Journal of Comparative Psychology 2015, Vol. 129, No. 4, 329 –333
© 2015 American Psychological Association 0735-7036/15/$12.00 http://dx.doi.org/10.1037/a0039450
The Generation Effect or Simply Generating an Effect? Jack Staniland, Michael Colombo, and Damian Scarf
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
University of Otago The Generation Effect is the phenomenon wherein attempting to retrieve or generate information from memory leads to better encoding and retention than passive rehearsal. Kornell and Terrace were the first to provide evidence for the Generation Effect in nonhuman animals, demonstrating that two rhesus monkeys performed markedly worse when tested following a passive learning condition relative to an active learning condition. In Experiment 1, using the same paradigm as Kornell and Terrace, we demonstrate that pigeons also display this effect. However, an assumption underlying the Generation Effect is that, under passive learning conditions, subjects will still display some evidence of learning but less than that displayed in active learning conditions. In Experiment 2, we examined this issue by pretraining pigeons on a list with hints and then comparing their acquisition of that same list to animals that did not receive any pretraining. Again, we found no evidence that pretraining on a list with hints conferred any advantage when learning that list without hints, a manipulation that Kornell and Terrace did not undertake. In summary, our data raise doubts about the evidence for the Generation Effect in nonhuman animals. Keywords: Generation Effect, passive learning, active learning, serial-order task
human participants. That is, the monkeys performed better on the test following the active (i.e., No-Hint) condition than the passive (i.e., Auto-Hint and Manual-Hint) conditions. However, an assumption underlying the Generation Effect is that even under passive learning conditions, participants will display some evidence of learning. This is the case in most, if not all, of the studies conducted with human participants. For example, in one of the first studies of the Generation Effect, Slamecka and Graf (1978) had participants either read or construct the second item of 100 word pairs. In the construct condition, participants were given the first letter of the second word (e.g., light-l____) and were asked to guess the correct word based on a rule (e.g., association rule: light-lamp). Recognition memory was then tested by presenting participants with the first word of each word pair and three other words, one of which was the correct second word. Again, although the Generation Effect was observed (e.g., association rule: read 71% vs. construct 88%), participants in the read condition still performed well above chance (1 in 3, 33%). With respect to Kornell and Terrace’s monkeys, given they performed at baseline following the Auto-Hint and Manual-Hint conditions, this finding does not appear to have been upheld. Does the absence of any learning in the passive conditions of Kornell and Terrace’s (2007) study weaken their claim to have demonstrated the Generation Effect in monkeys? We think it does. Although Kornell and Terrace accept that “hints produced no measurable learning” (p. 684), they also define the Generation Effect as the finding that “active retrieval of information from memory results in more learning than passive observation of the same information” (emphasis added) and accompany this definition with the Slamecka and Graf (1978) study described above, which clearly adheres to this definition. One possibility, however, is that Kornell and Terrace’s monkeys simply did not have enough time to demonstrate evidence of learning following the passive learning conditions. Indeed, Kornell and Terrace’s monkeys re-
The Generation Effect is the phenomenon wherein attempting to retrieve or generate information from memory leads to better encoding and retention than passive rehearsal (Jacoby, 1978; Slamecka & Graf, 1978). Kornell and Terrace (2007) were the first to provide evidence for the Generation Effect in nonhuman animals. Two rhesus monkeys learned several five-item lists trained under an active (i.e., No-Hint) condition, where the item order was discovered by trial and error, or under one of two passive conditions, where responses were guided by a frame (i.e., Hint) appearing around the correct item to press. In the Auto-Hint condition, the frame appeared automatically around the correct item, whereas in the Manual-Hint condition, monkeys had to request the hint. For each of the three conditions (No-Hint, Auto-Hint, and ManualHint), monkeys received three training sessions followed by a single test session where no hints were available. In both the Auto-Hint and Manual-Hint conditions, the monkeys performed extremely well during training, but their performance dropped dramatically on the test session. In fact, performance on the test sessions for both passive conditions was appreciably no different from baseline performance (i.e., first session performance on a novel list; Figure 1a). In contrast, as would be expected, performance on the test session following training in the No-Hint condition was markedly above baseline (Figure 1a). On the surface, the performance of Kornell and Terrace’s (2007) monkeys is consistent with the Generation Effect observed with
This article was published Online First July 6, 2015. Jack Staniland, Michael Colombo, and Damian Scarf, Department of Psychology, University of Otago. Correspondence concerning this article should be addressed to Damian Scarf and Michael Colombo, Department of Psychology, University of Otago, PO Box 56, Dunedin 9054, New Zealand. E-mail: damian@psy .otago.ac.nz and [email protected] 329
STANILAND, COLOMBO, AND SCARF
330
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Method
Figure 1. (a) The performance of the monkey Oberon on the Auto-Hint, Manual-Hint, and No-Hint conditions during training (Days 1–3) and test (Day 4). The monkey data are redrawn from Terrace and Kornell (2007). (b) The average performance of the eight pigeons on the Auto-Hint and No-Hint conditions during training (Days 1–10) and test (Day 11). The error bars represent 95% confidence intervals.
ceived only a single test session, giving them little opportunity to display evidence of learning. In the current study, we assess this possibility using pigeons as a model for Kornell and Terrace’s monkeys. First, we replicate Kornell and Terrace’s findings (Experiment 1) and then assess whether with extensive training there is any evidence of learning in the passive hint condition (Experiment 2).
Experiment 1 In Experiment 1, we trained pigeons on the No-Hint and AutoHint conditions used by Kornell and Terrace (2007). We excluded the Manual-Hint condition because Kornell and Terrace’s monkeys performed identically on the Manual-Hint and Auto-Hint conditions. Similar to Kornell and Terrace’s monkeys, the pigeons in the present study had extensive experience learning lists.
Subjects. The subjects were eight pigeons (Columba livia). Each pigeon was maintained at 85% of its free-feeding weight with grit and water provided ad lib in its home cage. The room in which the pigeons were housed was on 12-hr light-dark cycle, with overhead fluorescent lights turned on at 7 a.m. and turned off at 7 p.m. Apparatus and stimuli. Subjects were trained in standard operant chambers. The front wall of each chamber housed a Perspex panel with six 60-mm ⫻ 60-mm openings arranged in two rows of three. Each of the six openings provided access to a 15-in. LCD monitor on which the stimuli were displayed. Housed between the Perspex panel and LCD monitor was an infrared touch frame used to record subjects’ responses. The stimuli were 120pixel ⫻ 120-pixel gray-scale images (List A: stadium— car— radio; List B: helmet—fish— handbag). Procedure. Subjects were trained on two three-item lists, with one list trained using hints and the other list trained without hints. In the No-Hint condition, the list was learned by trial and error. The procedure was as follows. At the conclusion of a 10-s intertrial interval (ITI), the three list items were presented simultaneously on the screen. For ease of exposition, the three stimuli will be referred to as A, B, and C. All three list items remained on screen for the duration of each trial until either a correct response to all three items had been made or an error to one item occurred. Responses to individual stimuli were accompanied by a 100-ms feedback tone. A houselight situated on the rear wall of the chamber was on during the entire session except during the timeout period following an error. A correct response required pressing the three stimuli in the order A¡B¡C, after which the stimuli were turned off and the subject rewarded with a 2-s access to wheat via a hopper. Repeat responses to an item were permitted as long as they did not violate the prescribed order (e.g., A¡A¡B¡B¡B¡C). Any violation of the prescribed order was considered an error and resulted in trial termination and a 10-s timeout period during which the houselight was turned off, followed by the ITI. There were four types of errors a subject could make: three forward errors (¡B, ¡C, and A¡C) and one backward error (A¡B¡A). For lists trained with a hint, the hint consisted of a 3-mm black border that appeared around the correct item to press. When the item was pressed, the hint disappeared and reappeared around the next correct item to press. Once Item C was correctly pressed, the hint border disappeared and the subject was rewarded. With three stimuli and six possible on-screen positions, there were a total of 120 possible stimulus configurations. Each session consisted of 96 trials with the configuration on each trial drawn at random from the 120 possible configurations. Subjects received 10 training sessions followed by a single test session. The order of conditions was counterbalanced such that half of subjects received the AutoHint condition first followed by the No-Hint condition, whereas the other half received the No-Hint condition first followed by the Auto-Hint condition. The lists subjects were trained on (i.e., List A and List B) were counterbalanced across conditions.
Results and Discussion Figure 1b shows the performance of pigeons over the training (Sessions 1–10) and test (Session 11) sessions. Identical to Kornell
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
GENERATION EFFECT OR SIMPLY GENERATING AN EFFECT
and Terrace’s (2007) monkeys (Figure 1a), the performance of pigeons on the test session following the Auto-Hint condition was significantly lower than their performance in the final training session (M ⫽ 12.76% vs. M ⫽ 93.88%), t(7) ⫽ 20.24, p ⬍ .001, Cohen’s d ⫽ 7.157, and no different from baseline performance (M ⫽ 12.76% vs. M ⫽ 12.37%), t(7) ⫽ .094, p ⫽ .928, d ⫽ ⫺.033. In contrast, the performance of pigeons on the test session following the No-Hint condition was significantly higher than their performance in the final training session (M ⫽ 57.94% vs. M ⫽ 52.47%), t(7) ⫽ 3.19, p ⫽ .015, Cohen’s d ⫽ 1.129, and significantly above baseline (M ⫽ 57.94% vs. M ⫽ 12.37%), t(7) ⫽ 13.97, p ⬍ .001, d ⫽ 4.938. Finally, when directly compared to chance (.33 ⫻ .5 ⫻ .5 ⫽ .0825 or 8.25%), the performance of pigeons on the test session following the Auto-Hint condition was no different to chance, t(7) ⫽ 1.15, p ⫽ .288, Cohen’s d ⫽ .407, whereas their performance on the test session following the NoHint condition was significantly above chance, t(7) ⫽ 11.11, p ⬍ .001, Cohen’s d ⫽ 3.929. Our findings replicate those of Kornell and Terrace (2007). In both studies, performance on the test session following the NoHint condition was superior to performance on the test session following the Auto-Hint condition.
Experiment 2 Although superficially similar to the Generation Effect reported in studies with human subjects (see Bertsch, Pesta, Wiscott, & McDaniel, 2007, for a review), the drop to baseline performance levels on the test session following training on the Auto-Hint condition suggests our pigeons learned nothing about the list order from the hints. We suspect that during training with the hints, the pigeons may have simply adopted the strategy of following the hint border. In Experiment 2, we test this account by comparing the acquisition of a list under the No-Hint condition when it had been pretrained under the Auto-Hint condition and when it had not. If pigeons passively encode any information about the list order under the Auto-Hint condition, then one would expect that, when presented with that list without hints, they would learn that list at a faster rate than pigeons that did not receive the pretraining.
Method Subjects. The subjects were the same pigeons used in Experiment 1. Apparatus and stimuli. The apparatus was identical to that used in Experiment 1. The lists and images were as follows: List C, flower—iPad—insect; List D, snake—a person’s face—motorcycle; List E, a person’s arm— bird— house; and List F, dog and cat— bowtie—shirtless man. Procedure. The serial-order task procedure was that described in Experiment 1. The design was as follows. The subjects were divided into two groups, and both training and testing consisted of blocks of 10 sessions. In Phase 1, Group 1 was trained on List C (Auto-Hint condition) and then tested with List C again (No-Hint condition), whereas Group 2 was trained on List D (Auto-Hint condition) and then tested with List C (No-Hint condition) (see Table 1). The experiment was then repeated and the conditions for each group reversed. That is, in Phase 2, Group 1 was trained on List E (Auto-Hint condition) and then tested with List F (No-Hint
331
Table 1 The Order of Conditions in Experiment 2 Phase 1 Train
Phase 2 Test
Train
Test
Group 1 List C (hint) List C (no hint) List E (hint) List F (no hint) Group 2 List D (hint) List C (no hint) List F (hint) List F (no hint)
condition), whereas Group 2 was trained on List F (Auto-Hint condition) and then tested again with List F (No-Hint condition) (see Table 1). Data for the “Same” condition were generated by collapsing the data for Group 1 in Phase 1 and Group 2 in Phase 2, and data for the “Different” condition were generated by collapsing the data for Group 2 in Phase 1 and Group 1 in Phase 2.
Results and Discussion Figure 2 shows the performance of the pigeons in the Same and Different conditions. A three-way analysis of variance with Condition (2: Same, Different), Block (2: Training, Testing), and Session (10: 1–10) as factors revealed no effect of Condition, F(1, 7) ⫽ .126, p ⫽ .733, 2p ⫽ .018; a significant effect of Block, F(1, 7) ⫽ 350.822, p ⬍ .001, 2p ⫽ .98; but critically no Condition ⫻ Block interaction, F(1, 7) ⫽ .238, p ⫽ .641, 2p ⫽ .033, suggesting that training with hints conferred no advantage to learning that same list with no hints.
General Discussion The performance of our pigeons in Experiment 1 is identical to that displayed by Kornell and Terrace’s (2007) monkeys in that their performance was at baseline on the test session following training on the Auto-Hint condition. Experiment 2 confirmed that, rather than this drop being a temporary lapse, it was because subjects acquired no knowledge of the list order under the Auto-Hint (i.e., passive learning) condition. We would argue that if trained on a comparable procedure to that employed in Experiment 2, Kornell and Terrace’s monkeys would have displayed a similar pattern of behavior. Of course, differences between Kornell and Terrace’s (2007) study and the present study require us to be cautious when stating that their monkeys would have behaved in a similar manner to our pigeons. For example, with respect to the procedural differences, Kornell and Terrace trained their monkeys on five-item lists composed of colored images, whereas we used three-item lists composed of gray-scale images. Although gray-scale images may be more difficult to distinguish than color images, our pigeons displayed acquisition curves comparable to what we have observed in previous studies employing three-item lists of colored stimuli (e.g., Scarf & Colombo, 2010, 2011). With respect to the issue of three-item versus five-item lists, our use of shorter lists was an attempt to equate the number of trials pigeons and monkeys required to acquire their three- and five-item lists, respectively. As the training phase of Figure 1 clearly shows, however, the pigeons still required markedly more sessions to reach a similar level of performance to that displayed by Kornell and Terrace’s monkeys after just three sessions. Although one could argue that the combination of gray-scale images and difficulty acquiring the three-
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
332
STANILAND, COLOMBO, AND SCARF
From our perspective, what may make designing a valid Generation Effect paradigm difficult is the high likelihood that the hint will attain stimulus control. Indeed, unless the animal can foresee that the strategy of following the hint will be maladaptive in the future (i.e., when tested without hints), it will likely always learn to simply follow the hints. Generation Effect studies with nonhuman animals, therefore, may be limited by the fact that nonhuman animals display only very rudimentary planning and foresight abilities (Beran, Pate, Washburn, & Rumbaugh, 2004; Bourjade, Thierry, Call, & Dufour, 2012; Dekleva, van den Berg, Spruijt, & Sterck, 2012; Paxton & Hampton, 2009; Scarf & Colombo, 2009; Scarf, Danly, Morgan, Colombo, & Terrace, 2011; see Scarf, Smith, & Stuart, 2014, for a review). One potential solution to this problem is for future studies to take a learning set approach (Harlow, 1949), transferring animals from a hint condition to a test condition across several lists or problems, to help instill in animals the future payoff of learning under the passive learning condition. Figure 2. The average performance of the eight pigeons under the Same and Different conditions. The error bars represent 95% confidence intervals.
item lists may have made our pigeons more reliant on the hints than Kornell and Terrace’s monkeys, the performance of the monkeys in the 50% Hint condition suggests they were also heavily reliant on the hints. Indeed, on trials where no hints were available, the monkeys performed extremely poorly (Oberon: 1%, 0%, and 0%; Macduff 2: 0%, 3%, and 9%) despite performing extremely well on the same list when hints were available. One might still argue that although our pigeons failed to display the Generation Effect, Kornell and Terrace’s (2007) monkeys would have. That is, despite the clear correspondence demonstrated in Experiment 1 and the performance of the monkeys in the 50% Hint condition, it is still possible that the monkeys would have behaved differently from our pigeons if run on a comparable procedure to that employed in Experiment 2. This argument, however, fails to acknowledge the ever growing body of data demonstrating the clear similarity between the cognitive abilities of pigeons and monkeys (Macphail & Barlow, 1985; Scarf & Colombo, 2010, 2011; Scarf, Hayne, & Colombo, 2011; Wright, 1997). In the end, we believe our findings shift the burden of proof to those conducting work with monkeys to demonstrate that monkeys would display an outcome different from that reported in Experiment 2. If the Generation Effect does not underlie the difference in performance following the active and passive learning conditions in our study and Kornell and Terrace’s (2007) study, then what does it reflect? A strong possibility is that in both studies, the border attained stimulus control and, rather than it serving as a hint indicating the correct item to respond to, it simply served as a discriminative stimulus, with subjects adopting the strategy of following the border (Herrnstein, 1990). Another possibility is that the hint, due to its salience, overshadowed processing of the picture stimulus (Spear & Kucharski, 1984). Although these alternative accounts may be considered “killjoy” (Shettleworth, 2010), we acknowledge that the commonly used epigram, absence of evidence is not evidence of absence, still holds true and the issue of whether the Generation Effect can be demonstrated in nonhuman animals is still an open question.
References Beran, M. J., Pate, J. L., Washburn, D. A., & Rumbaugh, D. M. (2004). Sequential responding and planning in chimpanzees (Pan troglodytes) and rhesus macaques (Macaca mulatta). Journal of Experimental Psychology: Animal Behavior Processes, 30, 203–212. http://dx.doi.org/ 10.1037/0097-7403.30.3.203 Bertsch, S., Pesta, B. J., Wiscott, R., & McDaniel, M. A. (2007). The Generation Effect: A meta-analytic review. Memory & Cognition, 35, 201–210. http://dx.doi.org/10.3758/BF03193441 Bourjade, M., Thierry, B., Call, J., & Dufour, V. (2012). Are monkeys able to plan for future exchange? Animal Cognition, 15, 783–795. http://dx .doi.org/10.1007/s10071-012-0502-1 Dekleva, M., van den Berg, L., Spruijt, B. M., & Sterck, E. H. (2012). Take it or leave it: Transport of tools for future use by long-tailed macaques (Macaca fascicularis). Behavioural Processes, 90, 392– 401. http://dx .doi.org/10.1016/j.beproc.2012.04.003 Harlow, H. F. (1949). The formation of learning sets. Psychological Review, 56, 51– 65. http://dx.doi.org/10.1037/h0062474 Herrnstein, R. J. (1990). Levels of stimulus control: A functional approach. Cognition, 37, 133–166. http://dx.doi.org/10.1016/0010-0277(90) 90021-B Jacoby, L. L. (1978). On interpreting the effects of repetition: Solving a problem versus remembering a solution. Journal of Verbal Learning & Verbal Behavior, 17, 649 – 667. http://dx.doi.org/10.1016/S0022-5371 (78)90393-6 Kornell, N., & Terrace, H. S. (2007). The Generation Effect in monkeys. Psychological Science, 18, 682– 685. http://dx.doi.org/10.1111/j.14679280.2007.01959.x Macphail, E., & Barlow, H. B. (1985). Vertebrate intelligence: The null hypothesis. Philosophical Transactions of the Royal Society of London Series B, Biological Sciences, 308, 37–51. http://dx.doi.org/10.1098/rstb .1985.0008 Paxton, R., & Hampton, R. R. (2009). Tests of planning and the BischofKöhler hypothesis in rhesus monkeys (Macaca mulatta). Behavioural Processes, 80, 238 –246. http://dx.doi.org/10.1016/j.beproc.2008.12.016 Scarf, D., & Colombo, M. (2009). Eye movements during list execution reveal no planning in monkeys (Macaca fascicularis). Journal of Experimental Psychology: Animal Behavior Processes, 35, 587–592. http://dx.doi.org/10.1037/a0014020 Scarf, D., & Colombo, M. (2010). Representation of serial order in pigeons (Columba livia). Journal of Experimental Psychology: Animal Behavior Processes, 36, 423– 429. http://dx.doi.org/10.1037/a0020926
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
GENERATION EFFECT OR SIMPLY GENERATING AN EFFECT Scarf, D., & Colombo, M. (2011). Knowledge of the ordinal position of list items in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 37, 483– 487. http://dx.doi.org/10.1037/a0023695 Scarf, D., Danly, E., Morgan, G., Colombo, M., & Terrace, H. S. (2011). Sequential planning in rhesus monkeys (Macaca mulatta). Animal Cognition, 14, 317–324. http://dx.doi.org/10.1007/s10071-010-0365-2 Scarf, D., Hayne, H., & Colombo, M. (2011). Pigeons on par with primates in numerical competence. Science, 334, 1664 –1664. http://dx.doi.org/ 10.1126/science.1213357 Scarf, D., Smith, C., & Stuart, M. (2014). A spoon full of studies helps the comparison go down: A comparative analysis of Tulving’s spoon test. Frontiers in Psychology, 5, 893. http://dx.doi.org/10.3389/fpsyg.2014 .00893 Shettleworth, S. J. (2010). Clever animals and killjoy explanations in comparative psychology. Trends in Cognitive Sciences, 14, 477– 481. http://dx.doi.org/10.1016/j.tics.2010.07.002
333
Slamecka, N. J., & Graf, P. (1978). The Generation Effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4, 592– 604. http://dx.doi.org/10.1037/0278-7393.4.6.592 Spear, N., & Kucharski, D. (1984). Ontogenetic differences in the processing of multi-element stimuli: Potentiation and overshadowing. In H. Roitblat, T. Bever, & H. Terrace (Eds.), Animal cognition (pp. 545–567). Hillsdale, NJ: Erlbaum. Wright, A. A. (1997). Concept learning and learning strategies. Psychological Science, 8, 119 –123. http://dx.doi.org/10.1111/j.1467-9280.1997 .tb00693.x
Received January 6, 2015 Revision received April 15, 2015 Accepted April 22, 2015 䡲
E-Mail Notification of Your Latest Issue Online! Would you like to know when the next issue of your favorite APA journal will be available online? This service is now available to you. Sign up at http://notify.apa.org/ and you will be notified by e-mail when issues of interest to you become available!
© 2015 American Psychological Association 0735-7036/15/$12.00 http://dx.doi.org/10.1037/a0039450
The Generation Effect or Simply Generating an Effect? Jack Staniland, Michael Colombo, and Damian Scarf
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
University of Otago The Generation Effect is the phenomenon wherein attempting to retrieve or generate information from memory leads to better encoding and retention than passive rehearsal. Kornell and Terrace were the first to provide evidence for the Generation Effect in nonhuman animals, demonstrating that two rhesus monkeys performed markedly worse when tested following a passive learning condition relative to an active learning condition. In Experiment 1, using the same paradigm as Kornell and Terrace, we demonstrate that pigeons also display this effect. However, an assumption underlying the Generation Effect is that, under passive learning conditions, subjects will still display some evidence of learning but less than that displayed in active learning conditions. In Experiment 2, we examined this issue by pretraining pigeons on a list with hints and then comparing their acquisition of that same list to animals that did not receive any pretraining. Again, we found no evidence that pretraining on a list with hints conferred any advantage when learning that list without hints, a manipulation that Kornell and Terrace did not undertake. In summary, our data raise doubts about the evidence for the Generation Effect in nonhuman animals. Keywords: Generation Effect, passive learning, active learning, serial-order task
human participants. That is, the monkeys performed better on the test following the active (i.e., No-Hint) condition than the passive (i.e., Auto-Hint and Manual-Hint) conditions. However, an assumption underlying the Generation Effect is that even under passive learning conditions, participants will display some evidence of learning. This is the case in most, if not all, of the studies conducted with human participants. For example, in one of the first studies of the Generation Effect, Slamecka and Graf (1978) had participants either read or construct the second item of 100 word pairs. In the construct condition, participants were given the first letter of the second word (e.g., light-l____) and were asked to guess the correct word based on a rule (e.g., association rule: light-lamp). Recognition memory was then tested by presenting participants with the first word of each word pair and three other words, one of which was the correct second word. Again, although the Generation Effect was observed (e.g., association rule: read 71% vs. construct 88%), participants in the read condition still performed well above chance (1 in 3, 33%). With respect to Kornell and Terrace’s monkeys, given they performed at baseline following the Auto-Hint and Manual-Hint conditions, this finding does not appear to have been upheld. Does the absence of any learning in the passive conditions of Kornell and Terrace’s (2007) study weaken their claim to have demonstrated the Generation Effect in monkeys? We think it does. Although Kornell and Terrace accept that “hints produced no measurable learning” (p. 684), they also define the Generation Effect as the finding that “active retrieval of information from memory results in more learning than passive observation of the same information” (emphasis added) and accompany this definition with the Slamecka and Graf (1978) study described above, which clearly adheres to this definition. One possibility, however, is that Kornell and Terrace’s monkeys simply did not have enough time to demonstrate evidence of learning following the passive learning conditions. Indeed, Kornell and Terrace’s monkeys re-
The Generation Effect is the phenomenon wherein attempting to retrieve or generate information from memory leads to better encoding and retention than passive rehearsal (Jacoby, 1978; Slamecka & Graf, 1978). Kornell and Terrace (2007) were the first to provide evidence for the Generation Effect in nonhuman animals. Two rhesus monkeys learned several five-item lists trained under an active (i.e., No-Hint) condition, where the item order was discovered by trial and error, or under one of two passive conditions, where responses were guided by a frame (i.e., Hint) appearing around the correct item to press. In the Auto-Hint condition, the frame appeared automatically around the correct item, whereas in the Manual-Hint condition, monkeys had to request the hint. For each of the three conditions (No-Hint, Auto-Hint, and ManualHint), monkeys received three training sessions followed by a single test session where no hints were available. In both the Auto-Hint and Manual-Hint conditions, the monkeys performed extremely well during training, but their performance dropped dramatically on the test session. In fact, performance on the test sessions for both passive conditions was appreciably no different from baseline performance (i.e., first session performance on a novel list; Figure 1a). In contrast, as would be expected, performance on the test session following training in the No-Hint condition was markedly above baseline (Figure 1a). On the surface, the performance of Kornell and Terrace’s (2007) monkeys is consistent with the Generation Effect observed with
This article was published Online First July 6, 2015. Jack Staniland, Michael Colombo, and Damian Scarf, Department of Psychology, University of Otago. Correspondence concerning this article should be addressed to Damian Scarf and Michael Colombo, Department of Psychology, University of Otago, PO Box 56, Dunedin 9054, New Zealand. E-mail: damian@psy .otago.ac.nz and [email protected] 329
STANILAND, COLOMBO, AND SCARF
330
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Method
Figure 1. (a) The performance of the monkey Oberon on the Auto-Hint, Manual-Hint, and No-Hint conditions during training (Days 1–3) and test (Day 4). The monkey data are redrawn from Terrace and Kornell (2007). (b) The average performance of the eight pigeons on the Auto-Hint and No-Hint conditions during training (Days 1–10) and test (Day 11). The error bars represent 95% confidence intervals.
ceived only a single test session, giving them little opportunity to display evidence of learning. In the current study, we assess this possibility using pigeons as a model for Kornell and Terrace’s monkeys. First, we replicate Kornell and Terrace’s findings (Experiment 1) and then assess whether with extensive training there is any evidence of learning in the passive hint condition (Experiment 2).
Experiment 1 In Experiment 1, we trained pigeons on the No-Hint and AutoHint conditions used by Kornell and Terrace (2007). We excluded the Manual-Hint condition because Kornell and Terrace’s monkeys performed identically on the Manual-Hint and Auto-Hint conditions. Similar to Kornell and Terrace’s monkeys, the pigeons in the present study had extensive experience learning lists.
Subjects. The subjects were eight pigeons (Columba livia). Each pigeon was maintained at 85% of its free-feeding weight with grit and water provided ad lib in its home cage. The room in which the pigeons were housed was on 12-hr light-dark cycle, with overhead fluorescent lights turned on at 7 a.m. and turned off at 7 p.m. Apparatus and stimuli. Subjects were trained in standard operant chambers. The front wall of each chamber housed a Perspex panel with six 60-mm ⫻ 60-mm openings arranged in two rows of three. Each of the six openings provided access to a 15-in. LCD monitor on which the stimuli were displayed. Housed between the Perspex panel and LCD monitor was an infrared touch frame used to record subjects’ responses. The stimuli were 120pixel ⫻ 120-pixel gray-scale images (List A: stadium— car— radio; List B: helmet—fish— handbag). Procedure. Subjects were trained on two three-item lists, with one list trained using hints and the other list trained without hints. In the No-Hint condition, the list was learned by trial and error. The procedure was as follows. At the conclusion of a 10-s intertrial interval (ITI), the three list items were presented simultaneously on the screen. For ease of exposition, the three stimuli will be referred to as A, B, and C. All three list items remained on screen for the duration of each trial until either a correct response to all three items had been made or an error to one item occurred. Responses to individual stimuli were accompanied by a 100-ms feedback tone. A houselight situated on the rear wall of the chamber was on during the entire session except during the timeout period following an error. A correct response required pressing the three stimuli in the order A¡B¡C, after which the stimuli were turned off and the subject rewarded with a 2-s access to wheat via a hopper. Repeat responses to an item were permitted as long as they did not violate the prescribed order (e.g., A¡A¡B¡B¡B¡C). Any violation of the prescribed order was considered an error and resulted in trial termination and a 10-s timeout period during which the houselight was turned off, followed by the ITI. There were four types of errors a subject could make: three forward errors (¡B, ¡C, and A¡C) and one backward error (A¡B¡A). For lists trained with a hint, the hint consisted of a 3-mm black border that appeared around the correct item to press. When the item was pressed, the hint disappeared and reappeared around the next correct item to press. Once Item C was correctly pressed, the hint border disappeared and the subject was rewarded. With three stimuli and six possible on-screen positions, there were a total of 120 possible stimulus configurations. Each session consisted of 96 trials with the configuration on each trial drawn at random from the 120 possible configurations. Subjects received 10 training sessions followed by a single test session. The order of conditions was counterbalanced such that half of subjects received the AutoHint condition first followed by the No-Hint condition, whereas the other half received the No-Hint condition first followed by the Auto-Hint condition. The lists subjects were trained on (i.e., List A and List B) were counterbalanced across conditions.
Results and Discussion Figure 1b shows the performance of pigeons over the training (Sessions 1–10) and test (Session 11) sessions. Identical to Kornell
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
GENERATION EFFECT OR SIMPLY GENERATING AN EFFECT
and Terrace’s (2007) monkeys (Figure 1a), the performance of pigeons on the test session following the Auto-Hint condition was significantly lower than their performance in the final training session (M ⫽ 12.76% vs. M ⫽ 93.88%), t(7) ⫽ 20.24, p ⬍ .001, Cohen’s d ⫽ 7.157, and no different from baseline performance (M ⫽ 12.76% vs. M ⫽ 12.37%), t(7) ⫽ .094, p ⫽ .928, d ⫽ ⫺.033. In contrast, the performance of pigeons on the test session following the No-Hint condition was significantly higher than their performance in the final training session (M ⫽ 57.94% vs. M ⫽ 52.47%), t(7) ⫽ 3.19, p ⫽ .015, Cohen’s d ⫽ 1.129, and significantly above baseline (M ⫽ 57.94% vs. M ⫽ 12.37%), t(7) ⫽ 13.97, p ⬍ .001, d ⫽ 4.938. Finally, when directly compared to chance (.33 ⫻ .5 ⫻ .5 ⫽ .0825 or 8.25%), the performance of pigeons on the test session following the Auto-Hint condition was no different to chance, t(7) ⫽ 1.15, p ⫽ .288, Cohen’s d ⫽ .407, whereas their performance on the test session following the NoHint condition was significantly above chance, t(7) ⫽ 11.11, p ⬍ .001, Cohen’s d ⫽ 3.929. Our findings replicate those of Kornell and Terrace (2007). In both studies, performance on the test session following the NoHint condition was superior to performance on the test session following the Auto-Hint condition.
Experiment 2 Although superficially similar to the Generation Effect reported in studies with human subjects (see Bertsch, Pesta, Wiscott, & McDaniel, 2007, for a review), the drop to baseline performance levels on the test session following training on the Auto-Hint condition suggests our pigeons learned nothing about the list order from the hints. We suspect that during training with the hints, the pigeons may have simply adopted the strategy of following the hint border. In Experiment 2, we test this account by comparing the acquisition of a list under the No-Hint condition when it had been pretrained under the Auto-Hint condition and when it had not. If pigeons passively encode any information about the list order under the Auto-Hint condition, then one would expect that, when presented with that list without hints, they would learn that list at a faster rate than pigeons that did not receive the pretraining.
Method Subjects. The subjects were the same pigeons used in Experiment 1. Apparatus and stimuli. The apparatus was identical to that used in Experiment 1. The lists and images were as follows: List C, flower—iPad—insect; List D, snake—a person’s face—motorcycle; List E, a person’s arm— bird— house; and List F, dog and cat— bowtie—shirtless man. Procedure. The serial-order task procedure was that described in Experiment 1. The design was as follows. The subjects were divided into two groups, and both training and testing consisted of blocks of 10 sessions. In Phase 1, Group 1 was trained on List C (Auto-Hint condition) and then tested with List C again (No-Hint condition), whereas Group 2 was trained on List D (Auto-Hint condition) and then tested with List C (No-Hint condition) (see Table 1). The experiment was then repeated and the conditions for each group reversed. That is, in Phase 2, Group 1 was trained on List E (Auto-Hint condition) and then tested with List F (No-Hint
331
Table 1 The Order of Conditions in Experiment 2 Phase 1 Train
Phase 2 Test
Train
Test
Group 1 List C (hint) List C (no hint) List E (hint) List F (no hint) Group 2 List D (hint) List C (no hint) List F (hint) List F (no hint)
condition), whereas Group 2 was trained on List F (Auto-Hint condition) and then tested again with List F (No-Hint condition) (see Table 1). Data for the “Same” condition were generated by collapsing the data for Group 1 in Phase 1 and Group 2 in Phase 2, and data for the “Different” condition were generated by collapsing the data for Group 2 in Phase 1 and Group 1 in Phase 2.
Results and Discussion Figure 2 shows the performance of the pigeons in the Same and Different conditions. A three-way analysis of variance with Condition (2: Same, Different), Block (2: Training, Testing), and Session (10: 1–10) as factors revealed no effect of Condition, F(1, 7) ⫽ .126, p ⫽ .733, 2p ⫽ .018; a significant effect of Block, F(1, 7) ⫽ 350.822, p ⬍ .001, 2p ⫽ .98; but critically no Condition ⫻ Block interaction, F(1, 7) ⫽ .238, p ⫽ .641, 2p ⫽ .033, suggesting that training with hints conferred no advantage to learning that same list with no hints.
General Discussion The performance of our pigeons in Experiment 1 is identical to that displayed by Kornell and Terrace’s (2007) monkeys in that their performance was at baseline on the test session following training on the Auto-Hint condition. Experiment 2 confirmed that, rather than this drop being a temporary lapse, it was because subjects acquired no knowledge of the list order under the Auto-Hint (i.e., passive learning) condition. We would argue that if trained on a comparable procedure to that employed in Experiment 2, Kornell and Terrace’s monkeys would have displayed a similar pattern of behavior. Of course, differences between Kornell and Terrace’s (2007) study and the present study require us to be cautious when stating that their monkeys would have behaved in a similar manner to our pigeons. For example, with respect to the procedural differences, Kornell and Terrace trained their monkeys on five-item lists composed of colored images, whereas we used three-item lists composed of gray-scale images. Although gray-scale images may be more difficult to distinguish than color images, our pigeons displayed acquisition curves comparable to what we have observed in previous studies employing three-item lists of colored stimuli (e.g., Scarf & Colombo, 2010, 2011). With respect to the issue of three-item versus five-item lists, our use of shorter lists was an attempt to equate the number of trials pigeons and monkeys required to acquire their three- and five-item lists, respectively. As the training phase of Figure 1 clearly shows, however, the pigeons still required markedly more sessions to reach a similar level of performance to that displayed by Kornell and Terrace’s monkeys after just three sessions. Although one could argue that the combination of gray-scale images and difficulty acquiring the three-
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
332
STANILAND, COLOMBO, AND SCARF
From our perspective, what may make designing a valid Generation Effect paradigm difficult is the high likelihood that the hint will attain stimulus control. Indeed, unless the animal can foresee that the strategy of following the hint will be maladaptive in the future (i.e., when tested without hints), it will likely always learn to simply follow the hints. Generation Effect studies with nonhuman animals, therefore, may be limited by the fact that nonhuman animals display only very rudimentary planning and foresight abilities (Beran, Pate, Washburn, & Rumbaugh, 2004; Bourjade, Thierry, Call, & Dufour, 2012; Dekleva, van den Berg, Spruijt, & Sterck, 2012; Paxton & Hampton, 2009; Scarf & Colombo, 2009; Scarf, Danly, Morgan, Colombo, & Terrace, 2011; see Scarf, Smith, & Stuart, 2014, for a review). One potential solution to this problem is for future studies to take a learning set approach (Harlow, 1949), transferring animals from a hint condition to a test condition across several lists or problems, to help instill in animals the future payoff of learning under the passive learning condition. Figure 2. The average performance of the eight pigeons under the Same and Different conditions. The error bars represent 95% confidence intervals.
item lists may have made our pigeons more reliant on the hints than Kornell and Terrace’s monkeys, the performance of the monkeys in the 50% Hint condition suggests they were also heavily reliant on the hints. Indeed, on trials where no hints were available, the monkeys performed extremely poorly (Oberon: 1%, 0%, and 0%; Macduff 2: 0%, 3%, and 9%) despite performing extremely well on the same list when hints were available. One might still argue that although our pigeons failed to display the Generation Effect, Kornell and Terrace’s (2007) monkeys would have. That is, despite the clear correspondence demonstrated in Experiment 1 and the performance of the monkeys in the 50% Hint condition, it is still possible that the monkeys would have behaved differently from our pigeons if run on a comparable procedure to that employed in Experiment 2. This argument, however, fails to acknowledge the ever growing body of data demonstrating the clear similarity between the cognitive abilities of pigeons and monkeys (Macphail & Barlow, 1985; Scarf & Colombo, 2010, 2011; Scarf, Hayne, & Colombo, 2011; Wright, 1997). In the end, we believe our findings shift the burden of proof to those conducting work with monkeys to demonstrate that monkeys would display an outcome different from that reported in Experiment 2. If the Generation Effect does not underlie the difference in performance following the active and passive learning conditions in our study and Kornell and Terrace’s (2007) study, then what does it reflect? A strong possibility is that in both studies, the border attained stimulus control and, rather than it serving as a hint indicating the correct item to respond to, it simply served as a discriminative stimulus, with subjects adopting the strategy of following the border (Herrnstein, 1990). Another possibility is that the hint, due to its salience, overshadowed processing of the picture stimulus (Spear & Kucharski, 1984). Although these alternative accounts may be considered “killjoy” (Shettleworth, 2010), we acknowledge that the commonly used epigram, absence of evidence is not evidence of absence, still holds true and the issue of whether the Generation Effect can be demonstrated in nonhuman animals is still an open question.
References Beran, M. J., Pate, J. L., Washburn, D. A., & Rumbaugh, D. M. (2004). Sequential responding and planning in chimpanzees (Pan troglodytes) and rhesus macaques (Macaca mulatta). Journal of Experimental Psychology: Animal Behavior Processes, 30, 203–212. http://dx.doi.org/ 10.1037/0097-7403.30.3.203 Bertsch, S., Pesta, B. J., Wiscott, R., & McDaniel, M. A. (2007). The Generation Effect: A meta-analytic review. Memory & Cognition, 35, 201–210. http://dx.doi.org/10.3758/BF03193441 Bourjade, M., Thierry, B., Call, J., & Dufour, V. (2012). Are monkeys able to plan for future exchange? Animal Cognition, 15, 783–795. http://dx .doi.org/10.1007/s10071-012-0502-1 Dekleva, M., van den Berg, L., Spruijt, B. M., & Sterck, E. H. (2012). Take it or leave it: Transport of tools for future use by long-tailed macaques (Macaca fascicularis). Behavioural Processes, 90, 392– 401. http://dx .doi.org/10.1016/j.beproc.2012.04.003 Harlow, H. F. (1949). The formation of learning sets. Psychological Review, 56, 51– 65. http://dx.doi.org/10.1037/h0062474 Herrnstein, R. J. (1990). Levels of stimulus control: A functional approach. Cognition, 37, 133–166. http://dx.doi.org/10.1016/0010-0277(90) 90021-B Jacoby, L. L. (1978). On interpreting the effects of repetition: Solving a problem versus remembering a solution. Journal of Verbal Learning & Verbal Behavior, 17, 649 – 667. http://dx.doi.org/10.1016/S0022-5371 (78)90393-6 Kornell, N., & Terrace, H. S. (2007). The Generation Effect in monkeys. Psychological Science, 18, 682– 685. http://dx.doi.org/10.1111/j.14679280.2007.01959.x Macphail, E., & Barlow, H. B. (1985). Vertebrate intelligence: The null hypothesis. Philosophical Transactions of the Royal Society of London Series B, Biological Sciences, 308, 37–51. http://dx.doi.org/10.1098/rstb .1985.0008 Paxton, R., & Hampton, R. R. (2009). Tests of planning and the BischofKöhler hypothesis in rhesus monkeys (Macaca mulatta). Behavioural Processes, 80, 238 –246. http://dx.doi.org/10.1016/j.beproc.2008.12.016 Scarf, D., & Colombo, M. (2009). Eye movements during list execution reveal no planning in monkeys (Macaca fascicularis). Journal of Experimental Psychology: Animal Behavior Processes, 35, 587–592. http://dx.doi.org/10.1037/a0014020 Scarf, D., & Colombo, M. (2010). Representation of serial order in pigeons (Columba livia). Journal of Experimental Psychology: Animal Behavior Processes, 36, 423– 429. http://dx.doi.org/10.1037/a0020926
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
GENERATION EFFECT OR SIMPLY GENERATING AN EFFECT Scarf, D., & Colombo, M. (2011). Knowledge of the ordinal position of list items in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 37, 483– 487. http://dx.doi.org/10.1037/a0023695 Scarf, D., Danly, E., Morgan, G., Colombo, M., & Terrace, H. S. (2011). Sequential planning in rhesus monkeys (Macaca mulatta). Animal Cognition, 14, 317–324. http://dx.doi.org/10.1007/s10071-010-0365-2 Scarf, D., Hayne, H., & Colombo, M. (2011). Pigeons on par with primates in numerical competence. Science, 334, 1664 –1664. http://dx.doi.org/ 10.1126/science.1213357 Scarf, D., Smith, C., & Stuart, M. (2014). A spoon full of studies helps the comparison go down: A comparative analysis of Tulving’s spoon test. Frontiers in Psychology, 5, 893. http://dx.doi.org/10.3389/fpsyg.2014 .00893 Shettleworth, S. J. (2010). Clever animals and killjoy explanations in comparative psychology. Trends in Cognitive Sciences, 14, 477– 481. http://dx.doi.org/10.1016/j.tics.2010.07.002
333
Slamecka, N. J., & Graf, P. (1978). The Generation Effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4, 592– 604. http://dx.doi.org/10.1037/0278-7393.4.6.592 Spear, N., & Kucharski, D. (1984). Ontogenetic differences in the processing of multi-element stimuli: Potentiation and overshadowing. In H. Roitblat, T. Bever, & H. Terrace (Eds.), Animal cognition (pp. 545–567). Hillsdale, NJ: Erlbaum. Wright, A. A. (1997). Concept learning and learning strategies. Psychological Science, 8, 119 –123. http://dx.doi.org/10.1111/j.1467-9280.1997 .tb00693.x
Received January 6, 2015 Revision received April 15, 2015 Accepted April 22, 2015 䡲
E-Mail Notification of Your Latest Issue Online! Would you like to know when the next issue of your favorite APA journal will be available online? This service is now available to you. Sign up at http://notify.apa.org/ and you will be notified by e-mail when issues of interest to you become available!
