Trapped by Experience:

The Acquisition of Automatic Response Biases through

Stimulus-Response Mapping and Categorization

Determined by a Compatibility Task

 

Yvonne Lippa

Indiana University

and

Max Planck Institute for

Psychological Research

Robert L. Goldstone

Indiana University

 

 

 

 

 

 

 

Running head: ACQUISITION OF AUTOMATIC RESPONSE BIASES

Abstract

Two experiments explored whether spatially neutral stimuli acquire the ability to automatically elicit spatial responses. In Experiment 1, participants associated line-drawings with either left or right key presses. Subsequently, the pictures were used in a Simon task wherein participants made left and right key presses based on the color of the picture, ignoring its shape. Participants responded more quickly when the key press previously associated with the picture matched, rather than mismatched, the response required by the picture's color. In Experiment 2, participants learned response categories that grouped spatially ambiguous line-drawings together with pictures of left- and right-pointing arrows and fingers. A subsequent Simon task again yielded compatibility effects, indicating that the spatially ambiguous pictures inherited the response biases of the other objects in their category. Thus, responses directly associated with shapes, and indirectly associated with shapes by category membership, are both automatically triggered even when the responses are irrelevant and inappropriate.

 

Trapped by experience:

The acquisition of automatic response biases

through stimulus-response mapping and categorization

determined by a compatibility task

Everyday life provides many examples of situations where different response alternatives are conceivable, but where we consistently prefer one response over another. Prominent examples include tastes and attitudes. Some people prefer to buy a sweater in red rather than blue, to drink wine rather than beer, or to vote for party A rather than for party B (????Perhaps some other examples and references????????).

These biases to respond the one and not the other way may have their origins in the learning history of a person. Stimuli, like a particular color or beverage, may not be preferable or non-preferable from the start of life, but rather experiences with them may provoke certain response biases to occur. In the present article, we review research in the fields of animal and human learning, and stimulus-response (S-R) compatibility showing that stimuli indeed acquire the ability to elicit arbitrary responses, if they have been involved in S-R mapping or categorization tasks.

The main interest of the present study was to determine if response biases, once acquired, exert control over future behavior automatically or whether they are a result of learning experiences being intentionally applied to a new situation. In other words, if a person had certain experiences with a stimulus in the past, does the same stimulus in a different setting elicit the previous behavior automatically or does it work as cue that helps a person to recall and transfer the previous behavior? The idea that we become slaves of our learning experience might sound odd, especially when we think of stimulus-response relationships that are arbitrary, like tastes or attitudes. It seems like people have an attitude towards a stimulus, i.e., that the stimulus is subject of the attitude, but not that the stimulus evokes the attitude. However, there is some evidence that we may be trapped by previous experience much more than we want to admit. In a problem solving study by Glucksberg and Danks (1968), for instance, the task was to complete a simple electrical circuit without having enough wire to do so. Participants were given several materials, among them a screwdriver whose metal blade could compensate for the lack of wire. In one group, all materials were labeled, including the screwdriver as SCREWDRIVER. In a second group, the screwdriver was labeled SCREWDRIVER: HANDLE, BLADE. In a third group, no labels were given. It was found that in the group with different labels for the screwdriver’s parts, more participants were able to solve the problem than participants in the groups with the label SCREWDRIVER or no labels. Thus, it turned out that problem solving behavior was governed by the stimulus information participants attended to. If the stimulus was the entire screwdriver, it evoked its usual function or response, namely to turn screws. Turning screws, however, did not help in conducting electricity and, therefore, the probability of solving the problem was low. If, in turn, the stimulus was the metal blade of the screwdriver, properties were evoked that helped solving the problem, namely electrical conductivity. In sum, the study of Glucksberg and Danks (1968) illustrates that, due to learning history, certain stimuli evoke certain responses sufficiently strongly that people have a hard time to chose a different, more appropriate response.

In the present study, we pursued the question of whether response biases are automatic or not by testing whether they modulate performance in a compatibility task, specifically in a Simon task (cf. Simon & Rudell, 1967). The Simon task measures the tendency of a stimulus property to elicit a response when it is not in the focus of attention, when it is known to be irrelevant, and when it varies along an irrelevant dimension. Two experiments were conducted to determine the influence of responses biases in the Simon task after having been acquired in a S-R mapping task (Experiment 1) and a categorization task (Experiment 2). First, we revie research on how response biases are acquired and whether they are automatic or strategic in nature.

How are response biases acquired?

The simplest account for why a stimulus acquires the ability to elicit a response is that it was previously assigned to the response. If subjects repeatedly experience that a particular stimulus goes with a particular response, it is likely that when the same stimulus is presented in a different task, subjects continue to do what they have done before. The acquisition of response biases through stimulus-response mapping has been demonstrated in S-R compatibility research. In a study by MacLeod and Dunbar (1988), for example, subjects were trained to associate four unfamiliar shapes with four different color names. In a subsequent test phase they participated in a Stroop task. Subjects were asked to name the colors of shapes when they appeared or to name the color words when they appeared in the form of the shapes. With only a brief training in color-naming the shapes, performance was solely determined by color: A match between color and color name facilitated performance, while a mismatch hindered it, irrespective of which shape was presented. However, after 20 days of practice in color-naming the shapes, the original asymmetry was reversed. Color-naming the shapes remained unaffected by the respective color, but color-naming showed interference from incongruent shapes. Thus, the shapes had acquired the ability to elicit a color name stronger than the colors themselves. A similar impact of practice has been reported by Proctor and Lu (1999). In one of therir experiments, some subjects received 1,800 trials of practice with an incompatible, spatial S-R mapping -- Letters that appeared on the left side of the screen were assigned to a right response and letters that appeared on the right side of screen were assigned to a left response. In a subsequent test phase subjects participated in a Simon task. Spatial position of the letters was now a task-irrelevant stimulus attribute and subjects gave left and right responses based only on the identity of the letter (H vs. S). The practice with previously incompatible stimulus-response mappings carried over to the Simon task. If the letter required a left response, this was easier to perform when the letter had earlier appeared on the right side. On the other hand, if the letter required a right response, this was easier to perform when the letter appeared on the left side. In sum, there is some evidence that learning experience with a consistent S-R mapping transfers to other tasks and causes response biases to occur.

Besides direct stimulus-response mapping, categorization has also been shown to produce response biases. When subjects learn that a number of stimuli belong to the same category, stimuli pass their characteristics on to the other category members. In other words, a stimulus acquires the ability to elicit a particular response because it joins a category of stimuli that evoke this response. Many of the studies that have investigated the acquisition of response biases through categorization come from the animal learning literature and, typically, use a three-step procedure (Urcuioli, Zentall, Jackson-Smith & Steirn, 1989, Exp. 2; Zentall, Steirn, Sherburne & Urcuioli, 1991; Urcuioli, Zentall & DeMarse, 1995; Wasserman, DeVolder & Coppage, 1992). In a study by Wasserman, DeVolder and Coppage (1992), for example, pigeons learned in a first categorization phase to classify perceptually dissimilar stimuli (cars, chairs, flowers, and people) into two arbitrary categories. For instance, Response R1 was reinforced when A or B was present and R2 was reinforced when C or D was present. In the second, shaping phase, the pigeons learned a new response to some members of the respective stimulus classes. For instance, R3 was reinforced in the presence of A and R4 in the presence of C. The final test phase replicated the second phase using all members of each category, maintaining a differential reinforcement for R3 and R4 when A and C were present and introducing a non-differential reinforcement when B and D were present. Pigeons choice of R3 and R4 exceeded chance when B and D appeared, respectively. Hence, despite the fact that B and D have never been assigned to R3 and R4, they elicited these responses because they inherited the response biases from their category members. The acquisition of response biases through categorization has also been investigated in humans (Eisman, 1955; Dougher, Augustson, Markman, Greenway & Wulfert, 1994; Grice, Henriksen & Speiss, 1972; Jeffrey, 1953; Malloy & Ellis, 1970). Jeffrey (1953), for instance, trained preschoolers in the categorization phase to classify the colors white, gray, and black into two categories by giving the verbal responses "black" and "white". Some subjects assigned white and gray to one response category and black to the other category, while other subjects assigned black and gray to one and white to the other category. In a second, shaping phase, the color gray was omitted and subjects learned to make pull and push responses to the colors white and black, while still labeling them. In the final test phase, the color gray was reintroduced and subjects continued the task of the second phase. It turned out that compared to a baseline ascertained prior to learning, the push and pull responses to the color gray were a function of category membership. If gray was grouped with white it elicited the response assigned to white, while if it was grouped with black it elicited the response assigned to black. In sum, there is evidence that a stimulus acquires the ability to elicit a particular response simply because it is grouped together with stimuli that elicit the response.

Are response biases automatic or strategic in nature?

According to the evidence just reviewed, there is no doubt that experience in S-R mapping and category learning produces response biases to occur in subsequent tasks. However, it is still an question why learning experience gains this influence. Is it because subjects cannot suppress previous learning experiences and response biases occur automatically or is it because subjects explicitly use what they learned before and response biases occur for strategic reasons?

For either the idea that response biases are automatic or that they are strategic in nature, one can come up with an explanatory account. An automatic response bias account follows the framework of associative network models (Rumelhart & McClelland, 19?, other references???????????). In the case of S-R mapping learning, it is assumed that repetitive assignment establishes an associative link between the stimulus and the response. This may be transient in the beginning, but if the S-R assignment is further practiced, the connection is strengthened and lasts over time. Thus, if after learning a stimulus occurs in a different task, it not only activates its representation. Due to spreading activation, it also activates the associated response and, thus, creates a bias. The influence of categorization is attributed to mediated or secondary generalization, a mechanism which was introduced by Hull (1939). In the categorization phase, a simple S-R association is established between the category members and their categorizing response, say A-R1 and B-R1. More precisely, Hull (1939) thought of the S-R associations being represented as A-R1-r1 and B-R1-r1, taking into account that each response entails perceivable effects (r1) that also enter the association chain. In the second phase, one category member, A, is associated with a new response, R2. This, however, does not only form the link A-R2, but since A evokes r1, also the link r1-R2. This mediating link between r1 and R2 is the reason why other category members inherit the response bias from A. If in the third phase B is presented, the associated response R1, and its corresponding effect r1, are activated, which, in turn, activates R2, thus creating a bias. In sum, according to an automatic response bias account, S-R mapping and category learning establish associative links between stimuli and responses that subsequently exert control over response selection processes. The formation of S-R associations is subject to the person’s voluntary behavior. The response bias resulting from the associations, however, is mandatory because, once formed, the associations transfer activation automatically.

According to a strategic response bias account, behavior is biased in a transfer task because previously learned rules are applied once again. Such an account may also utilize the framework of an associative network, but instead of transferring activation automatically, associative links are explicitly used to retrieve information from the memory. Assume that a person hasjust completed a task in which one stimulus (or a class of stimuli) required one response and another stimulus (or class of stimuli) required another response. Now, the person is supposed to solve a second task, which involves the same stimuli once again. While animals might treat this second task as a new, independent task, humans do surely not. They recognize the stimuli as previously shown items and they may even suspect that the transfer between the tasks is under test. Thus, to solve the second task, it would a reasonable strategy to recall what has happened before and apply these regularities once again. The responses in the second task would be biased because they reflect what has been experienced before. Yet, this bias is strategic in nature because the person can choose not to take advantage of the previous learning experience but to apply a new strategy.

In order to provide evidence in favor of the automatic response bias account or the strategic response bias account, experiments are required that employ a test task where performance is unlikely to be contaminated by strategic behavior. If such a task provides no evidence for response biases, we can conclude that the strategic component is a necessary variable in biased behavior. If, in turn, response biases do occur, then there is evidence that performance can be determined by previous experience automatically.

The above-mentioned experiments that investigated the acquisition of response biases through categorization are all ambiguous regarding the test task requirements. This holds in particular for the human learning studies. Jeffrey’s (1953) preschoolers, for example, were instructed to keep categorizing the stimuli (naming the colors), while testing whether they produced a pushing or pulling response bias. Thus, the use of a strategy was almost requested. In the test task employed by Malloy and Ellis (1970), subjects were exposed to all members of each category and asked to identify the stimuli from the shaping phase by giving the name learned in the shaping phase. This procedure is quite elegant because the subjects’ attention was drawn to what had happened in the shaping phase and, thus, pulled away from what happened in the categorization phase. However, the stimuli were variations of a prototype and, therefore, it is not clear whether the obtained transfer effects resulted from perceptual similarities, (i.e., primary generalization; Hull, 1939) or from categorization (secondary generalization). The most convincing evidence comes from studies using conditioned responses. For example, after having subjects categorize abstract figures, Dougher at al. (1994, Exp.1) paired one category member with electric shock. A subsequent test task using skin conductance as a measure then demonstrated that the conditioning transferred to the other category members. Since a change in galvanic skin response is less subject to voluntary control, this finding provides some evidence for response biases acquired through categorization to occur automatically. However, galvanic skin responses following electric shocks or eye blink responses following air puffs (Grice et al, 1972) are exceptional responses as they help avoiding pain and threat. They form a distinguished class of responses and it is still open to question to what extent the findings can be generalized to other more arbitrary responses. Another shortcoming that holds for all of the above-mentioned category learning studies is that the stimuli whose ability to elicit a response is in question are task-relevant in the test phase. That is, the task that must be solved in the test phase required humans and animals to consider stimulus aspects that have been relevant beforehand. Thus, even if category learning forms associations between the critical stimulus and the response, evidence for response biases does not necessarily indicate that these links exert control automatically. After all, the organism is supposed to respond to the critical stimulus and might therefore intentionally use the pre-existing connections, an argument in favor of the strategic response bias account.

A task that overcomes these problems and is a better test for whether a stimulus elicits a response automatically or not, is one where the critical stimulus is task-irrelevant. Subjects are exposed to the critical stimulus, but the problem they solve involves a second stimulus or stimulus attribute. That is, all task demands and strategies refer to the second stimulus aspect, so that if the critical stimulus aspect turns out to influence performance, it must be attributed to mandatory processing. Tasks that resemble these characteristics are compatibility tasks like the Stroop task and the Simon task, which were used in the previously mentioned studies on the acquisition of response biases through S-R mapping (MacLeod & Dunbar, 1988; Proctor & Lu, 1999). Since these studies were successful in showing that response biases, once acquired, are elicited automatically, we have chosen to adopt their technique. Specifically, we employed the Simon task as a test task. The Simon task is a choice-reaction task, in which the response varies on one dimension and the stimuli vary on two dimensions. For instance, subjects perform left and right key presses in response to the letter H and S that are randomly presented to the left or right of a central fixation point. The task-relevant stimulus dimension is the identity of the letters, while the position of the letters is irrelevant. For example, subjects are instructed to press as quickly as possible the left key if H appears and the right key if S appears, irrespective of the position of the letter. The typical finding, also termed the Simon effect, is that performance is facilitated and hindered if the irrelevant stimulus attribute matches and mismatches, respectively, the attribute of the required response. For the example, responses are faster and less error-prone when the H occurs on the left side and the S on the right side than when the H occurs on the right side and the S on the left side (Proctor and Lu, 1999, Exp. 1). Hence, the Simon effect demonstrates that a task-irrelevant stimulus is able to automatically elicit a response, indicated by either facilitation or impediment of the selection of the correct response.

In the present study, we took advantage of the Simon effect by using it as an indicator for whether stimuli elicit a response automatically or not. Specifically, in Experiments 1 and 2, subjects participated in a S-R mapping and category learning task, respectively, in which spatially neutral line-drawings acquired the ability to elicit left and right responses. After learning, in both experiments, subjects were transferred to the Simon task. Left and right key presses served as responses and the line-drawings as stimuli, with the shape of the drawings being task-irrelevant and their colors being task-relevant. We were interested in whether the preceding learning experience would generate Simon effects in the test task, indicating that responses are automatically triggered by a particular stimulus.

Experiment 1

Experiment 1 tested whether response biases that are acquired through stimulus-response mapping are automatic in nature. Like in the study by MacLeod and Dunbar (1988), we asked subjects to practice an arbitrary stimulus-response mapping. Stimuli were line-drawings of objects that did not exhibit spatially biased attributes (e.g., a butterfly or barrel) and were assigned to a left and right response. In contrast to MacLeod and Dunbar’s (1988) study, we employed a two-to-one mapping, with two stimuli assigned to one response instead of each stimulus assigned to a different response. In addition, our subjects practiced the assignment between stimulus and response only for a short period of time.

To test the impact of learning, we used the Simon task. Subjects gave left and right responses to the line-drawings, with the shape of the stimuli being irrelevant and color being the relevant stimulus attribute. If the acquired stimulus-response associations exert control automatically, performance in the subsequent Simon task should be a function of the respective picture-color combination. If the color requires a response that was previously assigned to the stimulus (compatible condition), responding should be faster and less error-prone than when the color requires the opposite response (incompatible condition). Besides the picture-color combination, we varied the time elapsing between the onset of the picture and the onset of the color. We manipulated stimulus-onset asynchronies (SOAs) to account for the potentially dynamic nature of response biases. Following the automatic response bias account, these effects occur because responses are activated through associative links. This activation is a time-consuming process, especially if not only one, but several associative links are involved. Thus, it could be that the pictorial information must precede the color information in order to produce interference. On the other hand, it has been shown that stimulus or response activation that is irrelevant to the task at hand decays over time (Hommel, 1994). A particularly large SOA might therefore produce no response bias at all because facilitating or interfering information has already decayed. Since the particular processing times of our stimuli and colors are unclear and, in addition, subjects may differ in processing the stimulus information, we chose three different SOAs. Each originally black picture turned red or blue after 0 ms, 100 ms, or 300 ms. This range should allow us to account for the potential dynamic of response biases as well as for individual differences.

Method

Subjects

Twenty-four (12 females and 12 males) undergraduate students from Indiana University served as subjects in order to fulfill a course requirement and were run in parallel sessions. They were aged between 18 and 28, right-handed by self-report, and had normal or corrected-to-normal vision.

Apparatus and Stimuli

Stimulus presentation and data acquisition were controlled by Dell Dimension XPS P133s computers and the software ERTS (Experimental Run Time System; Beringer, 1998). Six line-drawings from the Snodgrass and Vanderwart picture corpus (Snodgrass & Vanderwart, 1980) served as stimuli: the barrel, the butterfly, the clown, the light bulb, the shirt, and the tree (see Figure 1). They were presented on a white background on Sony Multiscan 15SFII monitors with a resolution of 800 x 600 pixels either in black, red, or blue(Hue-Saturation-Value model; 0/, 0%, and 0% for black; 0/, 100%, and 100% for red; 240/, 100%, and 100% for blue) depending on the task and condition. The pictures differed in width and height but all fit in a 6.0 x 5.3 cm fixation frame. The viewing distance was approximately 50 cm.

Procedure

The experiment was divided into a learning and test phase, consecutively completed within a one-hour session. In the learning phase, subjects performed a categorization task and in the test phase a Simon task.

S-R mapping task. Subjects learned to classify four of the six pictures into two groups of two pictures by pressing the left and right shift key (see Figure 1, left panel). Each subject received a randomly chosen picture-response assignment. On each trial, one picture was presented at the center of the screen and remained visible until the subject responded or 10 sec had elapsed. The order of picture presentation was randomized. If the subject responded correctly, the picture remained displayed for another 500 ms and then the next trial started after an inter-trial interval of 800 ms. In case of a wrong response, the trial was immediately terminated, a beep tone was presented and then a new trial started. Subjects completed 900 correct trials, with error trials and trials with response times equal or below 120 ms (mean anticipations: 1.3%) or longer than 10 sec (mean overtime responses: 0.2%) eliciting a beep and being repeated at a random position later in a block.

Simon task. Subjects performed a compatibility task, with the shape of the pictures being the irrelevant attribute and the color of the picture being the relevant stimulus attribute (see Figure 1, center panel). The task was to press as quickly and accurately as possible the left and right shift keys in response to the colors red and blue, irrespective of which picture was shown. The color-key mapping was counterbalanced across subjects. All six pictures were used. The four pictures categorized previously were the experimental stimuli. The other two, new pictures served as a neutral or baseline conditions to determine whether there was facilitation in the compatible condition and/or interference in the incompatible condition.

Each trial started by presenting a fixation frame (6.0 x 5.3 cm) in the center of the screen, remaining visible throughout the trial. After 500 ms, one randomly chosen picture was displayed until the subject responded or 1000 ms had passed (missing). To take time differences in processing the shape and color information into account, we used three Stimulus-Onset-Asynchronies (SOAs) between the onset of the picture and its color: The picture could appear colored from the very beginning on (SOA of 0 ms), or be presented in black and then turn red or blue 100 ms or 300 ms after its initial appearance. Each of the six pictures was presented eight times in red and blue for each of the three SOA conditions, such that 288 trials were completed overall. Auditory feedback was given on error, missing (1.0 %), and anticipation trials (reaction time equal or below 120 ms: 0.3 %), which were then repeated at a random position later in the block.

Results

Performance in the S-R mapping task

All subjects completed 900 correct trials with a mean reaction time of 701 ms. On average, 44 errors were made.

Performance in the Simon task

Mean reaction times (RTs; measured from color onset) and mean percentages of error were calculated separately for the three SOA conditions and the three compatibility conditions of compatible (C), incompatible (IC), and neutral (N) (see Figure 1, right panel). A picture-color combination was considered compatible if the color required the same response that was assigned to the picture in the categorization task. If the color required the opposite response, it was considered an incompatible trial. In the neutral condition pictures were presented that did not occur in the categorization task. On RT and error data, analyses of variance (ANOVAs) were conducted, using the within-subject variables condition (compatible, neutral, and incompatible) and SOA (0, 100, and 300 ms).

The main effect of SOA was significant, showing a decrease in reaction time, F(2, 46) = 42.82, MSE = 739.85, p < .001, and an increase in errors, F(2, 46) = 5.98, MSE = 25.72, p < .01, with increasing SOA. That is, the more time allowed to process the pictorial information before the color information was available, the faster and more error-prone was the response to the color. The main effect of condition was reliable in RT data, F(2, 46) = 4.79, MSE = 384.48, p < .05, as well as error data, F(2, 46) = 3.71, MSE = 14.41, p < .05. Overall, subjects responded 10 ms faster and made 1.7 % fewer errors in the compatible condition than in the incompatible condition [t(23) = 2.77, p < .05, two-tailed, and t(23) = 2.70, p < .05, two-tailed, for RT and errors data, respectively]. According to RT data, the performance in the neutral condition was on average 6 ms worse than in the compatible condition and 4 ms better than in the incompatible condition [t(23) = 1.65, p = .112, two-tailed, and t(23) = 1.72, p = .10, two-tailed, for C vs. N and IC vs. N, respectively]. According to the error data, the performance in the neutral condition did not differ from the performance in the incompatible condition, t < 1, but, with a difference of 1.1 %, it was significantly worse compared to the performance in the compatible condition, t(23) = 2.14, p < .05, two-tailed. There was no interaction between condition and SOA, F < 1.2.

To check whether the compatibility effect was influenced by practice, we calculated the same ANOVA as described above, with RT and error data broken down in two halves (first half of correct trials in each condition vs. second half of correct trials in each condition) (see Table 1). Subjects responded faster in the first half (432 ms) than in the second half (456 ms), F(1, 23) = 25.92, MSE = 2321.92, p < .001, but this did not affect the compatibility effect, F < 1. The compatibility effect obtained in error data was also not modulated by practice, F < 1, or by a higher-order practice x condition x SOA interaction, p > .10.

Discussion

Experiment 1 demonstrated the acquisition of automatic response biases through direct stimulus-response association. If, for example, the spatially neutral picture of a butterfly was repeatedly assigned to a left response, the butterfly continued to elicit a left response even when its shape was an irrelevant attribute to the task at hand. This response bias was acquired very quickly. With only 225 correct stimulus-response assignments for each of the four stimuli (about half an hour), the learning phase was rather short. It is likely that the effects would increase if learning were prolonged. Once acquired, the response bias yielded effects in all SOA conditions. This indicates that irrespective of when the relevant color information is available the picture’s shape is mandatorily processed and determines performance over a longer period of time. There was also evidence that the response bias facilitated as well as hindered the selection of a response. At least by number, we found that the performance in the compatible and incompatible condition was better and worse, respectively, compared to the performance in the neutral condition. The finding that the compatibility effect did not change with increasing practice suggests that the response bias is relatively robust and lasts over time. In sum, Experiment 1 showed that a short-lasting, but consistent mapping between spatially neutral stimuli and spatial responses results in enduring response biases of these stimuli that automatically facilitate and hinder response selection.

Experiment 2

Experiment 2 investigated whether response biases that are acquired through categorization are automatic in nature. The learning procedure we adopted deviated from the procedure of previous secondary generalization studies in a critical point. We dropped the shaping phase, in which some of the category members are associated with new responses and used stimuli that already possessed the ability to elicit left and right responses, namely left- and right pointing arrows and fingers. Thus, the learning consisted of only one phase, the categorization task, in which subjects grouped these spatially biased stimuli together with stimuli that were expected to acquire response biases. This entailed a reversal of the usual order of learning phases. Typically, stimuli are first categorized and then some members are associated with a new response. Since in Experiment 2 we used stimuli that were already spatially biased, the categorization phase follows the shaping phase. Swapping the training steps is an interesting manipulation because it tests whether the knowledge of category membership is a necessary precondition to generalize characteristics or whether is can applied retrospectively. According to Hull’s (1939) analysis, the categorizing response (R1) must precede the shaping because it provides the stimulus (r1) which establishes the critical link to the new response (R2). Thus, according to Hull (1939), the reversed training order of Experiment 2 should prevent secondary generalization from occuring. Urcuioli (1999) tested Hull’s prediction with pigeons and obtained equivalent transfer for both training orders. Experiment 2 was an attempt to test Hull’s prediction in human learning.

The impact of learning was again measured by the Simon task. All members of each category served as Simon task stimuli. For the spatially biased stimuli, the pointing arrows and fingers, a strong Simon effect was predicted: A match between pointing direction and response position should produce better performance than a mismatch. If categorization causes stimuli to inherit the response biases of the other members of their category and if this bias is automatic in nature, a Simon effect should also be obtained for the stimuli grouped together with the spatially biased stimuli.

Method

Subjects

Thirty-two (16 females and 16 males) undergraduate students from Indiana University served as subjects in order to fulfill a course requirement and were run in parallel sessions. They were aged between 18 and 23, right-handed by self-report, and had normal or corrected-to-normal vision.

Apparatus, Stimuli, and Procedure

Experiment 2 was a replication of Experiment 1, except for the following changes in stimulus material and procedure. Six new line-drawings of the Snodgrass and Vanderwart picture corpus (Snodgrass & Vanderwart, 1980) served as stimuli: a left- and right-pointing version of the finger and the arrow, the eagle, and the French horn (see Figure 2). The left- and right-pointing pictures will be referred to as spatially biased stimuli because they already possessed the ability to elicit left and right responses. The eagle and the horn will be referred to as spatially ambiguous stimuli. Unlike the stimuli of Experiment 1, they are not symmetric, but exhibit spatially ambiguous attributes. When focusing on the eagle’s body and the horn’s end, they can be considered pointing to the left, and when focusing on the eagle’s beak and the horn’s mouthpiece, they can be considered pointing to the right. The pictures differed in width and height but again fit in a 6.0 x 5.3 cm fixation frame. The viewing distance was approximately 50 cm.

Categorization task. Subjects learned to assign two left-pointing stimuli and one spatially ambiguous picture to one category and two right-pointing stimuli and the other spatially ambiguous picture to a second category (see Figure 2, left panel). Whether the eagle or the horn was paired with the left- or right-pointing stimuli was counterbalanced across subjects. Subjects indicated category membership by clicking with the left mouse key on a top or button key (3.2 cm wide and 1.6 cm high), which were shown on the screen throughout the experiment. The top and bottom keys were located 4.3 cm above and below the center of the screen and were always labeled "Category 1" and "Category 2", respectively. The category that contained the left-pointing stimuli was always assigned to the top button and the category containing the right-pointing stimuli was always assigned to the bottom button. This up-left/down-right assignment is counter to the up-right/down-left preference reported in the literature (Weeks and Proctor, 1990), so that any effects we might find would not be attributed to a pre-existing stimulus-response mapping preference. Subjects completed 592 correct trials (148 replications for the eagle and the horn and 74 replications for each of the arrows and fingers), with error trials and trials of a response time equal or below 120 ms (mean anticipations: 0) or longer than 10 sec (mean missings: 1) producing a beep as feedback, and being repeated at a random position in a block. After that, subjects could end the training session by fulfilling the criterion of 12 correct trials within an additional 64 trials.

Simon task. This was identical to the task in Experiment 1. There was no neutral condition, instead, the left- and right-pointing stimuli were presented in addition to the critical pictures of the eagle and horn (see Figure 2, center panel). The eagle and the horn were each presented 16 times in red and blue for each of the three SOA conditions, while each of the two arrows and fingers was presented eight times. Overall, 384 trials were completed. Auditory feedback was given again on error, missing (2.8 %), and anticipation, which were then repeated at a random position later in the block.

Results

Performance in the categorization task

All subjects completed the 592 correct trials with a mean reaction time of 1195 ms. On average, 25 errors were made. The learning criterion was fulfilled without any errors within 12 trials with a mean reaction time of 1076 ms.

Performance in the Simon task

RT end error data were calculated separately for the two stimulus sets (spatially biased stimuli vs. spatially ambiguous stimuli), the three SOA conditions, and the two compatibility conditions (C vs. IC) (see Figure 2, right panel). In the case of spatially biased stimuli, trials were considered compatible and incompatible if there was a correspondence and non-correspondence, respectively, between the response required by the color and the pointing direction of the stimulus. In the case of the spatially ambiguous stimuli, a picture-color combination was considered compatible if the color required by the response matched the pointing direction of the stimuli with which the picture was previously paired. If the color required by the response mismatched the pointing direction of the stimuli with which the picture was grouped together, it was considered an incompatible trial. Separate ANOVAs were conducted on error and RT data for spatially biased and ambiguous stimuli, using the within-subject variables condition (compatible and incompatible) and SOA (0, 100, and 300 ms).

Spatially biased stimuli. The main effect of SOA was significant, indicating a decrease in reaction time, F(2, 62) = 44.93, MSE = 870.37, p < .001, and an increase in errors, F(2, 62) = 4.33, MSE = 15.64, p < .05, with increasing SOA. In addition, the main effect of condition was highly significant for reaction time, F(1, 31) = 132.40, MSE = 444.12, p < .001, and errors, F(1, 31) = 27.60, MSE = 28.74, p < .001. Performance in the compatible condition (454 ms and 2.8 % errors) was better than in the incompatible condition (489 ms and 6.8 % errors). The condition x SOA interaction did not approach significance, p > .20 and p > .10 for RT and error data, respectively. An ANOVA including the variable half (first half or second half of Simon task trials) showed that the compatibility effect observed in RT data was the same in the first and second halves, p > .15 (see Table 2). For error data, we found that the compatibility effect was slightly larger in the first half (4.4 %) than in the second half (2.8 %), F(1, 31) = 2.99, MSE = 20.28, p = .094. This is probably only due to the fact that subjects made more errors in the first half (5.0 %) than in the second half (3.7 %), F(1, 31) = 4.63, MSE = 35.73, p < .05.

Spatially ambiguous stimuli. The main effect of SOA was highly significant in reaction time, F(2, 62) = 62.71, MSE = 615.77, p < .001, and marginally reliable in errors, F(2, 62) = 3.08, MSE = 11.71, p = .053, indicating, again, that the longer the delay between picture and color onset, the faster and more error-prone the response. The main effect of condition failed to reach significance in error data (3.5 vs. 3.8 % for C and IC, respectively), F < 1, but was significant in RT data, F(1, 31) = 4.50, MSE = 393.17, p < .05. Subjects were faster in the compatible (459 ms) than in the incompatible condition (465 ms). As revealed by an ANOVA including the variable half, this compatibility effect occurred mainly in the first half (a 11 ms difference between C and IC) and not in the second half (1 ms difference), F(1, 31) = 5.38, MSE = 381.15, p < .05 (see Table 2). The absence of the compatibility effect in error data was true for the first as well as second half, F < 1.

After the experiments, subjects were asked to judge the pointing direction of the eagle and the horn. 7 subjects gave no valid answer (four subjects in the condition where the eagle and horn were grouped with left- and right-pointing stimuli, respectively, and three with the reversed mapping). Of the 25 subjects that gave valid answers, 4 gave only part answers, judging only the eagle or only the horn. Thus, instead of 50 observations, there were only 46. When an answer of a left pointing direction was expected, left was indicated 16 times and right 6 times. When an answer of a right pointing direction was expected, right was indicated 17 times and left 7 times, chi-square (1, N = 46) = 8.712, p < .001. Thus, for the majority of subjects, the pointing judgment for the spatially ambiguous stimuli was a function of category membership.

Discussion

As expected, Experiment 2 yielded strong effects of automatic responses biases, hence a Simon effect, for pointing arrows and fingers: A match between pointing direction and response location produced better performance than a mismatch, even though the pointing direction was irrelevant to the task at hand. This finding can be interpreted as further support for the acquisition of automatic response biases through S-R mapping experience. There is evidence that pointing gestures are not innately meaningful, but rather that looking where others point must be acquired and is not yet fully developed at 10.5 months in humans (Leung & Rheingold, 1981). Thus, it could be that, like the butterfly in Experiment 1 or the forms in an arbitrary alphabet, the pointing arrows or fingers have started out as a spatially neutral line-configuration that only through life-long associations with a left or right responses have acquired the ability to elicit a left and a right response.

Most interestingly, Experiment 2 yielded evidence for secondary generalization. Stimuli acquire the ability to elicit responses simply by grouping them together with stimuli that evoke these responses. This finding is not consistent with the secondary generalization mechanism proposed by Hull (1939) because, logically, the categorization took place after some of the category members were associated with new responses. If generalization across category members requires already established categories, one would have expected no transfer in Experiment 2. In the general discussion, we will tackle this issue in more detail and discuss alternative explanations.

The fact that we found effects of secondary generalization in the Simon task shows that the acquired responses biases exert control over behavior automatically. The effect of response bias occurred only in reaction times and vanished over time. In the first half of trials, it amounted to 11 ms and was as large as the effect in Experiment 1. Thus, if present, response biases acquired through categorization appear to be as strong as response biases acquired through S-R mapping. However, the lack of an effect in the second half of trials suggests that they are subject to more rapid extinction. From an associative network point of view, one may argue that the associations between the stimuli and the categorizing responses weakened rapidly. This may be due to the fact that the associations have not been very strong in the first place. Each of the spatially biased stimuli was associated with the categorizing responses only 74 times and each of the spatially ambiguous stimuli only 178 times. It is likely that an increase in the number of valid stimulus-response pairings would strengthen the associations and, thus, the response bias.

Finally, the effect of response biases could also be observed in judgment behavior because the majority of subjects indicated that the eagle and horn were pointing in the direction of their category members. These judgments, however, may be contaminated by strategy since subjects might have tried to judge the stimuli in accordance with what was experienced in the categorization task. For this reason, the judgments provide further evidence for categorization being effective in evoking response biases, but they are not conclusive with respect to the question of whether the response biases occur automatically or not.

General discussion

The present study was interested in whether response biases that are acquired through S-R mapping and categorization automatically exert control over behavior in a subsequent Simon task. In Experiment 1, we found a Simon effect for pictures of spatially neutral objects after they have been repeatedly assigned to spatial responses. In Experiment 2, we found a Simon effect for pictures of spatially ambiguous objects that have been repeatedly categorized with stimuli that naturally elicit spatial responses. Since the stimulus aspect that was relevant in the learning tasks was irrelevant in the Simon task, our findings show that its ability to elicit a response was not completely suppressed and became effective even when it led to an incorrect response. Hence, we found evidence for that acquired response biases influence performance in a subsequent task automatically.

The automatic response bias account provides a straightforward explanation for the results of Experiment 1. A consistent and repetitive mapping between stimuli and responses results in associative links that last and, therefore, govern response selection processes automatically at a later point in time. However, to explain the results of Experiment 2 within the framework of the automatic response account some reconsideration is needed. As noted earlier, the secondary generalization mechanism proposed by Hull (1939), cannot account for the obtained response biases. Hull (1939) conjectured that a stimulus A passes on its new response characteristics R2 to an unrelated stimulus B only if at the moment of shaping a response R1 exists that is common to A and B and functions as a mediator between B and the new response R2. In the setting we employed, the categorization phase logically followed the shaping phase, so that the order in which R1 and R2 are associated with each other changed. With categorization first and shaping second, R1 precedes R2 and, therefore, can trigger R2. With shaping first and categorization second, R2 precedes R1, so that R1 may be a result of, but not the cause for R2.

However, there is reason to question whether it is appropriate to apply Hull’s (1939) ideas to Experiment 2 at all. Hull’s (1939) analysis is based on the assumption that in the categorization task the stimulus and response are two sequential events that do not overlap and that, therefore, the formed associative links are uni-directional in nature. In our experimental setting, however, the stimulus was present during and even 500 ms after the subjects responded. Hence, stimulus and response were presented simultaneously, which might have given rise to backwards or bi-directional associations (Hall, 1996; Zentall, 1998). If this is true, the occurrence of B in the test task is capable of activating, due to the categorizing response R1, the category member A, which in turn triggers its newly associated response R2. According to this explanation, Experiment 2 produced automatic response biases not because B itself evoked R2, but because it shared a category with a stimulus, namely A, which evoked R2. Bi-directional associations, however, could have been formed not only between the categorizing response and the categorized stimuli, but also between the categorizing response and the new response. Since the appearance of A also evokes an internal representation of the previously associated response R2, the categorizing response R1 not only co-occurs with A but with R2, as well. According to this explanation, Experiment 2 produced automatic response biases because B itself, via the categorizing response R1, elicited R2.

These explanations focus on the idea that automatic response biases are acquired because a response is attached to the stimulus either directly or via mediating links. Hence, it was assumed that learning influences the associative links between stimuli and responses, but not the representation of the stimulus. However, one might also pursue the idea that learning alters the stimulus representation itself. According to this notion, the butterfly, for example, acquires the ability to elicit a left response not because it is somehow attached to a left response, but because it acquires the perceptual feature "Left." Just as our learning experience with letters makes us able to say that a scribble ‘a’ looks like ‘a,’ other events might exhibit after extensive learning the attribute with which they have co-occurred in the past (cf. Ach, Gerdessen, Kohlhagen, & Margaritzky, 1932). And because the response bias would be inherent in the stimulus, it should be automatically elicited as soon as the stimulus is processed.

The question of whether discrimination learning enriches the stimulus representation or changes it, is an old one and its discussion traces back, at least, to Postman (1955) and Gibson and Gibson (1955 a and b). While Postman (1955) followed the reasoning of Hull (1939) and other association theorists (Miller and Dollard, 1941), E.J. and J.J. Gibson (1955 a and b) argued in favor of the idea that what is perceived changes. The data of the current study cannot distinguish between these two alternatives. We only found that when response biases are acquired they occur automatically, but this can be attributed to changes in stimulus-response associations as well as to changes in stimulus representation (here mentioning Goldstone & Lippa???????).

In sum, the present study has shown that response biases that are acquired through S-R mapping and categorization are exhibited in a subsequent test task automatically. That is, our subjects appeared to be trapped by their learning experience much more than one might have expected. This does, of course, not imply that response biases are always automatic and never strategic in nature. Our experiments showed to what extent response biases become effective, if it is unlikely that behavior is governed by strategies. However, if the task allows strategies to be applied, then strategies surely remain a critical variable in determining behavior.

References

Ach, N., Gerdessen, H., Kohlhagen, F., & Margaritzky, S. (1932). Finale Qualitaet (Gefuegigkeit) und Objektion. / Final quality (pliancy) and objectification. Archiv fuer die Gesamte Psychologie, 2, 267-366.

Beringer, J. (1998). Experimental Run Time System [Computer program]. Frankfurt a.M.: BeriSoft Cooperation.

Dougher, M. J., Augustson, E., Markham, M. R., Greenway, D. E., & Wulfert, E. (1994). The transfer of respondent eliciting and extinction functions through stimulus equivalence classes. Journal of the Experimental Analysis of Behavior, 62, 331-351.

Eisman, B. S. (1955). Attitude formation: The development of a color-preference response through mediated generalization. Journal of Abnormal Psychology, 50, 321-326.

Gibson, J. J., & Gibson, E. J. (1955a). Perceptual learning: Differentiation or enrichment? Psychological Review, 62, 32-41.

Gibson, J. J., & Gibson, E. J. (1955b). What is learned in perceptual learning? A reply to Professor Postman. Psychological Review, 62, 447-450.

Glucksberg, S., & Danks, J. H. (1968). Effects of discriminative labels and of nonsense labels upon availability of novel function. Journal of Verbal Learning and Verbal Behavior, 7, 72-76.

Grice, G. R., Henriksen, K., & Speiss, J. M. (1972). Response mediated generalization in eyelid conditioning with reduced conflicting information. Journal of Experimental Psychology, 92, 398-404.

Hall, G. (1996). Learning about associatively activated stimulus representations: Implications for acquired equivalence and perceptual learning. Animal Learning & Behavior, 24, 233-255.

Hommel, B. (1994). Spontaneous decay of response code activation. Psychological Research, 56, 261-268.

Hull, C.L. (1939). The problem of stimulus equivalence in behavior theory. Psychological Review, 46, 9-30.

Jeffrey, W. E. (1953). The effects of verbal and nonverbal responses in mediating an instrumental act. Journal of Experimental Psychology, 45, 327-333.

Leung, E. H. L., & Rheingold, H. L. (1981). Development of pointing as a social gesture. Developmental Psychology, 17, 215-220.

MacLeod, C. M., & Dunbar, K. (1988). Training and Stroop-like interference: Evidence for a continuum of automaticity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 126-135.

Malloy, T. E., & Ellis, H. C. (1970). Attention and cue-producing responses in response-mediated stimulus generalization. Journal of Experimental Psychology, 83, 191-200.

Postman, L. (1955). Association theory and perceptual learning. Psychological Review, 62, 438-446.

Miller, N. E., & Dollard, J. (1941). Social learning and imitation. New Haven: Yale University Press.

Proctor, R. W., & Lu, C.H. (1999). Processing irrelevant location information: Practice and transfer effects in choice-reaction tasks. Memory and Cognition, 27, 63-77.

Simon, J. R., & Rudell, A. P. (1967). Auditory S-R compatibility: The effect of an irrelevant cue on information processing. Journal of Applied Psychology, 51, 300-304.

Snodgrass, J. G., Vanderwart, M. (1980). A standardized set of 260 pictures: Norms for name agreement, image agreement, familiarity, and visual complexity. Journal of Experimental Psychology: Human Learning and Memory, 6, 174-215.

Urcuioli, P. J. (1999). Acquired equivalence in pigeons. Invited paper presented at the 70th Annual Meeting of the Eastern Psychology Association, Providence RI, April 1999.

Urcuioli, P. J., Zentall, T. R., & DeMarse, T. (1995). Transfer to derived sample-comparison relations by pigeons following many-to-one versus one-to-many matching with identical training relations. The Quarterly Journal of Experimental Psychology, 48B, 158-178.

Urcuioli, P. J., Zentall, T. R., Jackson-Smith, P., & Steirn, J. N. (1989). Evidence for common coding in many-to-one-matching: Retention, intertrial interference, andtransfer. Journal of Experimental Psychology: Animal Behavior Processes, 15, 264-273.

Wasserman, E. A., DeVolder, C. L., & Coppage, D. J. (1992). Non-similarity-based conceptualization in pigeons via secondary or mediated generalization. Psychological Science, 3, 374-379.

Weeks, D. J., & Proctor, R. W. (1990). Salient-features coding in the translation between orthogonal stimulus and response dimensions. Journal of Experimental Psychology: General, 119, 355-366.

Zentall, T. (1998). Symbolic representation in animals: Emergent stimulus relations in conditional discrimination learning. Animal Learning & Behavior, 26, 363-377.

Zentall, T. R., Steirn, J. N., Sherburne, L. M., & Urcuioli, P. J. (1991). Common coding in pigeons assessed through partial versus total reversals of many-to-one conditional and simple discriminations. Journal of Experimental Psychology: Animal Learning Processes, 17, 194-201.

 

Authors note

We want to thank John Kruschke for making lab space and facilities available and Matthew R. Littell and Timothy J. Mahoney for running subjects. Correspondence concerning this article should be addressed to Yvonne Lippa, Department of Psychology, University of California, Santa Barbara, CA 93106-9660; e-mail: ?; fax: (805) 893-4303.

Table 1

Mean Reaction Times (RT, in Milliseconds) and Mean Errors (in Percentages) for Experiment 1 as a Function of SOA, Condition, and Half

SOA 0 SOA 100 SOA 300

 

 

C N IC

C N IC

C N IC

1. Half RT Errors

441 448 452

1.6 2.0 4.4

436 440 441

4.2 6.7 5.6

409 410 416

5.6 7.3 6.4

2. Half RT

Errors

467 480 484

2.1 4.0 4.6

456 460 465

3.5 2.6 5.3

424 431 436

5.2 5.0 4.9

 

Table 2

Mean Reaction Times (RT, in Milliseconds) and Mean Errors (in Percentages) for Experiment 2 as a Function of Stimulus Set, SOA, Condition, and Half

Spatially biased stimuli Spatially ambiguous stimuli

 

 

SOA 0 ms

SOA 100 ms

SOA 300 ms

SOA 0 ms

SOA 100 ms

SOA 300 ms

 

 

C IC

C IC

C IC

C IC

C IC

C IC

1. Half RT

Error

473 514

1.8 5.9

456 499

2.9 8.3

431 463

3.6 7.4

473 487

3.3 3.4

459 472

3.9 4.4

432 437

3.4 4.5

2. Half RT

Errors

485 511

1.5 3.8

449 489

1.8 6.1

431 461

3.4 5.4

491 484

2.2 1.5

464 470

3.2 3.5

434 439

3.1 4.3

jkjkjkjjkjkljkjkjFigure captions

Figure 1. Left and center panel: Illustrations of the tasks in the learning and test phase in Experiment 1, using an exemplary S-R mapping. The solid arrows indicate responses required by instruction and the dashed arrows indicate potential response biases. Right panel: The Simon effect obtained in

Experiment 1. Reaction times and errors are presented as a function of compatibility condition and SOA.

Figure 2. Left and center panel: Illustrations of the tasks in the learning and test phase in Experiment 2, using an exemplary S-R mapping. The solid arrows indicate responses required by instruction and the dashed arrows indicate potential response biases. Right panel: The Simon effect for spatially biases and spatially ambiguous stimuli obtained in

Experiment 1. Reaction times and errors are presented as a function of compatibility condition and SOA.