence, subjects indicated their decision concerning the presence or absence of a word in the target field by pressing the key located on either the right side ("present") or the left side ("absent") of the button box placed under the right hand.

The sequence of events for the 300 experimental trials administered in each session was similar to the event sequence on the practice trials, with the following two exceptions. First, a 50-msec SOA separated the onset of target field from the onset of the backward mask on all trials. Second, immediately following each detection decision, a pair of words was presented. Each word pair was presented to both eyes, and the words were positioned in the visual field such that each word was located either 14 mm (1.3o) above or below the center of the field. The task for the subjects was to decide which of the words may have been presented on that trial. The subjects indicated each decision by pressing either the upper or the lower key on the button box placed under the left hand.

Subjects. The twenty subjects were either undergraduate or graduate students at the University of Waterloo. The subjects were volunteers who participated with the understanding that they would receive $10 at the completion of the two days of testing. All subjects had normal or corrected to normal vision.

Detection. Mean overall detection performance is shown in Table 1. In addition to hits and false alarms, measures of perceptual sensitivity, dl, and criterion or bias, Cl, were computed for each subject. These particular measures of sensitivity and bias were chosen on the basis of their demonstrated independence (see Snodgrass & Corwin, 1988). The data clearly indicate that the subjects were able to discriminate between words and blank fields, and an inspection of the individual data for each subject revealed that all subjects had more hits than false alarms.

-------------------------------

In draft, Table 1 would appear here

-------------------------------

Recognition. The critical question addressed by this experiment concerns stimulus recognition conditional on hits and misses in the detection task. These conditional recognition scores are shown in Figure 1. Although recognition was better following hits than misses, t (19) = 4.84, p < .01, the important aspect of the data is that correct recognition was significantly greater than chance (i.e., 0.50) following both hits, t (19) = 7.25, p < .01, and misses, t (19) = 3.69, p < .01. Thus, even when the subjects claimed that no stimulus was presented, as indicated by a miss, sufficient information was nevertheless perceived on at least some proportion of these trials to guide subsequent recognition judgements.2 These results are completely consistent with previous findings (e.g., Lindner, 1968; Rollman & Nachmias, 1972) and they demonstrate that stimulus detection and recognition are dissociable measures of perception even when large sets of verbal materials are used.

-------------------------------

In draft, Figure 1 would appear here

-------------------------------

Although the present results suggest that perception in the absence of detection is a robust phenomenon, the dissociation between tasks may have occurred for another reason. In addition to requiring different stimulus discriminations, the detection and recognition tasks also differed in terms of retrieval environments. For the detection task, no stimulus information was available when the subjects made their decisions, whereas recognition judgements were made within the context provided by the two stimulus alternatives. Thus, it is entirely possible that this difference in the retrieval environments across tasks accounts for the observed dissociation between stimulus detection and recognition.

Experiment 2

The purpose of this experiment was to determine whether stimulus detection and stimulus recognition are dissociable measures of perception even when the retrieval environments are matched as closely as possible across tasks. In the experiment, the retrieval environments were matched by presenting the same two stimulus alternatives when subjects made their detection and recognition decisions. The only difference distinguishing the tasks was the question posed to the subjects. For the detection task, subjects were asked to decide if "either" stimulus alternative had just been presented, whereas for the recognition task, subjects were asked to decide "which" stimulus alternative had just been presented.

Apparatus and Stimulus Materials. With two exceptions, the apparatus and the stimulus materials were the same as in Experiment 1.

One difference between experiments was that touch-sensitive plates were used by the subjects in the present experiment to initiate display presentation and to indicate their detection and recognition decisions. The physical arrangement of these plates was similar to the arrangement of the button boxes in Experiment 1.

The second difference between experiments was that 240 rather than 200 words were selected for use on the practice trials. These practice words were paired randomly to form 120 pairs, with each pair containing two words of equal length. Pairs of practice words were required, as each subject was always shown two words following the presentation of each target word. Each pair of practice words was assigned randomly to two of six blocks of 40 practice trials, and each time a given pair was used, a different word was designated as the "target."

Procedure. As in Experiment 1, each subject was tested individually in two separate sessions administered on two consecutive days.

At the beginning of the first session, either 200 or 240 practice trials were presented to familiarize the subjects with the detection task. For each subject, the SOA between the target field and the backward mask was decreased systematically across successive blocks of trials. Subjects who had difficulty initially making presence/absence decisions at the longest SOA (i.e., 267 msec) were given two blocks of 40 trials at this SOA. Otherwise, all subjects received an identical sequence of decreasing SOAs (i.e., 267, 167, 100,67, and 50 msec) across successive blocks of practice trials. Within each block of 40 practice trials, a word was presented on 20 randomly-selected trials and a blank field was presented on the remaining trials.

The sequence of events on each practice trial was similar to Experiment 1. However, rather than simply indicating a detection decision following the presentation of each target field, the subjects were shown two words and required to decide if either word had been presented. The physical arrangement of the words was the same as used for the recognition task in Experiment 1. In addition, presentation of the words was accompanied by the simultaneous presentation at the top of the CRT display of the question "Was either word presented?" The subjects indicated their decision on each trial by touching the appropriate touch-sensitive plate located near their right hand.

The only procedural difference distinguishing the practice trials from the 300 experimental trials administered in each session was that on the experimental trials, the subjects were required to make a recognition decision following each detection decision. Once each detection decision was made, the same pair of words was presented again, but the question at the top of the CRT display was changed to "Which word was presented?" The subjects indicated each recognition decision by touching either the upper or lower touch-sensitive plate located near their left hand.

Subjects. The 24 subjects were undergraduate students. All subjects had normal or corrected to normal vision and were paid $15 for their participation.

Results and Discussion

Detection. Mean overall detection performance is shown in Table 1. As in Experiment 1, an inspection of the individual data indicated that all subjects had more hits than false alarms.

Recognition. The conditional recognition scores are shown in Figure 1. The overall pattern of results was similar to the one found in Experiment 1; correct recognition was higher following hits than misses, t (23) = 7.26, p < .01, and significantly better than chance following both hits, t (23) = 7.93, p < .01, and misses, t (23) = 3.07, p < .01. These results thus demonstrate that stimulus recognition can occur following a miss on a detection task even when the retrieval environments for the two tasks are matched as closely as possible.3 In addition, even though the data presented in Figure 1 suggest that the dissociation between detection and recognition may have been smaller in this experiment than in Experiment 1, an analysis of variance based on the data from both experiments indicated that the pattern of recognition scores following hits and misses did not interact with experiment, F (1,42) = 1.64. Given these findings, it appears that stimulus detection and recognition are dissociable measures of perception both with and without matched retrieval environments.

Experiment 3

The important question raised by the dissociations demonstrated in Experiments 1 and 2 is what mediates correct recognition in the absence of stimulus detection? The interesting answer, suggested by Marcel's (1983a) original studies, is that these dissociations indicate that familiar stimuli, such as words, may be perceived and discriminated even when there is insufficient information for stimulus detection. Alternatively, the uninteresting answer is that these dissociations are an artifact of the experimental procedures. For example, if the different discriminations required by the detection and recognition tasks lead to uncorrelated errors (see Eriksen, 1960), then the observed dissociations between tasks may simply reflect these different sources of experimental error.

The purpose of the present experiment was to establish whether dissociations between stimulus detection and stimulus recognition are inevitably observed whenever these two tasks are compared or whether dissociations between these tasks may be limited to particular conditions. The experiment was very similar to Experiment 2, except that the stimuli were nonwords rather than words. If the dissociations observed in Experiments 1 and 2 simply reflect an aspect of general methodology, then a similar dissociation should be observed in the present experiment. On the other hand, if the correct recognition in the absence of detection observed in Experiments 1 and 2 only occurs when the stimuli are familiar, then no dissociation between the detection and recognition of nonwords should be observed. Rather, the two measures should be highly correlated, as indicated by significantly better than chance recognition following hits but chance level recognition following misses.

General. With one exception, the general methodology was the same as used in Experiment 2. The major difference between the two experiments was the stimulus set. In the present experiment, all stimuli were nonwords. These nonwords were generated by randomly ordering the letters in each of the words used in Experiment 2. The only constraint on the randomization process was that the resulting ordering of letters must not spell an English word.

Subjects. The 24 subjects were undergraduate students at the University of Waterloo. All subjects had normal or corrected to normal vision and were paid $15 for their participation.

Detection. Mean overall detection performance is shown in Table 1. As in Experiments 1 and 2, all subjects had more hits than false alarms. However, a comparison of the present results with detection performance in Experiment 2 indicates that overall perceptual sensitivity was somewhat lower in the present experiment. This pattern of results across comparable experiments suggests that words may be more readily detected than nonwords. Doyle and Leach (1988) recently reported a similar advantage for the detection of words over nonwords.

Recognition. The conditional recognition scores are shown in Figure 1. The overall pattern of results is quite dissimilar to the results found in Experiment 2. Although correct recognition exceeded the chance level of performance following hits, t (23) = 5.00, p < .01, correct recognition following misses (.502), in contrast to the results of Experiment 2, closely approximated the chance level of performance. Thus, the present results demonstrate that stimulus recognition does not necessarily occur following a miss on a detection task. In other words, when the stimulus set consists of nonwords, stimulus detection and stimulus recognition are not dissociable measures of perception .

Given that stimulus detection was lower in the present experiment than in Experiment 2, it was important to determine if the chance level recognition performance following misses was due simply to this lower overall perceptual sensitivity. To evaluate this possibility, the 24 subjects were divided into two groups of 12 subjects on the basis of overall detection performance. Mean perceptual sensitivity was considerably greater for the 12 subjects in the "high" group (dl = 2.569) than for the 12 subjects in the "low" group (dl = 0.628). However, despite this rather large difference in perceptual sensitivity across the two groups, stimulus recognition following misses on the detection task approximated a chance level of performance for the subjects in both the "high" (.499) and the "low" (.505) groups. Thus, independent of the overall level of perceptual sensitivity, nonwords were not recognized following misses on the detection task.

The absence of any dissociation between the detection and recognition of nonword stimuli in the present experiment stands in marked contrast to the dissociations observed between these same measures in Experiments 1 and 2 for stimulus sets consisting of words. Taken together, the results of these three experiments indicate that dissociations between stimulus detection and recognition are not inevitable. Rather, whether or not dissociations are observed depends upon the characteristics of the stimulus materials. Given the present results for nonword stimuli, it is very improbable that the dissociations between the detection and recognition of words observed in Experiments 1 and 2 are due to some unidentified methodological artifact. To the contrary, the present results suggest that perception in the absence of detection may only occur when familiar stimuli are presented.

Experiment 4

In this experiment, the generality of the previous results was evaluated under somewhat different experimental conditions. The major difference between the present experiment and the previous studies was the recognition decision required on each trial. Following each detection decision, the subjects were required to decide whether the stimulus was a word or a nonword. Given that a lexical decision requires a more abstract categorization of a letter string than is necessarily required for any simple discrimination between alternative stimuli, a dissociation between stimulus detection and lexical decision would provide more compelling evidence that familiar stimuli are perceived even when they are not detected.

Although the experiment was similar in many ways to the previous experiments, two aspects of the previous procedures were changed to implement the lexical decision task. First, rather than two possible stimulus states, there were three possible stimulus states on each trial: words, nonwords, and blank fields. This change in procedure allowed assessments of each subject's sensitivity to word and nonword stimuli following both hits and misses. Thus, it was possible to evaluate within a single experiment whether recognition in the absence of detection is limited to familiar stimuli. The second procedural change concerned the retrieval environments for the detection and lexical decision tasks. Across tasks, the retrieval environments were matched in that no stimulus alternatives were presented when subjects performed either task.

With these procedural changes, the critical aspects of findings observed in the previous experiments were evaluated under conditions involving a different recognition decision and a different retrieval environment. If the previous findings reflect a general phenomena, then comparable patterns of results should be found in the present study.

Apparatus. The apparatus was the same as used in Experiment 1.

Stimulus Materials. The word pool used in Experiments 1 and 2 was culled, and 600 words, varying in length from 3 to 10 letters, were selected. The frequency of the selected words ranged from 847/million to 57/million, according to the Kucera and Francis (1967) norms.

Each of the 600 words used in the experiment had a nonword variant. Each nonword variant was generated by randomizing the order of the letters in each word. The only constraint on this randomization procedure was an inspection of the nonwords to ensure that no words were created.

Finally, the same pool of 200 practice words used in Experiment 1 was used again in the present experiment.

Procedure. The general procedure was similar to the one followed in the previous experiments. Prior to the beginning of the experimental trials, 200 practice trials were given, and the SOA between the target field and the backward mask was systematically decreased across successive trial blocks. Following the practice trials, 600 experimental trials were administered in the first session and an additional 600 experimental trials were administered in the second session. On all experimental trials, the SOA between the target field and the backward mask was 50 msec.

The experimental trials in each session were presented in six blocks of 100 trials. In each trial block, a blank field was presented on 50 trials, a word was presented on 25 trials, and a nonword was presented on the remaining 25 trials. The ordering of each trial type within each trial block was completely random. Likewise, the order in which words or nonwords were selected for presentation on each trial was based on a different random ordering of the 600 word/nonword pairs determined for each subject prior to the beginning of the experiment.

The major difference in procedure distinguishing the present experiment from the previous experiments was the recognition decision required on each trial. Following each detection decision, the subjects were required to decide whether a word or a nonword had been presented. The subjects indicated each word/nonword decision by pressing either the upper ("word") or lower ("nonword") key on the button box placed under their left hand.

Subjects. The 24 subjects were undergraduate students who were paid $10 at the completion of the experiment. All subjects had normal or corrected to normal vision.

Detection. Mean detection performance for the words and nonwords is shown in Table 1. An inspection of the individual data for each subject revealed that all subjects had more hits than false alarms for both words and nonwords.

The data presented in Table 1 indicate that the subjects were more sensitive to the presence of words than to the presence of nonwords. Inspection of the individual data revealed that 19/24 subjects had a higher hit rate for words than for nonwords, and a statistical comparison of the sensitivity indices (i.e., dl) for words and nonwords indicated that this small overall difference in perceptual sensitivity was significant, t (23) = 3.91, p < .01. This influence of lexical status on stimulus detection confirms the previous comparison made across the results of Experiments 2 and 3, and it is consistent with other recent findings indicating that stimulus detection is influenced by overall stimulus familiarity (Doyle & Leach, 1988; Purcell & Stewart, 1988). What all of these findings suggest is that it may be impossible to ever have a "pure" measure of detection that is completely unaffected by stimulus familiarity.

Lexical Decision. Table 2 shows the recognition scores conditional on hits and misses in the detection task for each stimulus condition. Each score reflects the proportion of trials on which the subjects decided that the stimulus was a "word." Given the nature of the lexical-decision task, chance performance was not necessarily equivalent to 0.50, as any bias to respond "word" or "nonword" would affect the actual chance performance. For this reason, correct recognition following both hits and misses was expressed in terms of A'.4 This nonparametric measure of recognition memory provides a good estimate of forced-choice performance (see Snodgrass & Corwin, 1988), and it allowed lexical decision accuracy in this experiment to be expressed on a scale equivalent to the proportion correct measure used in the previous experiments. To compute these A' measures, the proportions shown in Table 2 for word stimuli were considered hits, and the proportions for nonword stimuli were considered false alarms.

-------------------------------

In draft, Table 2 would appear here

-------------------------------

Figure 2 shows mean lexical decision accuracy (i.e., word vs. nonword discrimination) expressed in terms of A' conditional on hits and misses. The important aspect of these data is that the overall pattern of results is very similar to the ones found in Experiments 1 and 2. Although lexical decisions were more accurate following hits than misses, t (23) = 3.58, p < .01, lexical decisions were significantly better than chance following both hits, t (23) = 6.64, p < .01, and misses, t (23) = 2.33, p < .05. These results demonstrate that stimulus detection and lexical decision are dissociable measures of perception.5 More importantly, the results are completely consistent with the results of Experiments 1 and 2. Taken together, the results of these three experiments provide convincing evidence that a failure to detect a stimulus does not necessarily imply that another direct measure of perception requiring a different discrimination will also indicate that the stimulus was not perceived.

-------------------------------

In draft, Figure 2 would appear here

-------------------------------

Word and Nonword Discrimination. The results of the previous experiments indicated that stimulus recognition in the absence of stimulus detection occurs when the stimuli are words (Experiments 1 and 2) but not when the stimuli are nonwords (Experiment 3). To evaluate whether similar patterns of findings across word and nonword stimuli occurred in the present experiment, A' measures were computed to determine each subject's sensitivity to words and nonwords following both hits and misses. These measures were computed by considering the proportions shown in Table 2 for word and nonword stimuli as hits and the comparable proportions shown in the Table for the blank fields as false alarms.

The mean A' scores indicating sensitivity to word and nonword stimuli following both hits and misses on the detection task are shown in Figure 2. The important aspect of these results is that overall pattern is consistent with the critical findings obtained in the previous experiments. For words, sensitivity exceeded a chance level of performance following both hits, t (23) = 7.00, p < .01, and misses, t (23) = 2.85, p < .05, on the detection task. Thus, as found in Experiments 1 and 2, subjects were sensitive to words even when they failed to detect their presence. However, for nonwords, sensitivity only exceeded a chance level of performance following hits, t (23) = 3.33, p < .01, and approximated the chance level of performance following misses (.507). This pattern of results is completely consistent with the one found in Experiment 3 and indicates that the subjects were insensitive to the nonwords they failed to detect. Taken together, these findings for word and nonword stimuli indicate that in the absence of stimulus detection, subjects may perceive familiar stimuli but they are completely insensitive to unfamiliar stimuli.

General Discussion

The present results provide compelling evidence that a failure to detect a word does not necessary indicate that insufficient information was perceived to guide another stimulus discrimination based on a different characteristic of the word. This conclusion follows directly from the dissociations observed between both detection and recognition (Experiments 1 and 2) and detection and lexical decision (Experiment 4) when the stimuli consisted of words. Thus, at the empirical level, the present results are consistent with Marcel's (1983a, Experiment 1) earlier findings suggesting that stimulus recognition can occur in the absence stimulus detection. In addition, the present results are also consistent with the results of previous psychophysical studies based on small, non-verbal stimulus sets (e.g., Lindner, 1968; Rollman & Nachmias, 1973), Taken together, the present results and these previous findings indicate that dissociations between stimulus detection and other direct measures of perception reflect a very general phenomenon.

The present results also demonstrate an important boundary condition for observing dissociations between stimulus detection and other direct measures of perception. In contrast to the dissociations observed when the stimuli were words, the results of Experiments 3 and 4 indicate that subjects did not perceive nonword stimuli they failed to detect. In other words, subjects were completely insensitive to nonword stimuli when their detection decisions indicated that no stimulus had been presented. This absence of any sensitivity to nondetected, nonword stimuli suggests that the dissociations between stimulus detection and recognition for familiar words were not due to some artifact of the general methodology. More importantly, the different patterns of results for word and nonword stimuli indicate that stimulus familiarity was a critical factor mediating perception in the absence of detection. Thus, when a stimulus was not detected, only familiar stimuli with preexisting memory representations were recognized, whereas when a stimulus was detected, sufficient information was perceived to recognize both familiar and unfamiliar stimuli.

In previous studies, perception in the absence of stimulus detection has been interpreted as support for unconscious perception (e.g., Balota, 1983; Fowler et al., 1981; Groeger, 1984, 1986; Kemp-Wheeler & Hill, 1988; Marcel, 1980, 1983a). These studies were based on the intuitive notion that a failure to detect a stimulus indicates a complete absence of conscious perceptual experience. Thus, the implicit a priori assumption underlying these studies was that stimulus detection provides a completely adequate measure of conscious awareness If this assumption is accepted, then the present results provide overwhelming support for perception without awareness. With this assumption, the present studies provide convincing demonstrations that sufficient information is perceived without awareness (i.e., without detection) to allow an observer both to discriminate between alternative stimuli and to classify a stimulus as a word.

However, even though stimulus detection has considerable intuitive appeal as a measure of awareness, dissociations between stimulus detection and other measures of perception do not necessary support unconscious perception. In fact, given that detection, recognition, and lexical decision necessarily require different stimulus discriminations based on different task relevant information (see Duncan, 1985; Navon, 1986), there are at least two reasonable alternative interpretations of the dissociations observed in the present studies. First, each task may measure a different characteristic of consciously available information (see Erdelyi, 1986; Eriksen, 1960). From this perspective, the dissociations observed in the present studies may simply indicate a partial independence in conscious awareness for the different aspects of a stimulus. A second possible interpretation is based on a consideration of possible differences across tasks in bias or criterion. If criterion placement varied independently for each task, then dissociations between tasks would be expected (see Macmillan, 1986 for a related discussion).

Fortunately, the decision as to whether or not stimulus detection provides an adequate measure of conscious awareness does not have to be as arbitrary as the preceding discussion might suggest. One way to validate any measure of perception as a measure of awareness is to obtain satisfactory converging evidence. In other words, empirical evidence is needed to justify any assumption that a particular measure provides an adequate index of conscious awareness. One type of converging evidence that has been emphasized previously is the demonstration of qualitatively different behavioral consequences following what is assumed to reflect conscious or unconscious perceptual processing (e.g., Cheesman & Merikle, 1986; Groeger, 1988; Marcel, 1980). Thus, if "detect" and "nondetect" states lead to qualitatively different consequences, such findings can be used to validate the assumption that stimulus detection provides an adequate measure of conscious awareness.

In the present studies, the contrasting patterns of results for word and nonword stimuli provide a clear demonstration that the "detect" and "nondetect" states are qualitatively different. The recognition tasks indicated that only familiar stimuli were perceived when subjects failed to detect a stimulus but that both familiar and unfamiliar stimuli were perceived following stimulus detection. This pattern of results is completely consistent with the assumption that stimulus detection provides an adequate measure of awareness. In the nondetect or nonaware state, the subjects perceive only familiar stimuli with preexisting memory representations, whereas in the detect state both familiar and unfamiliar stimuli were perceived. Thus, this qualitative difference between word and nonword stimuli across the detect and nondetect states provides converging evidence that stimulus detection may in fact provide an adequate measure of perceptual awareness.

Although the present results certainly do not demand an interpretation that stimulus detection is an adequate measure of perceptual awareness, this qualitative difference across detect and nondetect states constrains possible alternative interpretations. For example, any alternative explanations of the dissociations between stimulus detection and recognition that are based solely on a consideration of task differences must also explain why stimulus familiarity interacts with these task differences. Thus, this qualitative difference between detect and nondetect states requires alternative explanations to become more complex, while at the same time, it provides converging evidence validating stimulus detection as a measure of awareness. If additional qualitative differences consistent with the assumption that stimulus detection provides an adequate measure of awareness can be demonstrated, then alternative explanations will become even less likely. Obviously, before stimulus detection can be justified as an adequate measure of awareness, many other qualitative differences that provide converging evidence consistent with this assumption need to be demonstrated.

The present experiments illustrate the importance of using converging evidence to validate a selected measure of consciousness. In fact, any successful approach to the measurement of awareness may require an operational definition of consciousness that can be validated by satisfactory converging empirical evidence. No matter how reasonable a measure may seem on an a priori basis, the assumption that any particular measure provides an adequate measure of awareness needs to be justified. This goal cannot be achieved entirely through any amount of definitional or methodological precision. Rather, any non-arbitrary measure of awareness requires converging empirical evidence before it can be assumed that the measure adequately distinguishes conscious from unconscious states. Consistent with this objective, the present studies provide preliminary empirical evidence suggesting that stimulus detection, a measure with considerable intuitive appeal, may in fact provide an adequate index of conscious awareness.

References

Adams, J. K. (1957). Laboratory studies of behavior without awareness. Psychological Bulletin, 54, 383-405.

Avant, L. L., & Thieman, A. A. (1985). On visual access to letter case and lexical/semantic information. Memory & Cognition, 13, 393-404.

Balota, D. A. (1983). Automatic semantic activation and episodic memory. Journal of Verbal Learning and Verbal Behavior, 22, 88-104.

Cheesman, J., & Merikle, P. M. (1984). Priming with and without awareness. Perception & Psychophysics, 36, 387-395.

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.

Diener, D. (1981). On the relationship between detection and recognition. Perception & Psychophysics, 30, 237-246.

Doyle, J. R., & Leach, C. (1988). Word superiority in signal detection: Barely a glimpse, yet reading nonetheless. Cognitive Psychology, 20, 283-218.

Duncan, J. (1985). Two techniques for investigating perception without awareness. Perception & Psychophysics, 38, 296-298.

Erdelyi, M. H. (1986). Experimental indeterminacies in the dissociation paradigm. Behavioral and Brain Sciences, 9, 30-31.

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.

Greenwald, A. G., Klinger, M. R., & Liu, T. J. (1989). Unconscious processing of dichoptically masked words. Memory & Cognition, 17, 35-47.

Groeger, J. A. (1984). Evidence of unconscious semantic processing from a forced-error situation. British Journal of Psychology, 75, 305-314.

Groeger, J. A. (1986). Predominant and non-predominant analysis: Effects of level of presentation. British Journal of Psychology, 77, 109-116.

Groeger, J. A. (1988). Qualitatively different effects of undetected and unidentified auditory primes. Quarterly Journal of Experimental Psychology, 40A, 323-339.

Hines, D., Czerwinski, M., Sawyer, P. K., & Dwyer, M. (1986). Automatic semantic priming: Effect of category exemplar level and word association level. Journal of Experimental Psychology: Human Perception and Performance, 12, 370-379.

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.

Kemp-Wheeler, S. M., & Hill, A. B. (1988). Semantic priming without awareness: Some methodological considerations and replications. Quarterly Journal of Experimental Psychology, 40A, 671-692.

Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, Rhode Island: Brown University Press.

Lindner, W. A. (1968). Recognition performance as a function of detection criterion in a simultaneous detection-recognition task. Journal of the Acoustical Society of America, 44, 204-211.

McCauley, C., Parmelee, C. M., Sperber, C. D., & Carr, T. H. (1980). Early extraction of meaning from pictures and its relation to conscious identification. Journal of Experimental Psychology: Human Perception and Performance, 6, 265-276.

Macmillan, N. A. (1986). The psychophysics of subliminal perception. Behavioral and Brain Sciences, 9, 38-39.

Marcel, A. J. (1980). Conscious and preconscious recognition of polysemous words: Locating the selective effects of prior verbal context. In R. S. Nickerson (Ed.), Attention and performance VIII (pp. 435-457). Hillsdale, NJ: Erlbaum.

Marcel, A. J. (1983a). Conscious and unconscious perception: Experiments on visual masking and word recognition. Cognitive Psychology, 15, 197-237.

Marcel, A. J. (1983b). Conscious and unconscious perception: An approach to the relations between phenomenal experience and perceptual processes. Cognitive Psychology, 15, 238-300.

Merikle, P. M. (1982). Unconscious perception revisited. Perception & Psychophysics, 31, 298-301.

Navon, D. (1986). On determining what is unconscious and what is perception. Behavioral and Brain Sciences, 9, 44-45.

Nolan, K. A., & Caramazza, A. (1982). Unconscious perception of meaning: A failure to replicate. Bulletin of the Psychonomic Society, 20, 23-26.

Parasuraman, R., Richer, F., & Beatty, J. (1982). Detection and recognition: Concurrent processes in perception. Perception & Psychophysics, 31, 1-12.

Purcell, D. G., & Stewart, A. L. (1988). The face detection effect: Configuration enhances detection. Perception & Psychophysics, 43, 355-366.

Purcell, D. G., Stewart, A. L., & Stanovich, K. E. (1983). Another look at semantic priming without awareness. Perception & Psychophysics, 34, 65-71.

Reingold, E. M., & Merikle, P. M. (1988). Using direct and indirect measures to study perception without awareness. Perception & Psychophysics, 44, 563-575.

Rollman, G. B., & Nachmias, J. (1972). Simultaneous detection and recognition of chromatic flashes. Perception & Psychophysics, 12, 309-314.

Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recognition memory: Applications to dementia and amnesia. Journal of Experimental Psychology: General, 117, 34-50.

Author Note

The authors are indebted to Jim Cheesman for many constructive discussions of the issues, and to Jan Williams for writing the software, testing the subjects and analyzing the data.

Footnotes

1. The visual angle subtended by each stimulus dimension is provided within parentheses following each linear measurement.

2. Even though no bias in criterion placement was evident in the overall detection performance, it is possible that the significant stimulus recognition on miss trials was due primarily to a subjects who adopted relatively conservative criteria for stimulus detection. To assess this possibility, correct stimulus recognition following misses was correlated with criterion placement (i.e., Cl) on the detection task. This correlation did not approach significance, r (18) = 0.21.

3. As in Experiment 1, the correlation between stimulus recognition following misses and detection criterion (i.e., Cl) did not approach significance, r (22) = -0.09.

4. A'=0.5+[(H-FA)*(1+H-FA)]/[(4H*(1-FA)] ; H=hits, FA=false alarms

5. The correlation between correct lexical decisions on miss trials and the criterion for stimulus detection (i.e., Cl) was not significant, r (22) = -0.24. In fact, the direction of this small correlation suggests that subjects with liberal detection biases, rather than subjects with conservative biases, may have been slightly more sensitive to lexical status on miss trials.