Elephant cognition in primate perspective

Elephant cognition
65
2009
Volume 4, pp 65-79
Elephant cognition in primate perspective
Richard W. Byrne and Lucy A. Bates
School of Psychology, University of St Andrews
Cynthia J. Moss
Amboseli Trust for Elephants
On many of the staple measures of comparative psychology, elephants show no obvious differences from other mammals,
such as primates: discrimination learning, memory, spontaneous tool use, etc. However, a range of more naturalistic mea-
sures have recently suggested that elephant cognition may be rather different. Wild elephants sub-categorize humans into
groups, independently making this classification on the basis of scent or colour. In number discrimination, elephants show
no effects of absolute magnitude or relative size disparity in making number judgements. In the social realm, elephants
show empathy into the problems faced by others, and give hints of special abilities in cooperation, vocal imitation and per-
haps teaching. Field data suggest that the elephant’s vaunted reputation for memory may have a factual basis, in two ways.
Elephants’ ability to remember large-scale space over long periods suggests good cognitive mapping skills. Elephants’ skill
in keeping track of the current locations of many family members implies that working memory may be unusually devel-
oped, consistent with the laboratory finding that their quantity judgements do not show the usual magnitude effects.
Keywords: Loxodonta, Elephas, cognitive maps, social knowledge, social memory, tool-use, classification learning, quantity discrimina-
tion, empathy, mirror self-recognition
Although the cognitive revolution was slow in coming to an- Relative to their body size, the brain size of elephants is
imal behaviour, in the last twenty years our understanding of large, although not in the extreme range of humans or dol-
animal cognition has advanced considerably. In particular, phins. The encephalization quotient varies between 1.3 and
the highly social non-human primates have been the focus 2.3, depending on species (Cutler, 1979; Jerison, 1973),
of intense interest, and research in comparative cognition comparable indices to those of apes (Eisenberg, 1981). En-
has recently expanded to other large-brained and often so- cephalization gives a useful measure of investment in brain
cial taxa, such as corvids (Bugnyar, 2002; Bugnyar & Hein- tissue relative to metabolic energy available in total, and
rich, 2005; Clayton & Dickinson, 1998; Emery & Clayton, thus indicates the cost to the animal of so large a brain (By-
2001, 2004), cetaceans (Connor, 1999; Herman, 1986; Reiss rne, 1996). It does not, however, give the best indication of
& Marino, 2001; Rendell & Whitehead, 2001), and social ‘brain-power’: as with any computational system, to assess
carnivores (Hare, Brown, Williamson, & Tomasello, 2003; power it is the actual number of processing units available
Holekamp, 2006; Manser, Seyfarth, & Cheney, 2002; Mik- for use that is important (Byrne, 1996). Elephants have the
losi, Topal, & Csanyi, 2004). Here we review what is known largest absolute brain size among land animals: up to 5.5
about the cognition of another large-brained and highly so- kg in Asian elephants and up to 6.5 kg in African savannah
cial group – elephants. Research conducted specifically into elephants (Cozzi, Spagnoli, & Bruno, 2001; Shoshani, Kup-
elephant cognitive abilities is still a fledgling enterprise, but sky, & Marchant, 2006). Although neurons are less densely
we consider what we do and do not know, and compare the packed in elephant brains than in primates’, elephant brains
observable pattern with what we know about the best-stud- nevertheless contain as many cortical neurons as do human
ied mammalian group, the primates.
brains (Roth & Dicke, 2005). Moreover, the pyramidal neu-
Send correspondence to Richard W. Byrne, Centre for Social rons are larger than in humans and most other species, with
Learning & Cognitive Evolution and Scottish Primate Research a large dendritic tree giving the potential for many more
Group, School of Psychology, University of St Andrews, St An-
connections, and perhaps hinting at superior learning and
drews, Fife KY16 9JU, United Kingdom (rwb@st-andrews.ac.uk).
memory skills (Cozzi, et al., 2001).
ISSN: 1911-#### doi: 10.####/ccbr.2009.300## © Richard W. Byrne 2009

---

Elephant cognition
66
There is a considerable body of theory and supportive data people working with Asian elephants that are held in semi-
to suggest that living in an extensive social network often captive conditions for use in the logging industry. Both
correlates with, and likely promotes, cognitive sophistica- observational studies and anecdotal reports have provided
tion (Byrne & Bates, 2007). To judge by the well-studied many examples of behaviors that appear to reflect advanced
African elephant Loxodonta africana, elephant society may cognitive processing, but often without appropriate control
be one of the most elaborate discovered among mammals or observations. In this paper, we attempt to bring these differ-
birds. Elephant ranging is fission-fusion in nature: female ent sources of information together to provide an evaluation
matrilineal relatives and dependent offspring form family of what can so far be judged of elephant cognitive skills,
units that usually travel, forage and socialize together, but and what needs to be examined in the future. Given the cur-
these family units can split up into smaller groups with ir- rent paucity of studies designed specifically to test elephant
regular composition and can also join with members of other cognition, all conclusions that we draw must be treated as
families to form larger groups (Moss, 1988). It is therefore tentative and requiring further verification.
possible that each individual elephant knows and differenti- In order to organize this somewhat patchy information, we
ates among several hundred other individual elephants, far turn to current knowledge of what must be the best-studied
in excess of anything found in non-human primates. It has order of mammals: the primates. The most comprehensive
also often been suggested that superior primate intelligence review of primate cognition remains that of Tomasello and
derives from the technical sophistication that is allowed by Call (1997). Although the literature it evaluates is now a de-
prehensile hands, and the flexible motor control of the hands cade out of date, we borrow their organisation as a structure
and fingers found especially in apes (Deacon, 1997; Napier, on which to hang our review of elephant cognition, allowing
1962). On this theory, also, a case may be made that the el- for straightforward comparison. (Note that this categoriza-
ephant might possess special abilities. The elephant’s trunk tion may not truly reflect the organisation of the underly-
is prehensile, and under exquisite brain control (Shoshani, ing cognitive apparatus.) Also like those authors, we take a
et al., 2006), such that it can pick up and put down an egg cognitive approach in theorizing, throughout. In many of the
without breaking it.
cases we review, post hoc associative accounts could also
The large brains, societal complexity and advanced pre- be constructed; but we consider that formalism an unhelpful
hension of elephants have evidently evolved in parallel with one for comparing animal abilities, particularly when flex-
those of primates. Elephants belong to the Afrotheria (Hedg- ible and sophisticated behaviour is deployed in real-world
es, 2001; Murphy, et al., 2001), an ancient placental group environments (see Byrne & Bates, 2006).
that evolved between 105 and 40 Mya when the continent
of Africa was isolated from other land masses. Among their
Elephant perceptual systems
closest relatives, elephants number golden moles, tenrecs,
dugongs and elephant-shrews—Afrotheria is not a lineage Perception and cognition are intimately linked, and since
in which large brains or social complexity are conspicuous. elephants are so distantly related to primates (and indeed to
This means that if the cognitive abilities of elephants prove any of the species traditionally studied in the psychologist’s
to resemble in any way those of humans, we can be sure that laboratory), we first provide a brief overview of what is so-
the resemblance results from convergent evolution rather far known of their capacities.
than primitive retention. This, in turn, would enable more Elephants are dichromats, with the same colour vision
confident identification of the ecological contexts that pro- pigments as human colour-blind deutaneropes (Yokayama,
mote these cognitive skills. Conversely, if elephant abilities Takenaka, Agnew, & Shoshani, 2005). With a greater con-
prove wholly strange to those we are familiar with from the centration of rods than cones, their eyesight is thought to be
study of humans and other primates, understanding elephant good in dull light, but considerably reduced in bright light,
cognition will broaden our knowledge of cognitive process- reaching a maximum range of 46 m (Sikes, 1971). Certain-
es in general. It would seem that elephant cognition is an ly, their eyesight is good enough to detect the sometimes
obvious topic for the highest priority.
subtle postural displays that appear to be very important to
Yet, to our knowledge, only 21 papers that describe spe- elephants, such as ear-flapping, ear folding and trunk curl-
cific attempts to assess elephant cognitive skills have to date ing (Kahl & Armstrong, 2000). Elephants are also extremely
been published in peer-reviewed scientific journals (see Ap- tactile animals, frequently touching one another using their
pendix 1 for a list). Elephant vocal and chemical communi- trunk, ears, tusks, feet, tail, and whole body. Tactile interac-
cation systems have been more extensively studied, but the tions between elephants can occur in aggressive, defensive,
cognitive implications of this work are usually not drawn affiliative, sexual, playful and exploratory contexts. How-
out. Most of what is known about elephant behaviour comes ever, it is the olfactory and auditory senses that are generally
either from structured observational studies of free-ranging considered to be the most significant to elephants, for obvi-
African savannah populations, or from the anecdotes of ous physical reasons.

---

Elephant cognition
67
Figure 1. At Amboseli, researchers can recognize over 1400 elephant individuals. Often the ears give the first clues to an
individual’s identity: overall shape, the pattern of veins and damage to the edges all differ. Photograph by the authors.
Elephants are capable of producing different sounds rang- air and at 248-264 m/sec through the ground (O’Connell-
ing from 5 Hz to over 9,000 Hz (Poole & Granli, 2003) Rodwell, Arnason, & Hart, 2000). Recent evidence suggests
resulting in a broad range of vocalizations from very low that elephants are able to detect these seismic vibrations, or
frequency rumbles to higher frequency trumpets and barks Rayleigh waves, through two possible means: bone con-
(Berg, 1983). However, elephants are specialists in the pro- duction and the use of massive ossicles of their middle ears
duction of low frequency sound, with the most often pro- (Reuter, Nummela, & Hemila, 1998), or possibly by mecha-
duced vocalizations being low frequency rumbles; the low- noreceptors such as Pacinian corpuscles in the toes or feet
est, infrasonic, components of these rumbles can be between that are sensitive to vibrations (O’Connell, Hart, & Arnason,
one and two octaves below the lower limit of human hearing 1998).
(Poole & Granli, 2003). Although the lower range limit of Olfaction and chemical communication is also very impor-
elephants’ hearing has not been established, they are known tant to elephants. Within the nasal cavity are seven turbinates
to be more sensitive to low frequency sounds, and less so the (dogs only have five), scrolls of bones with sensitive tissues
high frequency sounds, than any other mammal so far tested specialized for olfaction and hormone detection, containing
(Heffner & Heffner, 1982).
millions of olfactory receptor cells (Shoshani, 1997). An el-
Elephant groups are frequently highly dispersed; with ephant may be able to gain enough information by sniffing,
groups often spread over 100m in diameter and, due to the but if not it will collect the substance on the tip of trunk and
fission-fusion social organisation, kin sub-groups can often pass it to the Jacobson’s or vomeronasal organ on the roof
be separated by several kilometers. Elephants use their low of the mouth for further analysis, behaviour known as the
frequency rumbles such as contact calls to communicate flehmen response.
over these long distances (Payne, Langbauer, & Thomas, Sources of odours used in chemical communication be-
1986); these low frequency sounds are known to travel over tween elephants include urine, faeces, saliva and secretions
10km in optimal conditions (Garstang, Larom, Raspet, & from the temporal gland, a large multi-lobed sac with an
Lindeque, 1995).
orifice mid-way between the ear and eye (Langbauer, 2000;
When an elephant rumbles, a replica of the airborne sound Rasmussen & Krishnamurthy, 2000). Elephants can fre-
is also transmitted through the ground. Elephant sounds have quently be seen to raise their trunks up in the air to smell, or
been measured as travelling at 309 m per second through use the tips of their trunks to explore the ground, for urine

---

Elephant cognition
68
spots and faecal matter, as well as the genitals, temporal tree which held unripe fruit on their last visit, their decision
glands, or mouths of other elephants (Rasmussen & Krish- whether to check the tree depends on the weather since their
namurthy, 2000).
last visit: nor is this an effect of more enthusiastic ranging on
warm sunny days, as the effect persisted when the conditions
Knowledge of physical environment
on the day of travel were statistically controlled.
Although the logistics of studying elephants in the field
Space and objects
are entirely different to those in studying primates, there are
nevertheless some data on which to make comparison with
Extensive research has been carried out on how primates the primate picture.
understand objects in their immediate environment, includ- African savannah elephants are known to move over very
ing object permanence and the ability to track hidden dis- large distances in their search for food and water. Leggett
placements; however, these topics have not been studied in (2006) used GPS collars to track the movements of elephants
elephants, so no useful comparison can be made. When it living in the Namib desert. He recorded one group living
comes to the understanding of large-scale space, involving in the Namib desert travelling over 600 km in five months,
memory for and navigation within areas that cannot be ap- and Viljoen (1989) showed that elephants in the same region
prehended from a single viewpoint, there are data for both visited water holes, some of which were over 60 km apart,
species.
approximately every four days. Even more impressively, el-
It has long been suspected that primates possess powerful ephants inhabiting the deserts of northern and southern Af-
and efficient ‘cognitive mapping’ skills (Mackinnon, 1978; rica have been described travelling hundreds of kilometres
Milton, 1981), but only in the last few years have strong tests to arrive at remote water sources shortly after the onset of a
been made (Janson & Byrne, 2007). Experimental studies period of rainfall (Blake, Bouche, Rasmussen, Orlando, &
have shown that capuchin monkeys are able to head directly Douglas-Hamilton, 2003; Viljoen, 1989), sometimes along
for distant locations, and choose the most valuable to visit routes that researchers believe have not been used for many
first (Janson, 2007); however, they do not compute routes years. These remarkable feats suggest exceptional cognitive
that show efficiency over more than one target (i.e. solve mapping skills, reliant on the long-term memories of older
‘the travelling salesman problem’). Observations of ba- individuals who travelled that path sometimes decades ear-
boons have shown some ability in this direction, however: lier. Indeed, a recent study has confirmed that family groups
baboons take direct routes to distant, high quality resources with older matriarchs are better able to survive periods of
that are likely to be exploited by other groups, returning later drought (Foley, Pettorelli, & Foley, 2008). Families with
to consume other, more reliable resources by-passed earlier older matriarchs range over larger areas during droughts,
(Noser & Byrne, 2007b).
apparently drawing on the knowledge of the older females
All primates whose travel has been mapped utilize a net- about the locations of permanent, drought-resistant sources
work of familiar routes, whether these are arboreal runways of food and water.
or terrestrial paths. The impression of limited spatial mem- However, we cannot yet draw firm conclusions about the
ory that this hints at may be misleading, however; when cognitive mapping skills of elephants, as data are restricted
tamarins switched from their year-round fruit diet to feed to field observations for which the strength of evidence is not
on nectar, they travelled from tree to tree along quite differ- entirely apparent. For example, we now know that elephants
ent routes to their usual ones, routes that were highly direct can detect low frequency rumbles at distances of several
(Garber, 1988). A similar navigational efficiency was earlier kilometeres (Garstang, et al., 1995; Langbauer, 2000), and
noted in orangutans; when a favoured fruit crop failed and can even detect seismic rumbles (O’Connell-Rodwell, et al.,
individuals had to visit several trees, far apart in rainforest, 2006), so it may be that elephants can follow the sounds of
their travel was a least-effort route (Mackinnon, 1978). These distant thunder to reach fresh water sources. Further inves-
performances are often taken to imply Euclidian knowledge tigation is required to determine how well this hypothesis
of space; but an alternative exists, a richly interconnected accounts for the long-range movements of elephants. There
‘network-map’ that allows efficient search from any loca- are currently several study populations of African savannah
tion to any other (Byrne, 2000). Evidence that primate spe- elephants where the movements of multiple individuals are
cies do indeed rely on network-maps, rather than Euclidian being accurately mapped with GPS tracking devices, so fu-
knowledge equivalent to a 2D paper map, has come from ture insight into elephant cognitive mapping skills can be
both humans and baboons (Byrne, 1979; Noser & Byrne, expected.
2007a). Primate route planning goes beyond spatial informa- On a much smaller spatial scale, we recently showed that
tion: mangabeys have been shown to take into account the elephants are able to track the relative positions of their fam-
likely effect of warm, sunny weather on ripening fruit (Jan- ily members (Bates, et al., 2008b). We moved urine deposits
maat, Byrne, & Zuberbuhler, 2006). When passing close to a from known individuals to positions where they would be

---

Elephant cognition
69
discovered by target individuals. With samples from indi- breaking off a side branch or snapping them in half (Hart,
viduals who were at least 1 km away, urine from kin pro- Hart, McCoy, & Sarath, 2001). Elephants may thus be added
duced significantly more interest from the target individual to the small number of animals that make tools. However,
than samples from unrelated individuals. We also presented for an animal that frequently breaks branches while eating,
test elephants with urine deposits from related individuals the cognitive demands of extending this behaviour to manu-
actually present in their group that day, which were either facture fly switches are probably not great (Bates, Poole, &
walking some way ahead of the target elephant or behind Byrne, 2008). Elephant tool modification does not compare
it. We reasoned that if elephants are able to identify specific in complexity to the manufacture of ant and termite dipping
individuals from their urine, and each is continually updat- tools displayed by chimpanzees (Boesch & Boesch, 1990;
ing its memory of where other key individuals are, then dis- Goodall, 1986), which even includes making two different
covering a fresh urine deposit from an individual who was types of tool for different stages of the same task (Sanz &
walking behind should violate its expectations. Target indi- Morgan, 2007); or with the serrated leaf probes and hook
viduals investigated samples from family members behind tools produced by New Caledonian crows (Hunt, 1996,
them at the time of the test, significantly more than samples 2000a).
from individuals who were in front. From this, we concluded Detecting causal reasoning in animals has, understand-
that elephants are able continually to track the locations of ably, been a more elusive target. Several primate species
family members in relation to themselves, as either absent, have proved able to choose tools on the basis of relevant
present in front, or present behind (Bates, et al., 2008b). properties, even species like tamarins which do not naturally
These results suggest that elephants are able to hold in mind use tools (Hauser, 1997; Santos, Miller, & Hauser, 2003).
and regularly update information about the locations of at Chimpanzees also gave evidence of some sort of causal
least 17 other female party members, as well as implying understanding, when showing selectivity in what to copy
that they recognize individual identity from scent and have (Horner & Whiten, 2005). When presented with opaque
some understanding of invisible displacement and person puzzle boxes, chimpanzees copied two actions made by a
permanence. (We tested only scents of adult females, but human demonstrator to release food. When the same boxes
it is presumably likely that individuals keep track of males were transparent, revealing that one action was irrelevant,
and some immatures in addition.) That they can keep track they missed it out and copied only the relevant action. The
of so many, independently moving companions implies that judgement of relevance, however, may have been based on
elephants have particularly large working memory capacity. a simple parameter. For instance, because the ‘irrelevant’
action made no physical contact with the food, the chimpan-
Tools and causality
zees may only have attended to whether contact was made
when they choose which actions to copy. Even chimpanzees
With their opposable thumbs, primates are adept at object have not performed impressively when confronted with a
manipulation (Beck, 1980), with great apes having the great- ‘trap tube’ task, in which poking the desired food item from
est manual dexterity (Byrne, Corp, & Byrne, 2001; Napier, the wrong side causes it to fall into a well and be lost (Li-
1962). Tool use is widespread among animals, and is evident mongelli, Boysen, & Visalberghi, 1995); the few individuals
in all great apes and several monkey species (Beck, 1980). that solved the task did so only over several trials, raising
Tool use itself is not necessarily indicative of advanced cog- concerns that they did not understand the issue. In contrast,
nitive processing, however. Instead, systematic tool manu- rooks Corvus frugileus, another crow species that does not
facture or modification is recognised as cognitively more regularly use tools in the wild, solved the task rapidly and
demanding, and is evident in only a handful of species: all one individual showed immediate transfer to a different task
four great apes in captivity (McGrew, 1989), though only that relied on the same basic insight (Seed, Tebbich, Emery,
chimpanzees and orangutans in the wild (Fox, Sitompul, & & Clayton, 2006).
Van Schaik, 1999; McGrew, 1992); and the New Caledonian Evidence of causal understanding in elephants is sparse.
crow (Hunt, 1996, 2000b).
Nissani (2004) reports a string pulling experiment and a tube
Both Asian and African savannah elephants have been task conducted with two zoo-based female Asian elephants.
seen to use multiple tool types for up to six different func- The elephants were required to pull a retractable cord in or-
tions, mostly in the context of body care such as scratching der to obtain a food reward in the first experiment, and suck
and removing ticks, with one report of throwing mud during or blow through a tube to gain a reward in the second. Of the
a competitive encounter with a rhino over access to a water- two subjects, one performed reasonably well, the other less
hole (Chevalier-Skolnikoff & Liska, 1993; Wickler & Seibt, so, and careful analysis of the pattern of results suggested
1997). Asian elephants presented with branches that were that performance was dependent on trial and error learning,
too long or bushy to make effective fly switches (fly-swats), rather than an understanding of the causal relationships be-
a commonly used tool, modified them before use by either tween the action and the outcome. Nissani also reported an

---

Elephant cognition
70
apparent lack of causal understanding by working Asian el- learn the correct target, showing that she developed a learn-
ephants, in dealing with a modified discrimination task (Nis- ing set as do Old World monkeys and apes (Passingham,
sani, 2006). The subjects were trained to remove food from 1981). The elephant was re-tested with the same visual dis-
a bucket only after they had touched the bucket lid, which crimination pairs after a one-year delay, and she performed
during training was always placed on top of the bucket. In with 73-100% accuracy. As well as straightforward discrim-
the test phase, the lid was placed next to the bucket instead ination learning, there was some evidence that the elephant
of on top of it, so there was no need to touch it before ac- could transfer what she had learned about the features of the
cessing the food. Only in 3 of 77 trials did any elephants positive stimulus, and apply it correctly to novel stimuli. For
ignore the lid and reach straight into the bucket for the food. example, she appeared to generalize over orientation. In the
On the other trials, the elephants touched or even threw the initial tests a + symbol was the correct choice; when subse-
lid away, as they had previously been trained to do. Nissani quently presented with an x symbol for the first time, she im-
argued that this showed a lack of causal understanding by mediately recognised it as correct. Generalization over size
the elephants, but it might instead represent a lack of under- was also shown. The elephant was initially asked to discrim-
standing of the task demands. Working elephants are trained inate between two stimuli showing black and white stripes,
from an early age to follow precise sequences of behaviour, with the bands placed 2cm apart in the positive stimulus and
and are punished for any deviation. Thus, in trials where the 4cm apart in the negative stimulus. At a later date she was
lid was placed next to the bucket, the subjects may have seen presented with novel striped stimuli, this time with the spac-
that the lid was now irrelevant, but nevertheless interpreted ing either 1.5cm or 2cm. She chose the correct stimulus of
the task as one of following the trained practise. Until this the thinnest stripes – 1.5cm, even though in previous trials
experiment is repeated on animals that are normally allowed the 2cm stripes had been the correct choice. However, with
to exercise their behavioural choices freely, we should prob- only two such examples in this paper, it is difficult to draw
ably not draw conclusions about elephants’ lack of causal firm conclusions. In contrast to Rensch’s paper, a discrimi-
understanding from it.
nation task conducted on working Asian elephants (Nissani,
Indeed, some understanding of physical causality is sug- Hoefler-Nissani, Tin Lay, & Wan Htun, 2005) reported that
gested by the results of testing elephants on a well-known some of the 20 animals tested never learnt to pass the tests
Piagetian task that depends on an understanding of the con- although others did perform comparably to the young female
cept of support (Irie-Sugimoto, Kobayashi, Sato, & Hasega- tested by Rensch. There was an age effect in these results,
wa, 2008). Elephants were given a series of problems in each however, with more individuals over the age of 20-30 years
of which food bait was out of reach; but in some cases a bait unable to acquire the discriminations.
item was supported by a tray which the elephant was able to We conducted a field experiment in the Amboseli National
pull. In all cases, one of the two elephants tested performed Park, Kenya, that may be seen as a more ecologically valid
above chance in selecting the correct tray to pull, showing test of elephant feature learning and categorization, present-
that it took a means-end approach to the problem and had ing individually known elephants with garments that gave
some understanding of the notion of physical support.
either visual or olfactory information about their human
wearers (Bates, et al., 2007). We used garments that had
Discrimination learning, features and categories
been worn by members of two different ethnic groups that
pose different levels of danger to elephants. In the first set
Tomasello and Call (1997) review extensive evidence of trials, we separately presented elephant groups with three
that apes and monkeys are generally adept at learning fea- different red cloths, using a within-subjects design. Each of
ture discriminations and categories, both natural and artifi- the cloths had been worn either by a Masaai warrior, or a
cial, In this, they do not differ from pigeons and rats, the similar aged man from the agricultural Kamba tribe, or by
animals more typically used in animal learning laboratories. no one at all. The only thing that differed between the cloths
Evidence suggests that elephants are similarly capable of was therefore the smell, derived from the ethnicity or life-
discrimination learning, and moreover, that they have clas- style of the wearers. With access only to this olfactory in-
sification abilities that may exceed those of many animals.
formation, the elephants showed significantly greater flight
One of the first systematic tests of elephants’ discrimina- reactions to garments worn by Masaai warriors than similar
tion learning ability was conducted by Rensch (1957) as he age Kamba men. In a second experiment, elephants in the
explored the long-term retention of a captive Asian elephant. same population were presented with two cloths that had not
Rensch taught a juvenile female Asian elephant 20 different been worn by anyone, but here one was white - a neutral
visual discrimination pairs and six acoustic discrimination stimulus, and the other was red - the colour that is ritually
pairs; one pattern of each pair was rewarded. It took the fe- worn by Masaai warriors. With access only to these visual
male 330 trials to learn the first visual discrimination, but by cues, the elephants showed significantly greater reactions to
the fourth and subsequent pairs it took her only ten trials to garments worn by Masaai warriors than similar age Kamba

---

Elephant cognition
71
men, often including signs of aggression. We concluded that infants in similar relative quantity judgement tests (see An-
elephants are able to categorize a single species (humans) derson, Stoinski, Bloomsmith, & Maple, 2007; Beran, 2001;
into subclasses, an ability that relies on sophisticated dis- Boysen, Bernston, & Mukobi, 2001; Feigenson, Carey, &
crimination learning and classification (Bates, et al., 2007). Hauser, 2002; Xu & Spelke, 2000). It is not yet known what
The numerous vocal, gestural, and chemical signals used cognitive mechanisms underlie the Asian elephants’ numeri-
by elephants also suggest considerable classification abili- cal ability, but it may be that elephants are able to keep track
ties in elephants. Over a hundred different gestural displays of a larger number of items in immediate, working memory
have been described for African elephants (Kahl & Arm- than can great apes including humans, and that when relative
strong, 2000; Poole & Granli, 2003); tens of different vo- quantity judgements of larger numbers of items are explored
calisations have been recorded (Langbauer, 2000; Poole, then disparity and magnitude effects will emerge. If so, then
Payne, Langbauer Jr, & Moss, 1988); and chemical signals there is a clear parallel with the ability of African savannah
are known to accurately indicate sexual states in both males elephants to keep track of the movements and positions of a
and females (Rasmussen & Schulte, 1998). Although recog- remarkable number of family members. Indeed, the unusual
nition of most of the visual and vocal displays has not been numerical abilities of elephants may derive, in evolutionary
formally tested, the fact that so many distinct signals exist terms, from the elephant’s need to monitor and coordinate
is suggestive of subtle discrimination learning and elaborate movement of their extensive families.
categorisation. Where playback experiments have been used
to test reactions, this notion has been supported (McComb,
The social environment
Moss, Sayialel, & Baker, 2000; Poole, 1999). From their re-
actions, it was clear that elephant calls convey information Knowing about others and their interactions
about musth state, and some information about the caller,
at least familiarity and perhaps individual identity. Thus, al- Whilst individual recognition is considered an important
though primates have been tested much more extensively in component of animal social life, experimental demonstra-
this area, there is reason to think that knowledge of features tions of the ability are relatively sparse. Monkey and ape
and categories that can be learnt by elephants, and their abil- species, however, are known to understand both the direct
ity to categorize and classify entities in the world, may at and third party relationships of others (Cheney & Seyfarth,
least match those of any primate.
1990), and alliances and coalitions with specific individuals
are a prominent feature of monkey and ape sociality (Har-
Quantities
court & deWaal, 1992). Note that knowledge of third party
relationships is not restricted to large brained species. Some
Rensch (1957) presented his five year-old female Asian fish and bird species have also been found able to track the
elephant with number discriminations: she was able to dis- relative relationships of third parties, through what has been
tinguish 3 from 4 dots regardless of their arrangement and termed ‘eavesdropping’ on the behaviour of others (e.g.
spread. Irie-Sugimoto et al (2008) tested the relative quantity Oliveira, McGregor, & Latruffe, 1998; Otter, et al., 1999).
judgements of Asian elephants. They presented five captive Our urine-moving experiments, discussed above, showed
Asian elephants with two baskets containing different quan- that elephants have knowledge of individual identities, and
tities of food, and the elephant had to choose the basket with are able to recognise and keep track of at least 17 differ-
the larger amount. All five elephants chose the larger quan- ent female family members (Bates, et al., 2008b). McComb
tity significantly more often than the smaller, performing et al (2000), using experimental play back of long-distance
with 67-89% accuracy. Elephants were as good at picking contact calls in the same Amboseli population, showed that
the larger quantity when it was only slightly bigger (e.g. 6:5) each adult female elephant was familiar with the contact-
than when it was considerably bigger (e.g. 5:1), and perfor- call vocalisations of individuals in an average of 14 families
mance did not vary with the total number of items present- in the population, totalling around 100 elephants. When the
ed, up to 12. In a second experiment, four additional Asian calls were from a familiar family, i.e. one that had previously
elephants watched and listened to the baskets being baited, been shown to have a high association index with the test
but they could not see the final amounts in the baskets. All group, the test elephants contact-called in response and ap-
elephants chose the basket containing the larger amount, sig- proached the location of the loudspeaker. When a test group
nificantly more often than expected by chance, performing heard unfamiliar contact calls (from groups with a low asso-
at 72-82% accuracy. As in the first experiment, the elephants ciation index with the test group), their spatial cohesion in-
did not exhibit disparity or magnitude effects, in which per- creased, and they retreated from the area. Whether this vocal
formance declines with a smaller difference between quanti- familiarity is based on individual recognition is uncertain,
ties, or the total quantity increases, respectively, in striking however. It currently remains untested whether elephants
contrast to the performance of great apes and even human understand and take advantage of third-party relationships.

---

Elephant cognition
72
Field observations of African elephants show clear co- elephants renders deception an inappropriate way for them
alitions and alliances (Bates, et al., 2008a; Moss & Poole, to manipulate others.
1983), but reciprocity is less clear – there may be an ele-
ment of this in the formation of coalitions, but it has not been Social learning
demonstrated formally. Asian logging elephants frequently
work together to roll heavy logs up ramps, which is certainly Although as yet there are no formal studies of social learn-
cooperative, but they have been trained to work like this so ing in elephants, or observations of behavioural traditions, it
may not themselves understand the cooperation (Rensch, would be unwise to conclude that such a long-lived, slowly
1957).
maturing, highly social species does not learn through ob-
Cooperative problem solving is observed fairly regularly serving group members. The impressive spatial knowledge
in long-term behavioural studies of African elephants (Moss, shown in some populations, for instance, is surely acquired
1988). For example, two or more individuals may work to- by young individuals from following older, more knowl-
gether to help individuals that are trapped by muddy river edgeable relatives.
banks or drainage ditches (Bates, et al., 2008a), or to chase Elephants may eventually be shown to possess social
off vehicles when an individual is darted for veterinary pur- learning abilities absent in non-human primates. For in-
poses (Moss, personal observation). In the Amboseli popu- stance, there is some evidence of vocal imitation in African
lation, related individuals have also been observed to form elephants (Poole, Tyack, Stoeger-Horwath, & Watwood,
coalitions when attempting to retrieve infants that have been 2005), and even one intriguing report of a captive Asian ele-
commandeered by other, unrelated families (Moss, personal phant copying human speech, although this has not yet been
observation), and this could also be viewed as cooperative formally verified. Non-human primates lack any significant
problem solving.
abilities in vocal imitation (Janik & Slater, 1997). Also, cur-
rently unpublished data suggests that older female African
Social strategies and communication
elephants may teach young, naïve, nulliparous females how
to behave when they come into oestrous for the first time
African savannah elephants are known to have an exten- (Bates et al, in preparation). Despite some suggestive obser-
sive gestural and vocal repertoire (Poole & Granli, 2003). vations (Boesch, 1991), there is no consensus that any spe-
Moreover, as previously noted, they discriminate the contact cies of non-human primate is able to teach (Caro & Hauser,
calls of familiar and strange individuals (McComb, et al., 1992; Thornton & Raihani, 2008)
2000). Social knowledge apparently accrues with age: old Thus, while there is little evidence on which to compare
females have the best knowledge of the contact calls of other the social learning skills of elephants directly with those of
family groups (McComb, Moss, Durant, Baker, & Sayialel, the extensively studied primates, this is representative of a
2001). Monkey and ape vocalizations sometimes encode lack of research effort on the topic rather than a lack of social
‘functionally referential’ information, that is, hearers react learning ability in elephants. Reports of vocal imitation and
to the calls in the same way as they would to entities in the potential teaching behaviour suggest this is one area where
world, such as specific predators (Seyfarth, Cheney, & Mar- greater research effort could prove particularly fruitful and
ler, 1980; Zuberbuhler, 2000). Whether anything similar is instructive.
latent in elephant vocalizations, however, is unknown.
Deception and manipulation have been hot topics in pri- Theory of Mind
mate work since the 1980s (Byrne & Whiten, 1988; de Waal,
1982). Deceptive tactics have been reported in all taxa of Studies of primate cognition that are embraced by the
primates, with frequency of use well predicted by the spe- portmanteau term ‘theory of mind’ include those on under-
cies’ neocortex size (Byrne & Corp, 2004), and in great ape standing the gaze and perception of others, understanding in-
species at least some deception appears to be done intention- tentions and attention, understanding knowledge and beliefs,
ally, with some understanding of others’ mental states (By- and understanding the self. There has been much speculation
rne & Whiten, 1992). In elephants, there is only one report of as to the cognitive underpinnings of theory of mind and self-
possible deception (Morris, 1986): certain captive elephants recognition abilities, but as yet there is no consensus (Byrne
in a zoo were noted to finish their ration of hay quickly, and & Bates, 2006; Gallup, 1985; Povinelli & Vonk, 2003; To-
then move near others still eating, swinging their trunks in masello, Call, & Hare, 2003).
an ‘aimless’ manner, but occasionally eating some of their There is experimental evidence that many species of pri-
hay. It is not clear, however, whether the trunk-swinging mate follow gaze, even where the gaze is directed behind ob-
was used tactically, and whether the other elephants were structions (Tomasello, Hare, & Agnetta, 1999). Puzzlingly,
deceived. Researchers on wild populations report no decep- individuals of the same species often fail to give evidence
tion at all, so it may be that the extensive social network of that they can use gaze as a cue to object choice, but recent

---

Elephant cognition
73
to understand the operation of mirrors: continuing to make
social responses, as if to another individual of their species,
or habituating to mirrors altogether and paying no further at-
tention to them. Monkeys, however, do learn to understand
the geometric properties of mirrors, using them to detect the
presence of individuals that appear behind them, or to reach
food rewards more successfully by locating their reflection
in a mirror (Anderson, 1984). However, they signally fail
to recognize a reflection of their own face as such, despite
extensive experience. In contrast, members of all species of
great apes can learn to interpret a mirror reflection or CCTV
image of their face as ‘themselves’, although not all individ-
uals ever do so (Gallup, Jnr, 1970, 1982; Patterson & Cohn,
1994).
Recent analysis of behavioural records of the African sa-
vannah elephants of Amboseli NP suggests that they under-
stand the emotions and the intended goals of others, acting
empathically towards individuals who are distressed and
helping them in ways that were appropriate to their predica-
ment (Bates, et al., 2008a). A description of the reactions
to a dying and subsequently dead matriarch by both fam-
ily members and unrelated elephants further suggests that
elephants can act empathically (Douglas-Hamilton et al,
2006). However, Bates et al (2008a) did not find any posi-
Figure 2. Collecting earth soaked with recently-deposited
tive evidence for the capacity to understand others’ attention,
urine, for use in conducting “expectancy violation” experi- knowledge or beliefs. The same could be said of field studies
ments at Amboseli. The presence of an individual is artifi-
of apes, so explicit tests of elephants’ understanding of oth-
cially simulated by moving urine deposits. Photograph by
ers’ mental states may be worthwhile.
the authors.
Nissani (2004) modified experiments devised by Povi-
work suggests a resolution. Lemurs were found to use gaze nelli and Eddy (1996) to examine what two zoo-based Asian
successfully as a cue, by following gaze and then choosing elephants understood about visual attention. Povinelli and
objects in the direction they were looking, a tendency that Eddy had found that chimpanzees chose to beg from a per-
the researchers called ‘gaze priming’ (Ruiz, Gómez, Roeder, son whose whole body was oriented toward them rather than
& Byrne, 2009). However, neither gaze following nor gaze oriented away, and these two elephants performed similarly.
priming were done rigidly and success rates were relatively However, chimpanzees did not discriminate between people
low, perhaps explaining apparent failures in earlier work according to their head orientation alone: for instance, when
that did not test gaze following and cueing in the same task. offered a choice between two people standing sideways-on,
In addition, field observations suggest that monkeys and one with the head turned 90 degrees to look at the chimpan-
apes are able to understand the geometric perspective of oth- zee subject, and one not, or between two people both whole
ers (Byrne & Whiten, 1992); this has been confirmed ex- body oriented toward them but one with the face covered or
perimentally only in the case of the chimpanzee (Hare, Call, with a bucket over their head, the chimpanzees performed at
Agnetta, & Tomasello, 2000). The observational data, based chance. Nissani gave 8 trials of each sort to each of the two
on collated records of deception, also implied that only apes elephants, and none of the results were individually signifi-
are able to take account of the knowledge of competitors cant. However, with sideways body presentation it appeared
(Byrne & Whiten, 1992). Again, this was finally confirmed that the elephants were able to take account of the person’s
experimentally in the case of the chimpanzee (Hare, Call, & face orientation (6/8 and 7/8 trials successful, respectively).
Tomasello, 2001; Tomasello, et al., 2003) and the orangutan It may indeed be that straight copies of studies used with
(Cartmill & Byrne, 2007), after several years of explicit de- monkeys and apes are systematically unsuitable for prob-
nials of the possibility (Povinelli, 1993; Povinelli & Eddy, ing elephant cognition; new and creative experimental de-
1996; Tomasello & Call, 1997).
signs may be required. Monkeys and apes predominantly
An intriguingly similar distribution of ability across pri- use vision to learn about the social world. For elephants, au-
mate phylogeny has also been uncovered by study of ani- dition and olfaction are demonstrably more important than
mal reactions to mirrors. Most animals plainly cannot learn vision, with comparatively much larger brain areas dedicat-

---

Elephant cognition
74
ed to these areas than visual cortex (Hakeem, et al., 2005; so different to that observed in the great apes that recognize
Shoshani, et al., 2006). It therefore seems probable that tests themselves in mirrors that it cannot be considered definitive;
using auditory or olfactory stimuli will be more relevant to but we suspect that future data will show that elephants do
elephants, and more likely to accurately measure their men- indeed have the competence to recognize themselves in a
tal abilities.
mirror.
One area in which it will not be possible to avoid the visu-
al domain is that of mirror self-recognition; accounts of two
Summary
tests in Asian elephants have been published, with somewhat
contradictory results. Both relied on Gallup’s (1970) ‘mark A recurring theme throughout this survey has been the
test’ paradigm. In this, the subject is first given extensive stringent limitations on the extent to which elephant and pri-
experience with a mirror. If signs of self-recognition are ob- mate cognition can be satisfactorily compared. It is not just
served, such as self-monitoring while making repetitive or that less is known, for both practical and historical reasons,
unusual movements, then the mark test is applied. Surrep- about elephant cognition, although that is certainly true.
titiously, for instance while the subject is anaesthetized, a But in addition the positive things that we do know about
conspicuous mark is applied to its body in two places, one cognition within the two families are often hard to com-
visible directly and one not. The subject is then observed for pare. Repeatedly, the ready availability, ease of manipula-
a period without access to a mirror, to control for the possi- tion and straightforward motivation of captive primates has
bility that the marks can be detected by scent or tactile sen- enabled experiments to be done which are not easily done
sation. Provided the subject examines only the visible mark, with elephants, and have yet to be attempted. At other times,
and not the concealed one, the mirror is then restored. Then, experiments with elephants have revealed abilities that are
once the subject catches sight of its image in the mirror, the not proven with primates; for instance, our studies with ol-
critical observation – seen in many great apes – is that it factory social stimuli enabled demonstration of individual
suddenly touches and explores the concealed mark, while identity recognition in elephants, an ability not so far shown
monitoring the mirror to guide its hands. Povinelli (1989) experimentally in primates. Yet the differences we have re-
followed this procedure when testing two elephants: how- viewed are often more likely, perhaps, to reflect differences
ever, he observed no signs of self-recognition and both indi- in study methods and facilities available than to point to any
viduals subsequently failed the mark test (see also Nissani & profound difference in cognition.
Hoefler-Nissani, 2007). He concluded that Asian elephants Undoubtedly, elephants perform well on laboratory tests
do not show self-recognition. However, these elephants were of learning, discrimination and memory, and spontaneously
only given a few days exposure to the mirror prior to testing; engage in simple forms of tool use – no obvious differences
chimpanzees that have passed the mark test have typically from primates are apparent here. Similarly, it seems likely
had weeks or months of prior mirror exposure. In the sec- that elephants, like the great apes, have the cognitive ca-
ond experiment, because their three elephants could not be pacity to recognize themselves in a mirror. When it comes
separated, Plotnik et al (2006) used sham marking to control to categorization, some intriguing findings have recently
for the possibility that the subjects might be able to detect emerged that may point to capacities unusual for mammals.
a mark by non-visual means. They report that one of three In number discrimination, elephants like primates are able to
adult females they tested did show mirror self-recognition: distinguish quite small quantity differences – yet strangely
she touched the visible mark several times, but never the they showed no effect of the size difference in making their
sham mark. Moreover, all three elephants showed sugges- judgements, nor any variation in performance with the total
tive behaviour in front of the mirror, prior to any marking, number of items presented. One possibility is that this lack
most strikingly cases where “the elephant is standing AT the of disparity and magnitude effects is a result of unusually
mirror, and moves its head in and out of mirror view, like a large working memory capacity: for elephants, even groups
kid playing with his mirror image by running in and out of of 5-6 items can be appreciated and compared in the man-
view of it” (Josh Plotnik, personal comm.). However, with ner that humans and other great apes can appreciate groups
only one subject giving a positive response to the mark test, of 2-3. Wild elephants have been shown to sub-categorize
it is crucial that this response should be a convincing one. humans into groups according to the varying levels of risk
The initial response shown by the ‘successful’ elephant, af- to them that different groups present, and to make this cat-
ter spending time in front of the mirror, was to walk away egorization on the basis of scent or colour, independently.
for 7 minutes. Then she returned and moved in and out of In the social realm, there is no doubt that elephants show
mirror-view a couple of times, then moved away again, still empathy into the problems faced by others as well as react-
having shown no mark-touching. Finally, she moved away ing to their expressed emotions, but much more work needs
from the mirror and only then, when away from the mirror to be done before the initial hints of elephant abilities in
did she first touch the mark with her trunk. This pattern is cooperation, imitation and teaching can be properly under-

---