A Method for Studying Imagery in Animals

By: Julie J. Neiworth and Mark E. Rilling

The purpose of these studies were to try and determine if it was possible to confirm that animals think in a series of pictures, enabling them to hold and manipulate an image in their heads. It had already been suggested that, by default, animals must be thinking in a series of pictures since they do not have language (as we know it), but in the field of cognitive science nothing can be assumed, it must be proven.

The test on pigeons was modeled after a human imagery, where the subjects had to determine if an object had moved at the correct velocity while it could not been seen. These experiments utilized a clock design where one of the ‘hands’ would move at 90 degrees/second, either visible (perceptual) the whole time or missing (imagery) for 0.5-1 second. Violation trials meant that the hand had either traveled too far or not far enough. The pigeons had to peck a left key if the hand had traveled the appropriate distance and a right key if the hand had not traveled the appropriate distance.

The first experiment initially trained pigeons to discriminate correct and incorrect presentations of a stimulus rotating from 0 to 135 degree position. During imagery trials, the hand would disappear at the 90 degree mark for .5-1 second, with the hand always reappearing at the 135 degree mark. This presentation would be correct after .5 seconds, but incorrect after the 1 second trial, since the hand should have rotated to the 180 degree mark in that time. When pigeons were correctly identifying the trials at 80% accuracy, the novel location was introduced (158 degrees). At this point in the training, pigeons were using a simple discrimination, if it was a perceptual or short delay trial, peck left, if it was a long delay trial, peck right. This carried over to this new location, with longer delays leading the pigeon to peck right, even though the stimulus was in the correct location.

After the initial transfer test of 158 degrees, the pigeons were trained to discriminate correct and incorrect presentations of a stimulus rotating from 0-180 degrees.This way the timing contingencies were reversed, with .5 seconds being a violation trial and 1 second being a correct trial. When the pigeons reached 80% accuracy on these trials, pigeons experienced alternating sessions of 135 or 180 trials.

After pigeons were able to successfully alternate sessions of these two trials, they were able to transfer this discrimination to a novel location (158 degrees) after the first trial, indicating that they had learned the underlying rule. However, since the pigeons had been previously exposed to the 158 degree trials and received extensive practice on the 135 and 180 degree trials the successful transfer could have been the result of this practice.

Experiment 2 consisted of a similar paradigm where pigeons were first trained on 135 degree trials, then 180, and finally alternating sessions before being tested on 158 degree trials. When pigeons finally received the transfer test, which consisted of a novel location and novel delay, the pigeons could discriminate these novelties at the same accuracy of the training stimuli.

However, these results could still be the result of learning specific stimulus-time delay relationships, with a series of if-then rules (if a long delay and the stimulus stops at 180, then peck the left key).

Experiment 3 followed a similar procedure, utilizing birds that had been trained on 135, 158, 180 degree discrimination procedures. This final experiment was to determine if they could extrapolate this cognitive rule outside of the originally trained times. The results from the 158 and 202 transfer tests were analyzed for forward and backward errors, based on additional research from human work. Forward errors are when the pigeons would overestimate how far the stimulus had traveled and backward errors would underestimate how far the stimulus had traveled during delay periods. So during a .5 second delay, if a pigeon would  think that the stimulus had actually been able to travel to the correct location, where it would actually only be able to travel to the 135 mark.

Forward errors were predicted if the pigeons were using imagery since the momentum of the object was being accounted for. This also implies that a similar brain region is activated during both kinds of visual processes, whether it be through imagery or direct visual stimulus. The proportion of forward errors was greater in the 158 trials than the 202, and for 202 trials the difference between forward errors and imagery errors was not significant.  I  interpreted the imagery error score as the total number of errors on imagery trials.

While the series of experiments was interesting, I found the use of left versus right response a little confusing, especially when it came to error analysis. I also thought that inability to alternate 135 and 180 within session indicated that the birds were possible encoding the presentations differently or were at least not as flexible with their imagery. However, I do wonder how these would change after being exposed to more exemplars or if the stimulus had a smoother motion movement. Finally, a different study in pigeons investigating recognition of 3D objects that had been rotated does seem to corroborate these finding, that pigeons can use imagery to manipulate visual stimuli.

The Special Status of Actions in Causal Reasoning in Rats

By: Kenneth J. Leising, Jared Wong, Michael R. Waldmann and Aaron P. Blaisdell

Being able to identify cause-effect relationships is so intuitive it takes relatively few trials to determine how these relationships interact with other variables. When independent action is added to this mix, it is even easier for a person to confidently state whether or not their actions were the cause of said effect. As this paper demonstrates, humans may not be alone in this assumption.

This research is a further exploration of how rats determine causality, an area of cognition that has been strictly in the human domain for some time now. Some researchers argue that rats can only form associations between events which can resemble causality.This series of studies explores how actions influence the perception of causality and how this differs from predictions set by the associative model.

Experiment 1 had 4 groups of rats, (Intervene, Observe1, Exogenous cue and Observe2) that interacted with 3 or 4 different stimuli. A, X, and B were all neutral cues that consisted of a flashing light, tone, and click train respectively. F was a food reward of sucrose. All rats were trained on A-X in phase 1 and A-F in phase 2, thus experience a common cause model, where X and F are caused by A. At test the Intervene group had a lever press bar in the chamber that when activated created X, to which all the other groups were yoked. The Observe1 group also experience X whenever the Intervene group pressed the lever, but were unaware of how X was being activated. The Exogenous cue group was exposed to a new pairing, B-X, with the lever press triggering a 10s exposure of B before presenting X for 10 additional seconds. Finally, the Observe2 group had 10 seconds of silence before being exposed to 10 seconds of X when the lever was pressed. As expected by causality model, the effect of an action generating the cue X decreased the expectancy of food, as shown by significantly fewer nose pokes than any other group. The rate of nose poking by the Exogenous group strengthens the idea that there is more than retroactive interference at play.

Experiment 2 refined the timing of the stimuli presentations, particularly with the Exogenous cue group. The set up was similar to experiment 1, with the exception of Observe2, which was replaced with an Unpaired group, where they experienced A-X, but A and F were not temporally synced. Another change was that the groups were also yoked temporally, equating the length of the lever press with the subsequent presentation of X or B at test. The unpaired group and the intervene group had the same rate of nose poking, indicating that neither group thought that X had a relationship with food. Additionally, this discounts the idea that the separate temporal context of both cues together was effecting the results from experiment 1.

Experiment 3 examined the flexibility of this determination. In the causal model theory,the lack of nose pokes in response to X at test should only occur when they have the opportunity to X. Subsequent spontaneous presentations of X should still evoke frequent nose pokes, even after an experience with intervention. The associative theory would predict that the amount of nose pokes to X would decrease following subsequent, spontaneous presentations of X since a new, strong association has been formed between Lever Press – X. Experiment 3 also included a simultaneous Y (noise) and F pairing, indicating that Y was directly causing F.

Four groups were used, Common-Intervene, Common-Observe, Direct-Intervene, and Direct-Observe. All groups received the same training during phase 1 and phase 2. Phase 1 consisted of A-X forward pairings and phase 2 consisted of A-F and Y:F pairings. During test day 1, the intervene groups in the Common group lever pressed and activated X, while the Direct group lever pressed and activated Y. The Common group showed less nose pokes than the Direct group, since X was only predictive of F throughA. If the Common group had lever pressed and activated A, it would be reasonable to predict nose pokes at the same rate as the Direct group. On test day 2, the Intervene groups were exposed to spontaneous presentations of X and Y respectively. Contrary to the associative model, nose pokes for the Common group were just as high for the observed groups, indicating that the rats could distinguish between self-generated tones and spontaneous tone presentations.

I thought this was an interesting and well done series of studies to help strengthen the Causal model in rats. The role of intended goal direction action in this model was something that I had not considered, probably because my own conscious experience with how I interact with the world made it too obvious to be seen.  It would be interesting to lesion the part of the brain responsible for self-generated movement in the rat to see if they would nose poke at the same rate of the observe group after lever pressing in a Common cause condition.

This paper also brought up the argument that X should be negatively associated with food, creating fewer nose pokes in general, which obviously isn’t the case. However, I would have thought that the general excitation properties of the A-F condition would be enough to initial create excitement to X.

No Species Can Be Counted Out

Representation of the Numerosities 1-9 by Rhesus Macaques (Macaca mulatta)

By: Elizabeth M. Brannon and Herbert S. Terrace

A thorough experiment conducted by Brannon and Terrace on rhesus monkeys attempts to answer, unequivocally, if monkeys can truly utilize a ‘counting’ method similar to what humans use or if they were able to use other features, like size or surface area, to correctly respond on trials.

Three rhesus monkeys were used and were trained on various counting procedures. Macduff was initially trained on a non-monotonic order, (3-1-4-2) but was unable to reach above-chance performance and was subsequently switched to a monotonic sequence. Macduff and Rosencrantz were trained on an ascending list (1-2-3-4) while Benedict was trained on a descending list (4-3-2-1).

During training trials, all of the stimuli were presented simultaneously on a touch screen and the monkeys were required to touch all of the stimuli in the correct order. If they were able to do so, they received a food reward, but if they answered incorrectly, the trial was terminated and the screen was darkened for 15 seconds. The stimuli were counterbalanced across presentation so size and surface area could not be reliably used.  Additionally, all of the stimulus sets were randomized in terms of their location presentation so muscle memory could not be used. All of the monkeys were able to perform above chance on this large training set and transfer this rule to novel stimuli.

Experiment 2A investigated if the monkeys had formed a nominal understanding of numbers, assigning each to a category that was not meaningfully bound or if they had a true numerical understanding. This was tested by seeing if the monkeys could successfully transfer this rule to exemplars outside of their original training number in single presentations of familiar-novel and novel-novel non-rewarded probe trials. Now the test set contained numbers 1-9, whereas the training set only contained numbers 1-4. Familiar – familiar pairings were rewarded and intermixed with probe trials.

The monkeys that had learned the ascending order, Macduff and Rosencrantz were more successful at transferring their rule learning to F-N and N-N pairs, but the monkey that had learned the descending order, Benedict, performed at a lower level during F-N pairings and at chance for N-N pairings. However, at least for the F-N, Benedict would have had to suppress selecting a previously rewarded stimulus in favor of selecting a novel stimulus, which was not an issue with the ascending order group.

A final study was conducted , 2B, which investigated the similarities between humans and monkeys in terms of accuracy and latency as pairs shift along numerical distance and magnitude. As pairs increased their numerical distance, accuracy increased. However, if pairs maintained the same distance but the magnitude of the numbers increased, latency to respond increased. This pattern is also found in humans.

I thought this was a very thorough experiment in terms of ruling out other possible explanations for correctly selecting stimuli based on number. While they did not explicitly say so, I liked that they ruled out the ‘Clever Hans’ explanation, the original counting animal. I would be interested to know how this relates to symbolic learning of numbers and if being able to count in this way would also correlate with being able to learn other high-level cognitive tasks. As discussed previously, there was an overlap with language experience and reflexive symbol experience in understanding relational rules.

For experiment 2A, I was not sure how the inability of Macduff to learn the 3-1-4-2 pattern did not show that these were something other than nominal categories.

Time as Content in Pavlovian Conditioning

By: Hernan I. Savastano and Ralph R. Miller

This paper reviews the importance of the temporal relationship in Pavlovian conditioning. The Rescola-Wagner learning theory argues that the predictive ability between the CS-US is what drives the learning, where the temporal order can only strengthen or weaken this relationship- it is not enough on it’s own to form associations.  This was in response to the lack of a behavioral response in conditions that presented the US and CS simultaneously or presenting the US first, followed by the CS (backward conditioning). This paper provides vast amounts of evidence to support the idea that contiguity is enough for animals to form an association between two events, even if they are not biologically significant.

The Temporal Coding Hypothesis emphasizes the difference between learning and performance, since previous theories have relied on the behavior as a measure of learning. While the experiments that support the Temporal Coding Hypothesis also rely on behavior, the experiments are refined enough to enable the behavioral expression of learning.

More direct measures of associative strength show stronger conditioning when the US and CS are presented simultaneously, a situation with perfect contingency but little predictive ability. However, assessing the knowledge of temporal order in regards to backward conditioning required the use of second-order conditioning, in which two neutral CS are paired together. In testing, CS2 is forward paired with CS1, then  CS1 is backward paired with US. When the animals were then exposed to CS2, they showed strong CR, indicating that they had learned CS2-US-CS1, despite the fact that CS2 had never directly been forward paired with US. This second-order conditioning is also present in sensory-preconditioning, where CS2-CS1 are paired together, with no biological significance, before CS1 is paired with the US.  Again, at test, the animals showed strong CR to CS2, even though it did not hold predictive value before.

Temporal coding can also explain cue conditioning, where one of the US is made more salient compared to the other. At test, the CR was strongest to the most salient stimulus, presumably because the animal was learning the most from that stimulus. However, when the temporal element of these two CS are investigated, it had a direct effect on the behavior. This is true for blocking studies as well, where a redundant cue does not elicit a strong CR. However, when the initial cue undergoes extinction, the ‘blocked’ cue elicits a strong CR without additional training, indicating that the animal still learned about cue without strong predictive strength.

However, while I agree with the main point of this paper, that the correct experiments were needed to determine the knowledge of temporal coding, I was surprised that the idea of a safety stimulus was never mentioned during discussions of backward pairings that use foot shock  as the US. While it would be obvious if the animal was not performing the same CR that would be shown during a forward-paired condition, there could still be behavioral differences between a backward-paired group (with US – CS) than a group that received shocks with random presentations to the CS, like less freezing or more lever pressing.

Associative Bias of Landmark Learning and Intergration in Vertebrates

By: Kenneth J. Leising and Aaron P. Blaisdell

This paper is a review of how smaller scale navigation has been investigated over the years and the theories that have followed. Initially, a divide was found between habitual and goal-directed navigation. Under certain conditions (over-training, exceedingly simple maze designs) behavior could become so ingrained that the original goal, food, would be overlooked if it were in a new location. However, more complex studies went on to show that given a richer spatial environment and more goals, rats would use these extra-maze cues to navigate. In an ecologically modeled study of a city block, rats would spontaneously explore tunnels outside of their normal routes in obtaining food and water, which was always located in the same place. These lead to a study by Tolman, Ritchie, and Kalish to teach rats to reach a goal location in one of two ways, by using stable extra-maze cues or by always having to turn a certain direction, irrespective of start point, so the goal was always in a different location. The rats using the stable extra-maze cues learned faster, leading them to draw the conclusion that place learning was simpler than response learning. However, future studies have distinguished more of a shift between these two types of learning, with place learning being responsible for the initial learning, and response learning gradually taking over as the response becomes habitual.

Learning about the landmarks provided was thought to occur with a flexible place representation that had enough information to circumvent novel problems in usual routes, forming a rich cognitive map (Tolman 1948). However, he also believed that the response learning was under the control of a strip-map, that was not flexible and would be comprised easily. O’Keefe and Nadel (1978) modified and renamed these concepts into locale and taxon, basing these distinctions on early neuroscience work. This comprehensive map (locale) was using hippocampal cells and was done in a all-or-none manner, meaning that no individual feature was necessary to maintain the integrity of the map. The taxon system would be utilized if the context was deficient in multiple landmarks, with one ‘beacon’ available near the goal and more closely resembles a S-R relationship. The most recent system comes from Burgess (2006) who purports that landmark and beacon learning stem from egocentric representations and the passive learning about the greater context stems from allocentric processes. Interaction between these two systems is ultimately necessary for coding absolute location and translation, but Burgess did not believe that these two systems strongly interacted or competed with each other.

When navigating via landmarks proximity and redundancy effect how these landmarks are used in reaching the goal location. The vector sum theory (a vector being the distance between two objects), found that the altering the the landmark closest to the goal disrupted search behavior more then altering a landmark father away, indicating that the nearby landmark was weighed more heavily.

Vertebrates can also combine goal information across time, as shown in three phases. In the first phase, pigeons were given a compound land mark that had a consistent spatial relationship (AX). In phase two, one of these landmarks (A) had a consistent relationship with the goal location and the new landmark it was paired with did not. Finally, in phase three, when X was presented alone, the pigeons integrated the two relationships to find the goal in relation to X alone (Sawa et. al 2005).  A similar study was done in rats, where the group that had received a configuration of landmarks with a common element were better able to utilize novel configurations than rats without (Chamizo, Rodrigo, and Mackintosh, 2006).

Blocking has also been found in forming spatial maps, where a reliable landmark cue will ‘block’ learning about a new, redundant cue. However, if the original cue becomes unreliable or undergoes a change, unblocking will occur.

Another subtle associative effect is overshadowing, when a compound stimulus that was previously presented together are now presented singularly, only one of the stimuli produces the trained effect. Spetch (1995) trained pigeons to reach a goal with near and moderate (A) landmark or a moderate and far (B) landmark, with the moderate landmark being the same distance from the goal in both groups. When these groups were given the moderate landmark only, B group had a stronger performance than the A group.

Despite knowing a large amount on how different landmarks are weighted and what associative processes are involved, a best model has not been found or rigorously tested for. Elemental models, like the Vector-sum, have been supported, but so have some the associated and computational models. Additionally, even when models explain some elements of navigation, like the weighing of landmarks, it still cannot explain if there is a performance, attention or acquisition deficit.

 

Declarative and Episodic-Like Memory in Animals

By:  Nicola S. Clayton, D. P. Griffiths, and Anthony Dickinson

Episodic memory is unique in terms of its contents and its realm within the field of memory and cognition. It seems to be one of the few remaining aspects of cognition that is uniquely human, thanks in large part to the definition of Tulving and  Markowitsch where the conscious awareness of retrieving the memory is the key feature of a ‘true’ episodic memory. However, due to the linguistic component of this definition, this naturally falls outside of the domain of animal cognition since one of the oldest problems is having to ask questions and receive answers without using words.

However, there are some examples of episodic-like memory in a wide variety of species that rely on a simpler definition of what-where-when memory, with each component being a key part of the memory, the most notorious of which is the scrub jay.

The scrub jay is a bird species that caches food in hiding places in order to eat over the winter. This survival technique implies a great memory source in order to utilize these caches effectively. When put through a variety of tests, the memory ability of the scrub jay maintained its strength.

To determine if the scrub jays could utilize all three components of this memory type, Clayton and Dickinson created a series of experiments that took advantage of this natural behavior. This test group of birds was divided; one group learned that wax worms degrade over time (DG) and the other group did not learn that wax worms degraded since the experimenters replaced the worms 124 hours later (RG).

All of the birds were allowed to cache wax worms and peanuts (which do not degrade, but are not as preferable as wax worms) then retrieve their caches 4 to 124 hours later. After 4 hours, the DG birds retrieved the still fresh wax worms first, but after 124 hours the peanuts were retrieved. However, RG birds continued to retrieve wax worms after 124 hours.  This implies that the birds are aware of when they make the caches, not that the memory of the wax worm cache is forgotten faster or that they are predisposed to certain caches at certain times.

A modified paradigm was used to create a more spatially distinct memory, forcing the birds to cache peanuts on one side of a distinct tray and wax worms on another in a distinct part of the room, at different time points. Again, DG birds were able to retrieve wax worms at the appropriate time point from the correct part of the tray, while the RG birds were not.

The birds were also tested on their ability to remember which cache sites had already been recovered.  The caching sites and first controlled recovery was counterbalanced across birds, so when they were allowed to recover their cache a second time, they would have to remember they had originally cached the food, what caches they had already recovered, and which sites still contained their preferred food.  This paradigm was further controlled in the next study in which food preference was artificially degraded by pre-feeding the birds before being allowed to recover their caches. In both studies, birds performed well above chance in their retrieval selection.

The eternal problem of comparing animal and human cognition is the element consciousness. Without being able to state their own perceptions and states explicitly, there will always be a divide. However, being able to amp which brain regions are active during this what-where-when retrieval compared to a human’s could possible strengthen the argument that these two processes are more similar than different. Further unraveling of the molecular mechanisms will show key elements of what makes each species unique.

A brief comment on the Wright, Santiago, Sands, Kendrick, and Cook paper ‘Memory Processing of Serial Lists by Pigeons, Monkeys, and People’ while the results were still visible and significant, I was surprised that the n for pigeons and humans was so small.

Language-Naive Chimpanzees (Pan troglodytes) Judge Relations Between Relations in a Conceptual Matching-to-Sample Task

By: Roger K.R. Thompson, David L. Oden, and Sarah T. Boysen

It was initially purported that only chimpanzees trained in language were able to understand conceptual reasoning problems, where matching depending on an abstract concept, not physical appearance. With very intensive differentially rewarded training, language-naive chimps could learn an analogous proportions task that did not generalize to other conceptual tasks.

The purpose of this study was to determine if experience with language was truly a requirement for being able to judge relations between relations or if it could be accounted for via associative processes, experience with token learning, or general sophisticated test experiences.

The test was a conceptual matching to sample, so if the sample consisted of a set of identical objects (AA, or a pair of identical cups) then the correct choice would be a set of items identical to each other, but different than the sample (BB or a pair of identical boxes instead of two dissimilar objects (CD or a screwdriver and a flower). In the conceptual match to sample task, object valance was counterbalanced so objects previously associated with samples were put into conflict. All of the conceptual matching was nondifferentially rewarded. Ability to match objects based on physical similarity in real-world objects versus objects on a computer screen was also tested and differentially rewarded.

4 out of 5 chimpanzees were able to spontaneously correctly complete the matching task. 1 of the 4 had prior language experience, but the other 3 only had experience with token training in numerical concepts. The 1 monkey that was unable to correctly match the sets on a conceptual level or a physical level had no prior testing experience, including screen training.

These results indicate that language training is not a prerequisite for understanding abstract concepts, but token training may be the key in understanding relational concepts. Another previously purported solution was test sophistication, but this argument is weakened by the original study, where chimpanzees that were exposed to a number of tests (but none involving tokens) could not conceptually match items. Another theory was ‘dogged’ training, where over numerous differentially reinforced trials language-naive chimpanzees could eventually match proportionally similar objects. However, in this study chimpanzees were able to spontaneously match sets based on an abstract concept within the first session.

These results do indicate that previous experience with abstract token training, where the token representation becomes reflexive (heart token means 3, 3 means heart token)  enables chimpanzees to better identify abstract relations. However, performance of the naive chimp implies that this is not an innate way of thinking.  By not constricting performance on abstract tasks on the basis of language, many more animals are able to be studied under these conditions. Investigating token training and its similarities to language training would be helpful in determining the underlying mechanism that enables abstract concepts.

 

 

Amodal Completion and Illusory Perception in Birds and Primates

By: Kazuo Fujita, Noriyuki Nakamura, Ayumi Sakai, Sota Watanabe, and Tomokazu Ushitani

This chapter was a short review on research that has compared the nuances of the nonhuman primate and avian visual system compared to humans. Visual illusions and completion of ambiguous images in particular were examined.

When completing a vertical bar that is partially occluded by a horizontal, babies, adults, and nonhuman primates assume that it is one, long vertical bar, whereas pigeons perceive it as two separate shorter bars. This difference also effects how these species interpret length of bar when it is directly touching another object, with nonhuman primates indicating the bar is longer than it actually is, when pigeons do not show this bias.  While the pigeons’ perceptual system do not automatically assume completion, even with ecologically relevant stimuli, they can be trained to complete.

With visual illusions, pigeons and nonhuman primates also show consistent perceptual differences. In the Ponzo illusion, all species were effected, but even among closely related primate species, manipulations of the illusions created stronger effects for some species and weaker ones for others, indicating that the mechanism is not shared.

Pigeons are disproportionately biased to assimilation illusions, as shown in the strength of the Muller- Lyer illusion and interpretation of the Ebbinghaus-Titchener Circles illusion as an assimilation illusion. When the circle illusion was limited to a more local perspective (outer edges of inducing circles erased), this effect was more understandable from a human perspective.

Even though all of the species focused on in these studies are highly visual creatures, their particular visual systems lead to different visual perceptions and assumptions. This chapter points out an interesting caveat in comparing cognition specifically, when each species is working with different inputs. In addition to ensuring that experiments are only asking the question of interest, researchers need to ensure that the experiment is suited for their particular receptors.