**3. What/where/when memory in western scrub jays**

Clayton and Dickinson (1998) have been largely responsible for introducing and developing the concept of episodic memory in non-humans. They have demonstrated that Western scrub jays form integrated memories of what, where and when information in the context of caching and recovering food. Furthermore, they suggest that the types of caching behaviour shown by the scrub jays requires them to mentally travel forward and backward in time, which is a component of human episodic memory (Clayton et al., 2003a). However, because Clayton, Dickinson and their colleagues have not been able to demonstrate autonoetic consciousness (i.e., a sense of self) in scrub jays, they have stopped short of declaring that scrub jays have human-equivalent episodic memory. Instead, they have opted to conclude that scrub jays possess "episodic-like memory." This type of memory shares some characteristics with the definition of human episodic memory (Tulving, 1983), but avoids the currently impossible task of demonstrating consciousness without the use of verbal language (Clayton et al., 2003b).

26 Advances in Object Recognition Systems

of an event, the ability to recognize subjective time, and autonoetic consciousness (knowledge of self; Tulving, 1983, 2002). The main distinction between episodic memory and other forms of recall involves the recreation of a personally experienced event. Simple retrieval of discrete facts (e.g., Marconi received a wireless transmission at Signal Hill in 1901), does not require the self-consciousness nor the ability to mentally travel forward and backward in time that are indicative of episodic memory (e.g., I was on Signal Hill yesterday and read a sign about Marconi). Despite the acceptance of episodic memory in humans, its

In the absence of a measure of consciousness in non-human animals, it has not been possible to demonstrate episodic memory that is equivalent to humans. However, by studying food caching (Clayton & Dickinson, 1998), food finding (Babb & Crystal, 2006), fear conditioning (O'Brien & Sutherland, 2007), and object exploration (Eacott & Norman, 2004), researchers claim to have demonstrated a form of episodic memory in scrub jays (Clayton & Dickinson, 1998), pigeons (Zentall et al., 2001), mice (Dere et al., 2005), rats (Eacott & Norman, 2004; O'Brien & Sutherland, 2007), gorillas (Schwartz & Evans, 2001), rhesus monkeys (Hoffman

The interpretation of such studies is often controversial because there is no consensus regarding a definition of non-human episodic memory (Hampton & Schwartz, 2004). Schwartz, Hoffman and Evans (2005) outlined five operational definitions of non-human episodic memory including: (1) the demonstration of what/where/when memory (Clayton & Dickinson, 1998; Babb & Crystal, 2006), (2) the demonstration of what/where/which memory (Eacott & Norman, 2004), (3) the demonstration of spontaneous recall (Menzel, 1999), (4) the ability to recall an event when not expecting a test (Zentall et al., 2001), and (5) the ability to report on past events over a long term (Schwartz & Evans, 2001). Unfortunately, these definitions tend to be species-specific. For example, definitions of episodic memory based on research with food-caching birds (Clayton & Dickinson, 1998) often do not fare well when applied to non-caching species (Bird et al., 2003; Hampton et al., 2005). Consequently, alternative methods and definitions have been developed for rodents,

Clayton and Dickinson (1998) have been largely responsible for introducing and developing the concept of episodic memory in non-humans. They have demonstrated that Western scrub jays form integrated memories of what, where and when information in the context of caching and recovering food. Furthermore, they suggest that the types of caching behaviour shown by the scrub jays requires them to mentally travel forward and backward in time, which is a component of human episodic memory (Clayton et al., 2003a). However, because Clayton, Dickinson and their colleagues have not been able to demonstrate autonoetic consciousness (i.e., a sense of self) in scrub jays, they have stopped short of declaring that scrub jays have human-equivalent episodic memory. Instead, they have opted to conclude that scrub jays possess "episodic-like memory." This type of memory shares some characteristics with the definition of human episodic memory (Tulving, 1983), but avoids the currently impossible task of demonstrating consciousness without the use of verbal

et al., 2009), and chimpanzees/bonobos (Menzel, 1999; Martin-Ordas et al., 2010).

presence in non-human animals is controversial.

primates, and non-caching birds.

language (Clayton et al., 2003b).

**3. What/where/when memory in western scrub jays** 

Clayton and Dickinson (1998) took advantage of the scrub jays' natural food-storing behaviours and allowed each bird to cache both perishable, but preferred, worms and nonperishable peanuts in opposite sides of an ice-cube tray filled with sand. Initially, the scrub jays demonstrated the ability to recall the location ("where") in which they cached each type of food ("what"), and consequently retrieved the preferred food, worms, before peanuts. In subsequent trials, the researchers replaced freshly cached worms with decayed worms if worms were cached first (124 h before retrieval) and peanuts cached second (4 h before retrieval). In contrast, fresh worms were left in their cached locations if peanuts were cached first (124 h before retrieval) and worms cached second (4 h before retrieval). Remarkably, the scrub jays quickly learned to retrieve peanuts if worms were cached first (since decayed worms are unpalatable) and to retrieve worms if peanuts were cached first. A similar result, although less compelling, was found when jays were taught that worms were removed (pilfered) if they were cached 124 h before retrieval.

In numerous subsequent studies, Clayton and Dickinson further developed their case for episodic-like memory in scrub jays. Specifically, through allowing jays to cache peanuts and dog kibble and then recover these items on successive trials, they demonstrated that scrub jays update their memories about which cache sites contain food (Clayton & Dickinson, 1999). Furthermore, by making one food less preferable than another through pre-feeding, they found that jays successfully identified food caches that were both non-recovered and contained preferable food. Clayton and Dickinson (1999) argue that this ability indicates that scrub jays form episodic-like memories that integrate the type of food in a cache, the location of that cache, the last activity at that cache (recovery or caching) and how long ago food was stored. Clayton et al., (2005) have also shown that scrub jays use novel information about the decay of a food source to reverse their strategies for recovery, since jays cache more non-perishable food items if their caches are consistently degraded on recovery. Emery and Clayton (2001) found that scrub jays who have previously raided the food cache of a conspecific will re-cache food if they are observed during their own caching process. Recently, Cheke and Clayton (2011) examined caching in the Eurasian jay and demonstrated that birds distinguish between their current food preference (created by pre-feeding a specific food) and their future needs. This was evidenced by the birds overcoming motivation to cache currently desired food and instead caching currently non-preferred foods according to their future value. Taken together, these findings provide preliminary evidence that caching scrub and Eurasian jays make decisions based on past episodes and anticipated future needs. Because these results suggest that episodic-like memory includes aspects of the mental time travel involved in human episodic memory, further study in this area, including research on non-caching species, such as ant-following birds, is suggested (Clayton et al., 2003c; Logan et al., 2011).

## **4. What/where/when memory in other species**

Many researchers have used the basic what/where/when criteria proposed by Clayton and Dickinson (1998) in their attempts to demonstrate episodic-like memory in species such as pigeons (Skov-Raquette et al., 2006), primates (Hoffman et al., 2009; Martin-Ordas et al., 2010), mice (Dere et al., 2005), and rats (Babb & Crystal, 2006; Fortin et al., 2002; Kart-Teke et al., 2006; O'Brien & Sullivan, 2007). The majority of studies have been conducted using mice and rats, which has led to the development of several different testing paradigms.

Spontaneous Object Recognition in Animals: A Test of Episodic Memory 29

that rats quickly learn to restrict their searches to the locations that provide food indicating that they have learned the bipartite what/where code (Thorpe, et al., 2003). It is also known that rats can learn when in the day that they will receive food – or the bipartite what/when code (Means et al., 2000; Thorpe et al., 2003). However, it is only under certain conditions that rats combine these three components into a tripartite what/where/when code and successfully solve the task. For example, in situations in which there is a high cost of making a mistake, either in effort or in time, rats are more likely to solve the task (Widman et al., 2000). Given these findings, animals may be able to learn temporal information, but it may

In an attempt to avoid some of the confounds and problems involved in demonstrating "when" memory, Eacott and Norman (2004) used *context* to replace time as the "when" component of episodic-like memory, which broadens the definition of episodic-like memory to include integration of the "what, where, and *which*" details of an event. They argue that the function of the "when" aspect of episodic memory is simply to mark an event as being unique. Therefore, requiring animals to remember the discrete time at which an event occurred (e.g., 1 hour ago or 24 hours ago) is the same as having animals discriminate the context in which an event occurred (e.g., white-walled room vs. black-walled room; Eacott & Gaffan, 2005; Eacott & Norman 2004;). Either chronological time or context can serve as the reference point that identifies a specific event and allows it to be recalled. This idea is further supported by the fact that time does not appear to be an essential part of human episodic memory. Humans tend to use background cues that are present during an event, rather than

The paradigm used most often to assess what/where/which memory is the novel object recognition task. This clever but simple task takes advantage of a predisposition in many species to explore novel objects over familiar ones. Ennaceur and Delacour (1988) first reported the object recognition task, in which rats were exposed to objects during an acquisition trial and then tested on their ability to discriminate between familiar and novel objects, as a test of working memory. The object recognition test has been used to show that rats are sensitive to the location of objects (Dix & Aggleton, 1999; Ennaceur et al., 1997; Poucet, 1989), to the topological relationship between objects (Dix & Aggleton, 1999; Goodrich-Hunsaker et al., 2008; Harley et al., 2001; Lemon et al., 2009), to changes in the distance between objects (Goodrich-Hunsaker et al., 2008), to the context in which objects have been experienced (Dix & Aggleton, 1999; Eacott & Norman, 2004), and to changes in

In addition to the innovative what/where/which definition, Eacott and Norman's (2004) unique method of testing episodic-like memory meets the requirements of spontaneous recall (Menzel, 1999) and recall during an unexpected test (Zentall et al., 2001). Eacott & Norman (2004) found that rats can integrate memories of a specific object (what), its spatial location (where) and the context in which it occurs (which) to discriminate the more novel of two object/location/context combinations. Rats explored the locations (left or right) of each

not reflect the natural way events are encoded.

**6. Novel object recognition task** 

object compounds (Norman & Eacott, 2004).

**5. What/where/which episodic-like memory** 

time, to distinguish it from other similar events (Friedman, 1993).

Babb and Crystal (2006) developed a radial maze task that required rats to remember the type of food contained in different maze arms at different times. They showed that rats were able to integrate what/where/when memories to obtain preferred foods, and that rats changed their preferences if these preferred foods were devalued. Fortin et al. (2002) developed a task in which rats were required to remember a series of odour cues to obtain food from sand-filled cups. The rats were able to remember the odour and whether it occurred before or after another odour in the sequence. However, Clayton et al., (2003a) argued that rats may have solved this task using internal interval timing, and that this task does not demonstrate integrated memory for "where." O'Brien and Sutherland (2007) took advantage of the observation that rats need exposure to a context to form context-shock associations (Faneslow, 1990) and that the associations formed can be based solely on the memory of the context (Rudy et al., 2002). They (O'Brien & Sutherland, 2007) exposed rats to two distinctive boxes, one in the morning and the other in the evening. After the exposure, rats were exposed to a third box that was an amalgam of the morning and evening box. They were shocked in this box in either the morning or the evening session. Tests of freezing at an intermediate time interval in either the morning or the evening box demonstrated freezing to the box congruent with the time of day the shock had been received. This finding indicated that the rats had formed a time-place memory and that this memory had been updated at the time the shock had been administered. A recent study with chimpanzees, bonobos and orangutans adapted the methods of Clayton and Dickinson (1998) and showed that apes integrate what/where/when memories to choose between frozen juice (the preferred food after a 5 min rest interval, but not after a 1h rest interval because it melts and becomes unavailable) and a grape (the preferred food after a 1h rest interval because the juice is unavailable) (Martin-Ordas et al., 2010).

Although not exhaustive, the above list illustrates the main testing strategies that have been used to demonstrate what/where/when memory in non-caching species. The absence of caching behaviour in many species is a serious hindrance to replicating the results found in scrub jays (Bird et al., 2003; Hampton et al., 2005). Although numerous clever methods have been developed to test the what/where/when criteria, many of these cannot avoid alternate, more parsimonious explanations for results. With the possible exception of O'Brian and Sutherland (2007), this is particularly true for the "when" component of episodic-like memory. Even studies that have gone so far as to show that memories are flexible (i.e., a rat's change in food preference shown by Babb & Crystal 2006) are confounded by the possibility of relative memory strengths and internal time intervals experienced by subjects.

The problematic nature of the "when" aspect of memory is also demonstrated by distinct but related research in daily Time-Place Learning. In daily Time-Place learning tasks, animals are trained that a food reward is available in one location in morning sessions and in another location in afternoon sessions (Thorpe & Wilkie, 2006). This task is different from episodic tasks in that the subjects require repeated training prior to restricting their searches to the appropriate locations at the correct times of day. To solve this task, an animal must learn to associate event/place/time or what/where/when information in a single code. Paralleling the results in the episodic-like literature, pigeons learn this task relatively easily (Saksida & Wilkie, 1994); however, both fish (e.g., Barreto et al., 2006) and rats (e.g., Thorpe et al., 2003) have much more difficulty acquiring the task. Research has shown, however, 28 Advances in Object Recognition Systems

Babb and Crystal (2006) developed a radial maze task that required rats to remember the type of food contained in different maze arms at different times. They showed that rats were able to integrate what/where/when memories to obtain preferred foods, and that rats changed their preferences if these preferred foods were devalued. Fortin et al. (2002) developed a task in which rats were required to remember a series of odour cues to obtain food from sand-filled cups. The rats were able to remember the odour and whether it occurred before or after another odour in the sequence. However, Clayton et al., (2003a) argued that rats may have solved this task using internal interval timing, and that this task does not demonstrate integrated memory for "where." O'Brien and Sutherland (2007) took advantage of the observation that rats need exposure to a context to form context-shock associations (Faneslow, 1990) and that the associations formed can be based solely on the memory of the context (Rudy et al., 2002). They (O'Brien & Sutherland, 2007) exposed rats to two distinctive boxes, one in the morning and the other in the evening. After the exposure, rats were exposed to a third box that was an amalgam of the morning and evening box. They were shocked in this box in either the morning or the evening session. Tests of freezing at an intermediate time interval in either the morning or the evening box demonstrated freezing to the box congruent with the time of day the shock had been received. This finding indicated that the rats had formed a time-place memory and that this memory had been updated at the time the shock had been administered. A recent study with chimpanzees, bonobos and orangutans adapted the methods of Clayton and Dickinson (1998) and showed that apes integrate what/where/when memories to choose between frozen juice (the preferred food after a 5 min rest interval, but not after a 1h rest interval because it melts and becomes unavailable) and a grape (the preferred food after a 1h rest interval because the

Although not exhaustive, the above list illustrates the main testing strategies that have been used to demonstrate what/where/when memory in non-caching species. The absence of caching behaviour in many species is a serious hindrance to replicating the results found in scrub jays (Bird et al., 2003; Hampton et al., 2005). Although numerous clever methods have been developed to test the what/where/when criteria, many of these cannot avoid alternate, more parsimonious explanations for results. With the possible exception of O'Brian and Sutherland (2007), this is particularly true for the "when" component of episodic-like memory. Even studies that have gone so far as to show that memories are flexible (i.e., a rat's change in food preference shown by Babb & Crystal 2006) are confounded by the possibility of relative memory strengths and internal time intervals

The problematic nature of the "when" aspect of memory is also demonstrated by distinct but related research in daily Time-Place Learning. In daily Time-Place learning tasks, animals are trained that a food reward is available in one location in morning sessions and in another location in afternoon sessions (Thorpe & Wilkie, 2006). This task is different from episodic tasks in that the subjects require repeated training prior to restricting their searches to the appropriate locations at the correct times of day. To solve this task, an animal must learn to associate event/place/time or what/where/when information in a single code. Paralleling the results in the episodic-like literature, pigeons learn this task relatively easily (Saksida & Wilkie, 1994); however, both fish (e.g., Barreto et al., 2006) and rats (e.g., Thorpe et al., 2003) have much more difficulty acquiring the task. Research has shown, however,

juice is unavailable) (Martin-Ordas et al., 2010).

experienced by subjects.

that rats quickly learn to restrict their searches to the locations that provide food indicating that they have learned the bipartite what/where code (Thorpe, et al., 2003). It is also known that rats can learn when in the day that they will receive food – or the bipartite what/when code (Means et al., 2000; Thorpe et al., 2003). However, it is only under certain conditions that rats combine these three components into a tripartite what/where/when code and successfully solve the task. For example, in situations in which there is a high cost of making a mistake, either in effort or in time, rats are more likely to solve the task (Widman et al., 2000). Given these findings, animals may be able to learn temporal information, but it may not reflect the natural way events are encoded.
