Dolphins (Tursiops truncatus) comprehend the referential character of the human pointing gesture

Abstract

The authors tested a dolphin's (Tursiops truncatus) understanding of human manual pointing gestures to 3 distal objects located to the left of, to the right of, or behind the dolphin. The human referred to an object through a direct point (Pd), a cross-body point (Px), or a familiar symbolic gesture (S). In Experiment 1, the dolphin responded correctly to 80% of Pds toward laterally placed objects but to only 40% of Pds to the object behind. Responding to objects behind improved to 88% in Experiment 2 after exaggerated pointing was briefly instituted. Spontaneous comprehension of Pxs also was demonstrated. In Experiment 3, the human produced a sequence of 2 Pds, 2 Pxs, 2 Ss, or all 2-way combinations of these 3 to direct the dolphin to take the object referenced second to the object referenced first. Accuracy ranged from 68% to 77% correct (chance = 17%). These results established that the dolphin understood the referential character of the human manual pointing gesture.

Full text

Journal
of
Comparative Psychology
1999, Vol.
113,
No. 4,
347-364
Copyright
1999
by the
American Psychological Association,
Inc.
0735-7036/99/S3.00
Dolphins
(Tursiops
truncatus}
Comprehend
the
Referential Character
of
the
Human Pointing Gesture
Louis
M.
Herman
University
of
Hawaii
at
Manoa
and
Kewalo
Basin Marine Mammal Laboratory
Sheila
L.
Abichandani,
Ali N.
Elhajj,
Elia
Y. K.
Herman,
and
Juliana
L.
Sanchez
Kewalo
Basin Marine Mammal Laboratory
Adam
A.
Pack
Kewalo
Basin Marine Mammal Laboratory
and The
Dolphin Institute
The
authors
tested
a
dolphin's
(Tursiops
truncatus) understanding
of
human manual pointing
gestures
to 3
distal objects located
to the
left
of, to the right of, or
behind
the
dolphin.
The
human
referred
to an
object through
a
direct point
(Pd),
a
cross-body point
(Px),
or a
familiar
symbolic gesture
(S).
In
Experiment
1,
the
dolphin responded correctly
to 80% of Pds
toward
laterally placed objects
but to
only
40% of Pds to the
object behind. Responding
to
objects
behind
unproved
to 88% in
Experiment
2
after exaggerated pointing
was
briefly instituted.
Spontaneous comprehension
of Pxs
also
was
demonstrated.
In
Experiment
3, the
human
produced
a
sequence
of 2
Pds,
2
Pxs,
2 Ss, or all
2-way
combinations
of
these
3 to
direct
the
dolphin
to
take
the
object referenced second
to the
object referenced
first.
Accuracy ranged
from
68% to 77%
correct (chance
=
17%).
These
results
established
that
the
dolphin
understood
the
referential character
of the
human manual pointing gesture.
Manual
pointing
(use
of the
deictic gesture)
by
humans
serves
to
call another's attention
to an
object,
an
event,
or a
place
of
interest
to the
pointer.1
Human pointing
is
thus
a
social
triadic
transaction, involving
the
coordination
of the
attention
of the
pointer
and the
observer
to the
same target
or
event.
Implicit
in
adult human pointing behavior
is not
only
the
intent
to
manage
the
attention
of
another individual
but
also
an
expectation that
the
individual understands
the
referring
function
of the
pointing behavior. Likewise,
the
human
observer understands
the
intent
of the
pointer
to
direct attention toward
a
particular object, event,
or
place.
The
understanding
of the
referential
function
of the
Louis
M.
Herman, Department
of
Psychology, University
of
Hawaii
at
Manoa
and
Kewalo Basin Marine Mammal Laboratory,
Honolulu, Hawaii; Sheila
L.
Abichandani,
Ali N.
Elhajj,
Elia
Y. K.
Herman,
and
Juliana
L.
Sanchez, Kewalo Basin Marine Mammal
Laboratory, Honolulu, Hawaii; Adam
A.
Pack, Kewalo Basin
Marine Mammal
Laboratory,
Honolulu,
Hawaii,
and The
Dolphin
Institute, Honolulu, Hawaii.
Ali
N.
Elhajj
is now at the
Department
of
Psychology, University
of
Hawaii
at
Manoa.
Elia
Y. K.
Herman
is now at the
Independent
Studies Concentration, Brown University.
This project
was
funded
in
part
by
grants
from
Earthwatch
and
its
Research
Corps;
by financial
support
from
The
Dolphin
Institute;
and by
equipment donations
from
Apple Computer,
Inc.,
and
Mitsubishi,
Inc.
We
thank
the
staff,
students, interns,
and
volunteer
participants
at
Kewalo Basin Marine Mammal Laboratory
who
contributed
to
this
project.
We are
grateful
to
Matthias
Hoffmann-Kuhnt,
who
pre-
pared
the
computer images from
the
digital
videotape
record.
Correspondence concerning this article should
be
addressed
to
Louis
M.
Herman, Kewalo Basin Marine Mammal Laboratory,
1129
Ala
Moana Boulevard, Honolulu, Hawaii
96814.
Electronic
mail
may be
sent
to
lherman@hawaii.edu.
human
manual pointing gesture follows
a
developmental
progression
in
infants.
Lempers (1979) noted that beginning
at
9
months
of
age,
infants
were able
to
direct their attention
to
objects
50 cm
from
the
extended
arm and
index
finger
of
an
adult
but did not
understand points
to
objects
at a
distance
of 2.5
m
until
the age of 12
months. Lempers
thus
distinguished
between
prereferential
comprehension,
hi
which
the
object
and the
pointing
finger
must
be
close together,
and
referential
comprehension,
in
which
the finger can be
distant
from
the
object.
Butterworth
(1991)
and
Butterworth
and
Grover
(1988,
1990)
reported
findings
similar
to
those
of
Lempers:
Infants
younger than
12
months usually only
looked
at the
extended
finger or at
objects near
it
(see
also
Desrochers,
Morissette,
&
Ricard,
1995),
but
beginning
at
about
12
months, they could
follow
the
direction
of the
pointing
finger
toward more distant objects. Furthermore,
at
15
months,
infants
were able
to
disregard nearby objects
in
the
path
of the
point
to
look
at
more distant objects.
Morissette, Ricard,
and
Decarie
(1995)
generally
confirmed
this
progression, although they
found
a
somewhat longer
developmental
period.
At 12
months,
infants
attended primar-
ily
to
objects close
to the
pointer's hand. Attention
to
more
distant objects
did not
emerge until approximately
15
months
of
age.
The findings of
Morissette
et
al.
were based
on
a
longitudinal
study,
whereas those
of
Butterworth
and
Grover
were based
on a
cross-sectional approach.
In
addi-
tion,
Morissette
et al.
considered
nonresponses
by the
infants
(trials
on
which
the
infants
offered
no
response
to the
adult's
point)
as
incorrect, whereas Butterworth
and
Grover
ex-
cluded nonresponses
from
their data.
The
more complete
data
set of
Morissette
et al. and the
greater control over
1
For a
general discussion
of
deictic
expressions
in
human
language,
see,
for
example, Anderson
and
Keenan
(1985).
347

348
HERMAN
ET
AL.
participant variability
offered
by
longitudinal studies sug-
gest that,
on
average, infants' referential comprehension
of
pointing
may
develop
closer
to 15
months
of age
than
to 12
months
of
age.
By
approximately
15-18
months,
infants
also have devel-
oped
a
sense
of the
"other"
(Baldwin
&
Moses, 1994),
in
that
they recognize that
the
individual pointing
at an
object
is
subjectively
attending
to it.
This awareness
may be one of
the
early
precursors
of the
emergence
of a
"theory
of
mind"
in
children (Povinelli
&
Preuss,
1995). Early indicants also
have
been noted
in the
production
of
pointing gestures
by
infants.
At
approximately
10-15
months,
infants
begin
to
use
triadic
pointing gestures
to
attract
the
attention
of
adults
to
distal events, rather than
to
themselves,
and to
engage
in
communicative pointing
(protodeclarative
pointing, point-
ing
while looking back
at the
adult;
Desrochers
et
al.,
1995;
Tomasello
&
Camaioni,
1997).
The
close evolutionary relationship between humans
and
great apes
has
suggested
to
some researchers that
protolan-
guage
in the
form
of
referential gestures
may
occur among
great apes (see, e.g., Tomasello
&
Camaioni,
1997).
One
might expect,
for
example, that apes
in
their natural world
would
show capabilities
for
using
and
understanding manual
pointing gestures
to
refer other members
of
their species
to
distal events
or
objects
of
interest. However, such naturally
occurring behaviors have
not
been observed
(Goodall,
1986;
Povinelli
&
Davis, 1994; Povinelli
&
Preuss, 1995;
Premack,
1984; Savage-Rumbaugh, 1986; Tomasello
&
Call,
1994).
"Enculturated"
apes—apes
that have been reared
in a
human
environment
and
that have been explicitly
or
implic-
itly exposed
to the
functional
aspects
of a
human's pointing
gesture—do
learn
to
produce manual pointing gestures
in
social
triadic interactions
and to
direct
the
attention
of
humans
to
distal objects
(e.g.,
Greenfield
&
Savage-
Rumbaugh,
1990; Miles, 1990;
Woodruff
&
Premack,
1979).
Also,
the
primary chimpanzee subject studied
by
Leavens, Hopkins,
and
Bard
(1996)
exhibited whole-hand
or
single-digit points
to
food
items that were
out of
reach
significantly
more
often
when
a
human
was
present than
when
the
chimpanzee
was
alone
and
also directed
his
gaze
toward
the
human. This chimpanzee
had
been extensively
tested
in a
variety
of
experiments using
an
automated
joystick
for
control
of
stimuli
but was
neither language
tutored
nor
explicitly trained
to
point. Hence,
in
general,
pointing
may
emerge under
a
variety
of
conditions
in
which
chimpanzees have extensive human contact.
However,
in
contrast
to
these instances
of the
emergence
of
referential manual pointing
in
chimpanzees, Povinelli,
Reaux,
Bierschwale,
Allain,
and
Simon
(1997)
presented
data showing that adolescent common chimpanzees failed
to
referentially
comprehend
the
human pointing gesture.
All
seven
of the
chimpanzees
tested
by
Povinelli
et al. had had
prior extensive, though informal, exposure
to
pointing
by
humans.
A
formal test examined
the
chimpanzees' ability
to
use
the
experimenter's manual point
to
choose
between
two
boxes.
Prior
to a
chimpanzee subject entering
the
test
enclosure,
the
experimenter
sat at a fixed
distance
from
the
boxes while maintaining
a
pointing gesture toward
the box
containing food.
On
entering
the
test enclosure,
the
chimpan-
zee
viewed
the
boxes
and the
pointing posture
of the
experimenter through
a
Plexiglas partition.
Two
types
of
pointing gestures were used:
direct
pointing (the
arm
with
the
index
finger
extended
was
pointed directly toward
the
baited box)
and
cross-body pointing (the
arm on the
opposite
side
of the
target
box was
extended across
the
body
at the
midsection
with
the
index
finger
pointing toward
the
baited
box).
In
both conditions,
the
chimpanzees responded accu-
rately when
the
experimenter's outstretched
finger
almost
touched
the
referenced
box
but,
on
average, performed only
at
chance levels when
the box was at a
distance
of
approximately
90 to 130 cm
from
the finger. In
contrast,
3-year-old children tested similarly with
the
direct-point
procedure
and
2-year-old children tested
with
the
cross-
point
procedure responded almost without error
to the
correct box, regardless
of
distance. Povinelli
et al.
concluded
that
the
chimpanzees, unlike
the
children,
did not
interpret
the
pointing gesture
in a
referential manner.
Povinelli
et al.
(1997) argued
further
that three other
formal
studies
of
pointing comprehension
by
great apes
(Call
&
Tomasello, 1994;
Menzel,
1974; Povinelli, Nelson,
&
Boy
sen, 1992) also failed
to
establish referential under-
standing.
In two of
these studies (Call
&
Tomasello, 1994;
Povinelli
et
al.,
1992),
the
experimenter's hand
was
always
very
close
to the
target. Comprehension
of
pointing gestures
toward
distal objects
or
locations,
which
is
essential
for
evaluating
the
referential understanding
of the
pointing
gesture,
was not
assessed.
In the
third study (Menzel, 1974),
Povinelli
et al.
(1997) claimed that
the
methodological
information
that
was
reported
was not
sufficient
to
allow
for
an
assessment
of
referential understanding. Tomasello
and
Camaioni
(1997)
also noted
the
lack
of any
substantial
comprehension
of the
human pointing gesture
by the two
orangutans studied
by
Call
and
Tomasello, stating that
the
orangutan
reared
by
humans
(Chantek)
performed "barely
above
chance"
and the
other, trained specifically
to
point
for
food
within
a
specific context, showed
"no
signs
of
compre-
hension"
(p.
16). Savage-Rumbaugh
(1986),
although
not
reporting
a
formal
study, also remarked that
her
common
chimpanzees
did not
understand
the
referring
function
of the
human
manual pointing gesture. Call
and
Tomasello stressed
that,
for
infants
or for
chimpanzees, there
is no
clear-cut
relationship between comprehension
and
production
of
communicative gestures.
One may be
present
but not the
other (see also Herman, 1987; Herman
&
Morrel-Samuels,
1990).
In
this article,
we
report
on
studies
of
comprehension
of
the
human manual pointing gesture
by
bottlenosed
dolphins
(Tursiops
truncatus).
Referential understanding
of
this ges-
ture
by
dolphins,
if
demonstrated, would compel
a
broader
comparative analysis
of the
roots
and
meaning
of
such
understanding (cf. Tomasello
&
Camaioni, 1997).
Why
would
such understanding
be
demonstrable
in a
species that
has no
manual gestural system
in its
natural world?
Is
such
referential
understanding derivable simply because
the
manual
gesture
is an
abstraction
or a
symbol,
learnable
by
almost
any
highly social, large-brained species?
Or, is
such
referential
understanding dependent
on the
existence
of
some
form
of
triadic distal referencing
in the
natural world

DOLPHINS
COMPREHEND
HUMAN
POINTING
349
of
the
species
examined?
If the
latter
is the
case, then what
are the
selective advantages
for the
species
of the
evolution
of
such types
of
distal referencing? Prior brief investigations
of
comprehension
of
pointing
by a
dolphin
are
summarized
elsewhere
as
part
of
more general reviews
of
dolphins'
cognitive competencies (Herman, Pack,
&
Morrel-Samuels,
1993; Herman
&
Uyeyama,
1999).
In the
study reported
by
Herman
et
al.
(1993),
the
experimenter
sat on a
surfboard
at
the
approximate center
of the
dolphin's circular tank.
The
dolphin, Phoenix,
was
stationed next
to the
surfboard. Three
objects were positioned around
the
perimeter
of the
tank
at
radial distances
of
approximately
7.6
m
from
the
experi-
menter, forming
the
vertices
of an
equilateral triangle.
The
experimenter faced
one of the
vertices
and
pointed with
an
extended
arm and
hand
to one of the
objects, followed
quickly
by a
symbolic action gesture directing Phoenix
to
take that action
to the
indicated object. Phoenix correctly
responded
to the
indicated object
on 17 of 21
trials (81%,
p
<
.001, cumulative binomial test; chance probabil-
ity
=
.33).
In
further
probe tests,
the
experimenter pointed
to
a
location
at
which
an
object
was
normally positioned
but no
object
was
present.
For all
three probes given, Phoenix
responded
by
beginning
to
swim
in the
indicated direction
but
then quickly turned
and
swam
to one of the
remaining
objects, carrying
out the
requested action
to it.
These results
suggested that Phoenix responded
to the
manual pointing
gesture
as a
reference
to an
object
in the
indicated direction,
rather than simply interpreting
the
point
as an
instruction
to
swim
in the
indicated direction.
In
the
study reported
by
Herman
and
Uyeyama (1999),
Phoenix
was
shown
to be
capable
of
interpreting
a
sequence
of
pointing gestures. Three objects were arranged
to the
left
of,
to the right of, and
behind Phoenix
as she
faced
the
experimenter,
who was
positioned outside
of the
tank
and
immediately
in
front
of
Phoenix.
The
experimenter pointed
directly toward
one
object, signed
the
gestural
symbol
fetch,
and
then pointed directly toward
a
second object,
using
a
fully
extended
arm and
index
finger in
each case. This
sequence, point
+
fetch
+
point,
was
consistent with
the
linear grammar used
in the
symbolic acoustic language
taught
to
Phoenix (Herman, Richards,
&
Wolz,
1984)
and
requested
that
she
bring
the first
referenced object
to the
side
of
the
second referenced object. Over
the 18
trials given,
representing
three replications
of the six
possible unique
pointing sequences
of
this type (point
left
+
fetch
+
point
right,
point
left
+
fetch
+
point behind, etc.), Phoenix
re-
sponded wholly correctly
to 9 by
taking
the first
indicated
object
to the
second indicated object. Inasmuch
as the
probability
of a
correct response
to two
points
by
chance
alone
is 1 out of 6
(1A
X
y2),
Phoenix's level
of
performance
was
significant
(p < .01 by
cumulative binomial test).
In
the
present, more complete studies,
we
tested
for
referential
understanding
by the
dolphin
of
human pointing
by
using both simple
and
complex forms
of
pointing.
We
examined
the
dolphin's understanding
of
references
to
distant
objects
and to
objects behind
it
(out
of its
immediate
field
of
view).
We
used
not
only direct points
to
objects
but
also cross-body points. Additionally,
and
importantly,
we
tested
for an
understanding
of
complex sequences
of
points,
including sequences
of
direct points, sequences
of
cross-
body points, combinations
of
direct
and
cross-body points,
and
combinations
of
each
of
these forms
of
pointing with
symbolic gestural references
to
objects,
the
latter
as de-
scribed
by
Herman
et al.
(1984).
An
ability
to
respond
accurately
to
these various indicative forms would strongly
implicate referential understanding,
in the
same
sense
that
accurate responding
to
wholly symbolic
forms
can
imply
referential
understanding (Herman
et
al.,
1993).
Experiment
1:
Initial
Test
of
Pointing
Comprehension
and
Development
of
Final
Stimulus
Configuration
In
this
first
experiment, carried
out in two
parts,
we
compared
the
dolphin's understanding
of
human references
to
objects when made either
by
directly pointing
at an
object
or
by
referring
to it
through
a
symbolic gesture (Herman
et
al.,
1984). Objects were positioned
to the
left
of, to the right
of,
and
behind
the
dolphin,
and the
effects
of the
distance
of
the
objects
from
the
human pointer (and, therefore,
from
the
dolphin) were studied.
Part
A
Method
Subjects
and
background.
The
subjects were
Akeakamai
(Ake)
and
Phoenix,
two
adult female
bottlenosed
dolphins
(Tursiops
truncates).
Ake
served
as the
subject
in
this
and all
subsequent
experiments. Phoenix served
as the
subject
in
only
one
part
of
this
experiment.
Ake and
Phoenix were housed
together
with
two
other
bottlenosed dolphins
(Elele
and
Hiapo)
in two
interconnected
outdoor circular seawater tanks, each 15.2
m in
diameter
and 2.0 m
in
depth, located
at the
Kewalo Basin Marine Mammal Laboratory
in
Honolulu, Hawaii.
The
walls
of
these tanks
rose
approximately
1.2
m
above
the
surrounding deck.
To
interface
closely
with
the
dolphin, individuals
stood
on
0.6-m-high
platforms outside
of the
tank wall.
The
upper half
of
their body
was
thus visible
to the
dolphin.
All
testing took place
in one of the
tanks.
Both dolphins understood
symbolic
gestural
references
to ac-
tions. However, whereas
Ake
also understood symbolic gestural
references
to
objects, Phoenix's understanding
was
limited
to
acoustic references. These differences resulted
from
the
different
specializations
of
each dolphin:
Ake was
trained
in a
gesturally
expressed language,
and
Phoenix
was
trained
in an
acoustically
expressed language (Herman
et
al.,
1984).
Additionally, both
dolphins were familiar with
sequences
consisting
of a
symbolic
reference
to an
object (expressed through gestures
for Ake and
through
sounds
for
Phoenix) followed
by a
symbolic
reference to
an
action (expressed gesturally
for Ake and
either gesturally
or
acoustically
for
Phoenix).
Also, both dolphins have been exposed informally
to the
pointing gestures
of
humans over many years. This exposure
was
similar
in
principle
to the
informal exposure
to
human pointing
given
to the
chimpanzees studied
by
Povinelli
et al.
(1997).
Our
trainers
can ask a
dolphin
to
bring
a floating
object
to
them
by
pointing
in the
direction
of the
object
and
then giving
a
single
gestural sign
glossed
as
"fetch."
The
dolphin then
reliably retrieves
the
indicated object, although errors
may
occur
if the
indicated
object
is
relatively close
to
another object
or
lies
in the
path
of
another object.
Stimulus
objects.
Four objects having symbolic names
in
Ake's
gestural
language were used. These were
a
rectangular-shaped

350
HERMAN
ET
AL.
plastic laundry basket
(basket)
measuring
56 cm on its
longest side,
a
foam
"boogie"
board
(surfboard)
107 cm
long,
an
81-cm
square
plastic
hoop
(hoop),
and a
117-cm
length
of a
1.9-cm-diameter
plastic
pipe
(pipe).
Gestural references
to
these objects were
arbitrary
(noniconic)
in
appearance and, hence, symbolic. Through-
out the
study, only three
of the
four
stimulus objects were used
at a
time.
The
general plan
was to
create
a
triangular array
of
three
objects, such that
one
would
be to the
dolphin's
left,
another
to her
right,
and
the
third behind her. These orientations were relative
to
the
dolphin's
position
at the
tank wall
as she
faced
the
experi-
menter.
The
experimenter could then
refer
to any of
these objects
by
using either
a
pointing gesture
or a
symbolic gesture.
General
procedure.
An
individual, referred
to
here
as the
experimenter,2
stood
on the
platform adjacent
to the
tank wall.
The
dolphin,
in
turn,
was
stationed
in a
vertical position
next
to the
tank
wall,
with
her
head
out of the
water
and
facing
the
experimenter.
The
experimenter
provided
all
gestures,
pointing
or
symbolic,
following
a
preplanned schedule
and
under
the
control
of a
session
supervisor located
in an
observation tower
at the
side
of the
tank
opposite
to the
experimenter's location.
Initially,
we
placed
the
three objects
at a
considerable distance
from
the
dolphin,
as
shown
in
Figure
1
(Positions
LI ,
Bl,
and
Rl,
indicating
the
left,
behind,
and right
positions, respectively).
The
left
and right
objects were arrayed
along
the
tank perimeter, each
at
approximately
8.7-m
linear distance
from
the
dolphin's position.
The
third object, located behind
the
dolphin,
was at
approximately
14.1-m
linear distance
and
also
was
positioned along
the
perimeter
3.1m
3.1
m
Elevated
Tower
Figure
1.
Schematic diagram showing
the
arrangements
of
objects
to the
left
(L) of, to the right (R) of, and
behind
(B) the
dolphin
(D).
The
arrangement during
the first
part
of
Experiment
1
is
shown
as
LI,
Bl, and
Rl;
during
the
latter part
of
Experiment
1
and
for all
subsequent experiments,
the
arrangement
was L2, B2,
and
R2. The
location
of the
experimenter
(E)
also
is
shown,
as is
the
observation tower where
the
session supervisor
and the
naive
observer
who
labeled
the
dolphin's responses were located.
of
the
tank.
The
objects were held
in
place
by
assistants,
who
released them only when
the
trial began
and
then crouched down
behind
the
tank wall,
out of
sight
of the
dolphin. This procedure
ensured
that
the
objects remained close
to
their starting positions
throughout
the
trial.
Six
testing sessions were given.
The
particular
object
at
each
of the
three locations
was
maintained within
a
session
but was
changed across sessions. Within each
session,
each
object
was
referenced equally
often
across trials
by
pointing
directly (Pd)
at it
(either two, three,
or
four
times),
followed
immediately
by a
gesturally
given
action
(A)
command,
to
form
a
Pd
+ A
sequence. Also,
the
same number
of
object references were
made
through symbolic
(S)
gestures,
with
each object gesture
followed
immediately
by an
action
gesture,
to
form
an S
-I-
A
sequence. These
two
types
of
trials, indicative
and
symbolic, were
embedded among
a
variety
of
simple behavior
trials.
Simple
behavior trials (fillers)
are
behaviors performed without reference
to any
object, such
as
back dive,
spiral
swim,
and
twisting jump.
Each
of
these behaviors
is
elicited
by a
single unique gesture.
Simple
behavior trials
are
almost always carried
out
correctly
and
were
used here
to
increase
the
behavioral variation
of the
session
and
to
increase
the
overall level
of
success during
the
session.
An
assistant located behind
the
experimenter maintained
the
session plan
and
before each trial whispered trial directions
to the
experimenter,
but out of
earshot
of the
assistants
holding
the
side
objects.
The
session supervisor waited
for the
experimenter
to
give
a
"pay attention" gesture
to the
dolphin
(by
raising
one
arm,
elbow
bent, index
finger
extended)
and
then called
out
"ready,"
to
signal
that
signing should begin. This ready call also signaled
the
assistants
to
release their objects
and to
step down
from
the
side
of
the
tank
and out of
view
of the
dolphin.
On
hearing
the
"ready"
call,
the
experimenter
ceased
the
pay-attention
sign,
by
dropping
his right
hand
to his
side,
and
after
a
pause
of
approximately
2-3 s
began
the
gestural sequence. Direct
points
to the
left
and right
objects were made with
the arm on the
same side
as the
object.
The arm was
extended
left
or right at
shoulder height,
with
the
index
finger
extended. Points
to the
object
behind
the
dolphin were made
with
the arm
extended forward. Half
of
the
points
to the
object behind were made with
the
left
hand
and
half
with
the right
hand.
After
completion
of the
study,
the
duration
of
the
experimenter's points
was
measured through
a
sampling
of
the
videotaped trials. Points averaged less than
1 s in
duration.
Gestural symbols (gestural names
of
objects
and
names
of
actions)
were
as
described
by
Herman
et
al.
(1984). During
the
trials,
the
experimenter wore dark glasses,
to
guard against
any
eye-gaze
cues
that
might help
the
dolphin orient,
and
looked straight ahead
and
downward.
An
observer located
in the
tower,
and
having
no
knowledge
of
which
signs were given
to the
dolphin, categorized
the
dolphin's
response
by
calling
out the
object
or
objects responded
to and the
type
of
action given, using
Ake's
language system.
If
this
was
indeed
the
instruction given
to
Ake,
the
supervisor called
out
"yes,"
and the
experimenter blew
a
whistle
and
gave
fish and
social
praise
to the
dolphin.
If the
dolphin's response
was
incorrect,
the
supervisor instructed
the
experimenter
to
call
the
dolphin back
to
the
station. Reinforcement
was
omitted.
All
sessions
were video-
taped
to
provide
an
archival record
of
each trial.
The
first 2
testing sessions consisted
of 16
trials each,
the
next
three
of 32
trials
each,
and the
last
of 34
trials.
The
number
of Pd +
A
trials
and S + A
trials were balanced within each session:
6
each
for
Sessions
1 and 2 and 12
each
for
Sessions
3-6.
The
remaining
trials
at
each session consisted
of fillers.
2
The
same individual (Louis
M.
Herman)
was
used
as the
experimenter
throughout these studies.

DOLPHINS
COMPREHEND
HUMAN
POINTING
351
Results
and
Discussion
Figure
2
summarizes
the
results
for
pointing references
and for
symbolic
references
to
objects
for
each
of the
three
object
locations.
These
results
are for
object errors only
(there were only
five
action errors
in the 120
trials given
and
only
two
errors
in 42
filler
trials).
Ake
successfully
re-
sponded
to the two
side objects when referenced either
by
pointing
or
symbolically. However,
she had
difficulty
in
responding reliably
to the
object behind
her
when
it was
referenced
by a
point
but not
when
it was
referenced
symbolically.
All
performance levels, except that
to the
object behind
her
when
it was
referenced through pointing,
were significantly above chance
(p <
.002, cumulative
binomial test; chance probability
=
.33). Accuracy
in re-
sponding
to the
object behind when referenced symbolically
significantly
exceeded
accuracy when referenced
by
point-
ing,
x2(l,
N = 40) =
5.01,
p <
.05.
No
other
differences
between pointing
and
symbolic references
to a
given loca-
tion
were significant. Thus,
Ake
showed
a
good understand-
ing
of
pointing references
to
laterally displaced objects that
were
in her
visual
field (a
dolphin's eyes
are
laterally
located)
but did not
appear
to
grasp reliably
the
significance
of
a
point
to an
object behind her.
In
the
present
case,
in
response
to a
pointing
gesture
to the
object behind her,
Ake
often
began
to
swim toward that
object
but
then abruptly changed direction
to
choose instead
one of the
side objects.
It was as if she
defined
the
arena
in
which
an
object could
be
referenced
by a
pointing gesture
as
limited
to
something
less
than
the
full
diameter
of her
tank.
This
was not the
case
for
symbolically referenced objects,
however,
as was
shown
by her
high performance levels
no
matter which object
was
referred
to
symbolically. Clearly,
then,
her
major
difficulty
was in
understanding
a
pointing
gesture that referred
to an
object behind her.
Behind
Right
Figure
2.
Experiment
1:
The
dolphin
Ake's
accuracy
in
respond-
ing
to
direct-pointing
references
and to
symbolic
references
to
objects
to her
left
(8.7
m),
to her right
(8.7
m),
or
behind
her
(14.1
m).
The
number
of
trials
is
given
in
each bar.
The
horizontal line
indicates chance performance level (33%).
Pd =
direct
point;
A =
action
command;
S =
symbolic
gesture.
PartB
To
examine whether
the
results
for Ake
were limited
to
her or
were more
general,
we
tested
the
dolphin
Phoenix
in
the
same configuration using
the
same procedures
as in
Part
A,
except that
no
trials were given symbolically referencing
objects.
As
previously noted, Phoenix
was
unfamiliar with
symbolic gestural references
to
objects. Phoenix, too,
had
difficulty
with
the
object behind her, although
she
performed
significantly
(p <
.01)
above chance levels, responding
correctly
on 12 of 20
(60%) pointing trials.
In
contrast,
Phoenix
made only
a
single error over
the 19
pointing trials
referencing
the two
side objects. Thus,
the
effect
of the
distant
displacement
of the
object behind
was
similar
for
both dolphins, although somewhat less troublesome
for
Phoenix
man
for
Ake.
We
next tested Phoenix with
a
closer array
of
objects,
using
the
second configuration shown
in
Figure
1
(L2,
B2,
and
R2). Laterally displaced objects were
now
positioned
at
a
distance
of
3.1
m and the
object behind
at
approximately
2.7 m.
Over
the six
sessions with this
new
configuration,
Phoenix responded correctly
to the
object behind
her on 19
of
24
(79%) pointing trials
(p <
.001)
and
made
no
errors
at
all
to the
object behind
her on the
last
two of
these
six
sessions. Performance
on the
side objects remained high.
There
were
only
three
errors
in the 48
trials referencing
side
objects.
Phoenix then continued
as the
subject
to
complete
the
study with
her
described earlier (Herman
&
Uyeyama,
1999).
After
the
study with Phoenix
was
completed,
we
resumed studies
with
Ake,
as
reported
in the
following
experiments. Throughout these
new
studies,
we
used only
the
second stimulus
configuration
of
Figure
1
(L2,
B2,
and
R2).
Experiment
2:
Direct Pointing Gestures
in
a New
Configuration
The
results
of
Experiment
1
revealed that both dolphins
responded highly reliably
to
laterally displaced objects
referred
to by
direct pointing.
The
distance
to
these objects
(8.7
m in
Part
A) was
well beyond
the
extended
finger of the
experimenter. However, both dolphins
had
difficulty
in
responding reliably
to
pointing references
to the
object
behind them
(14.1-m
distant
in
Part
A),
although Phoenix
(but
not
Ake) performed
significantly
above chance. Phoe-
nix's performance improved
further
when
the
objects were
moved closer
(3.1
m for
lateral objects
and 2.7 m for the
object
behind)
but
still
distally
located well beyond
the
experimenter's extended
arm and finger.
In
Experiment
2, we
resumed testing
Ake in the
closer
configuration
used with Phoenix (L2,
B2, and R2 in
Figure
1). We
continued
to
make
use of
Ake's
knowledge
of
symbolic gestural references
to
objects
to
make comparisons
with
pointing references and,
in
subsequent additional tests,
to use
combinations
of
pointing
and
symbolic references
to
objects.
In
Part
A, we
tested Ake's understanding
of Pd + A
and
S + A
sequences.
In
Part
B, we
introduced exaggerated
pointing
as a
technique
to
help improve Ake's response
accuracy
to the
object behind her.
It was of
interest,
of

352
HERMAN
ET
AL.
course,
to see
whether
the
difficulty
in
responding
to
points
to
objects
behind
her
could
be
overcome.
Part
A
A
period
of 19
days
had
elapsed
since
completion
of the
work
with
Ake
described
in
Experiment
1.
The new
testing
with
her
began
immediately
thereafter.
Ake was
given
no
prior
exposure
or
training
in the
second
configuration
shown
in
Figure
1
before beginning testing.
Method
As in
Experiment
1,
assistants held
the
objects
in
place
to the
left
and
the
right
of the
dolphin until
the
trial began
and
then released
them
and
crouched down behind
the
tank wall.
The
experimenter
positioned
and
held
the
object behind
the
dolphin
in
place
by
using
a
long boat
pole.
He
then
released
the
object before initiating
the
pay-attention sign.
If the
object behind drifted
out of
position
before
signing began, because
of
wind conditions,
the
experimenter
repositioned
it.
Four sessions were given using
the
basket, hoop,
and
surfboard.
The
arrangement
of the
three objects
was
different
across sessions
but
remained
the
same within
a
session. Each session consisted
of
12
Pd + A
trials,
6 S + A
trials,
and 6 filler
trials. Each
of the
three
objects
was
referred
to
four
times using pointing gestures
and two
times using symbolic gestures.
A
session began with
3 S + A
trials,
1 to
each object, followed
by a filler
trial.
After
this warm-up,
the
remaining
trial types were randomly arranged, with
the
constraint
that there would
be no
more than
2 Pd
-I-
A
trials
in a
row.
In
addition
to
these preplanned trials,
11
supplementary
Pd + A
trials
were given:
6 at the end of
Session
1,
3 at
the
end of
Session
2,
none
at
Session
3, and 2 at the end of
Session
4.
These additional trials
were intended principally
to
provide
further
data
on the
accuracy
of
responses when
the
experimenter pointed
to
objects behind
the
dolphin.
All
pointing gestures were
the
same
as
those described
in
Experiment
1.
Results
and
Discussion
Figure
3
A
summarizes
the findings for the Pd + A and the
S + A
sequences.
As in
Experiment
1,
action errors
were
rare
(three errors
in 83
trials)
and
were
not
considered
in the
data
shown
here. Also, errors
on filler
trials
were
again
rare
(only
one
error
in 24
trials).
Responses
to
pointing
references
to the
left
and right
objects
yielded
only
two
errors each,
and
symbolic
refer-
ences
to
these
objects
yielded
only
a
single error. These
differences
were
not
significant,
x2(l »
N
= 52) =
0.30,
p >
.05.
In
contrast,
points
to the
object
behind
yielded
only
48%
correct
responses
(p >
.05,
cumulative
binomial
test;
chance
probability
=
.33),
whereas
symbolic
references
were
with-
out
error.
The
difference
between
the two
types
of
references
to
objects behind
was
significant,
\2(1,
N = 31) =
6.81,
p <
.01.
This
finding of
substantial
error
to the
object
behind
when
referenced
by a
point
is
consistent
with
the
results
for
Ake in
Part
A of
Experiment
1 and
also
with
the
relatively
greater
error
of
Phoenix
when
responding
to the
object
behind
as
compared
with
objects
placed
to
either side
of her
(Experiment
1,
Part
B).
Clearly,
Ake
understood
a
symbolic
Left
Behind
Right
Figure
3.
Experiment
2:
Ake's
percentage
of
correct responses
to
different
sequence types.
The
number
of
trials
is
given
in
each bar.
The
horizontal
lines
indicate
chance
performance
level
(33%).
A:
Accuracy
in
responding
to
direct-pointing references
and to
symbolic references
to
objects
to the
left
(3.1
m)
of, to the right
(3.1
m)
of, or
behind (2.7
m) Ake in
Part
A.
Results
for
direct-pointing
references
were before
the use of
exaggerated pointing.
B:
Accu-
racy
in
responding
to
direct-pointing references
and to
cross-point
sequences
in
Part
B.
Exaggerated pointing
was
used
for
direct-
pointing
gestures during some sessions
and
trials. Note that there
were
no
cross-point
sequences
when
the
object
was
behind Ake.
Pd
=
direct point;
A =
action command;
S =
symbolic gesture;
Px
=
cross-point.
reference
to the
object
behind
her.
Her
difficulty
in
under-
standing
a
pointing
reference
to an
object
behind
her may
mean
that
points
to
objects
out of her field of
view
are not
easily
understood,
as is the
case
for
human
infants
younger
than
12
months
(e.g.,
Butterworth
&
Grover,
1988).
At the
same
time,
unlike
the
case
for
chimpanzees,
the
dolphin's
difficulty
did not
seem
to lie in the
distal
displacement
of
objects,
inasmuch
as the
left
and right
objects
were
even
farther
away
than
the
object behind
and all
were well
away
from
the
extended
index
finger. A
closer
examination
of the
object
location
incorrectly
chosen
by Ake
after
points
to the
object
behind
her
revealed
no
left-right
bias
(five
errors
were
to the
left
distractor
object
and six
errors
were
to the
right
distractor
object),
but 8 of the
11
incorrect
choices
were
to the
surfboard—the
largest,
most
visible
object.

DOLPHINS
COMPREHEND
HUMAN
POINTING
353
PartB
We
attempted
to
improve
Ake's
responding
to the
object
behind
her by
using exaggerated pointing.
The
experimenter
leaned forward when pointing
at the
object behind
Ake and
leaned
to the
side when pointing
at the
laterally displaced
objects.
The
exaggerated motion
was
intended
to
help orient
Ake
toward
the
indicated object. Although this exaggerated
motion
was not
necessary
for the
side objects, which were
already responded
to
highly reliably,
it was
used
to
preclude
the
dolphin
from
using body motion alone
as a cue to
respond
to the
object behind.
Additionally, cross-pointing (Px)
was
introduced
for the
first
time.
For a
cross-point,
the
hand opposite
to a
side
object
was
brought across
the
front
of the
body
in a
pointing
motion,
at
about chest height,
and was
held
for
less than
2 s.
The
index
finger
was
extended
and
reached slightly beyond
the
opposite shoulder line. Deliberate exaggerations were
not
used with cross-points, although
at
times there
was a
slight
natural rotational motion
of the
body produced
as the
arm
swept across
the
body.
The
experimenter, however,
made
efforts
to
minimize
or
eliminate
any
rotational motion.
Each cross-point,
as for the
direct points,
was
immedi-
ately followed
by an
action gesture,
to
form
the
sequence
Px
+ A.
There
was no
pretraining
for
cross-points. Cross-
points
had not
been
used previously with
Ake (or
Phoenix)
and
were introduced
to
test
for
spontaneity
of
understanding
of
this
new
type
of
deictic gesture. Cross-points also served
as
a
good control test
to
ensure that
the
dolphin's accuracy
in
responding
to
direct points
to
left
or
right objects
was not a
function
of
associating
the
experimenter's
left
or
right arms
with
the
object appearing
to
that side.
Method
Other
than
the use of
exaggerated
gestures
for
some
of the
direct-pointing
trials,
the
procedure
was the
same
as
that
used
in
Part
A.
There
were
five
sessions
altogether.
No S + A
trials
were
given.
The first 2
sessions
consisted
of
direct
points
only,
for a
total
of
32
trials,
plus
11
filler
trials.
The
remaining
three
sessions
(31
trials
each
for
Sessions
3 and 4 and 16
trials
for
Session
5)
contained
both
direct
points
and
cross-body
points
interleaved
among
simple
behavior
trials.
Cross-body
points
thus
occurred
for
the
first
time
in the
dolphin's
experience
on
Session
3.
During
Session
3 and the
remaining
two
sessions,
cross-body
points
were
given
in low
numbers
to
avoid
any
negative
emotional
responding
should
frequent
errors
have
occurred.
Cross-body
points,
of
course,
can
reference
only
left
and right
objects,
whereas
direct
points
can
reference
all
three
locations—left,
right, and
behind.
For
direct
points,
there
were
approximately
twice
as
many
references
to the
object
behind
as to
either
the
left
or right
objects.
For
direct
points,
the use and
degree
of
exaggeration
was
gradually
reduced
over
the first 3
sessions.
Also,
there
were
trials
when
no
exaggeration
was
used,
interspersed
among
trials
having
only
a
slightly
exaggerated
motion. Beginning with
Session
4,
exaggeration
was no
longer
used.
Results
and
Discussion
As
in
Experiment
1 and in
Experiment
2,
Part
A,
errors
to
actions
in
these
sequences were rare (only three errors
in 92
trials).
Also, there were only
two
errors
in 31 filler
trials.
Figure
3B
summarizes object selection performance
on
Pd
+ A and Px + A
sequences.
For
direct points, responses
to the
object behind were much more accurate than they
were
in
Part
A,
with only
five
errors occurring
in 42
trials,
X2(l,
N = 42) =
56.70,
p <
.001,
including
no
errors during
the
last
two
sessions
(18
trials), both given without exagger-
ated pointing. Responding
to the
left
and right
points
continued
to
remain very accurate, yielding only
one
error
in
37
trials.
Inasmuch
as
cross-body points
can be
made only
to
laterally
displaced objects, there
are no
data under
Px + A
trials
in
Figure
3B for the
object behind.
For the
laterally
displaced objects,
Ake
responded correctly
to
11
of the
13
(85%)
cross-points given
(8 to the right and 5 to the
left;
p
<
.001,
cumulative binomial test; chance probabil-
ity
=
.33). Both errors were
a
response
to the
object behind
Ake
when
the
cross-point
was
directed
at the
object
to her
right.
Inasmuch
as
these
13
cross-points were
the
very
first
that
were given,
her
high level
of
performance implies
spontaneous
understanding
of the
referring
function
of the
cross-point. Spontaneity
is
clearly evidenced
by the
data
from
Session
3, the first
session
at
which cross-points were
given.
During this session,
Ake
responded correctly
to all 4
cross-point trials given,
2 to her
left
and 2 to her right
(p
<
.01, cumulative binomial test; chance probabil-
ity
=
.33).
One of
Ake's
two
cross-point errors occurred
at
Session
4 and the
other
at
Session
5.
Overall, then, these data
revealed
an
immediate understanding
of the
referring
func-
tion
of the
cross-point
to
distal objects located
to the
dolphin's
left
and right.
The findings
from
the first two
experiments show that
the
dolphin
understood
the
pointing gestures
of a
human
in a
way
not yet
demonstrated
for
great apes,
in
that
the
dolphin
readily understood pointing references
to
distal objects
laterally placed (cf. Povinelli
et
al.,
1997). Pointing refer-
ences
to the
object behind
the
dolphin proved
difficult
for
her at first.
However, this initial
difficulty
was
overcome
by
bringing
the
object closer (although still
at a
distance
of
approximately
2.7 m) and by
using exaggerated points
involving
moving
the
torso, rather than just
the arm and
hand,
in the
direction
of the
object.
After
the
dolphin
had
some experience with these exaggerated points, reliance
on
them
was no
longer necessary.
Experiment
3:
Relational Points
Within
Ake's learned artificial gestural language (Herman
et
al.,
1984),
she is
familiar with
the
symbolic relational
sequence
SI
+ S2 + R,
where
SI
and S2 are
different
objects referred
to
through symbolic gestures,
and R is
die
relational term fetch. This sequence requires
the
dolphin
to
bring
the
second named object (S2)
to the
side
of the first
named object
(SI).
For
example,
in
response
to the
symbolic
gestural sequence basket
+
ball
+
fetch,
the
dolphin
is
required
to
retrieve
the
ball
floating in her
tank
and
transport
it
to the
side
of the
basket, also
floating in her
tank. This
inverse grammar contrasts with
a
linear (left-to-right) gram-
mar, studied
by
Herman
et al.
(1984) with
the
dolphin

354
HERMAN
ET
AL.
Phoenix,
in
which
the
same instruction
was
given
as
ball
+
fetch
+
basket.
The
inverse grammar
was
developed
to
dissociate
the
order
in
which gestural symbols appear
from
the
order
in
which responses
to
those symbols must
be
executed. Thus,
in the
inverse grammar,
the
destination
object
is
named
first but is the
last object
to
which
the
dolphin must respond.
The
transport object
is
named second
but
is the first
object
to be
acquired.
The
type
of
relationship
required
(e.g.,
fetch
=
place
beside
vs. in =
place
inside
of
or on top of) is
specified only
at the end
of
the
sequence.
Hence, with
the
inverse grammar,
the
dolphin cannot
fully
organize
a
response until
the
entire sequence
has
been
received, precluding
any
simple word-by-word (linear)
processing
of
language items.
As
with
the S + A
sequences,
points
can
replace
the
gestural object names
in
relational
sequences
to
produce
the
following three variants:
(a)
PI
+
P2
+ R, (b)
PI
+ S2 + R, and (c)
SI
+ P2 + R.
Moreover,
either direct points (Pd)
or
cross-body points (Px)
can be
used.
These various conditions were tested
in the
present
experiment.
Would
Ake
understand
a
sequence consisting
of
manual
points
to two
different
objects,
Pdl
+ Pd2 + R or
Pxl
+
Px2
+ R, for
example,
as an
instruction
to
take
the
object
pointed
to
second
to the
object pointed
to first? If the
dolphin
does
referentially
understand
the
manual
pointing
gesture
as
indicating
a
particular
object,
she
should
be
able
to
interpret
faithfully
sequences
of
points that obey
the
grammatical
structure
normally used with sequences
of
symbolic ges-
tures. That
is, she
should
be
able
to
substitute sequences
of
indicative gestures (either
Pd or Px or
combinations
of
these)
for
sequences
of
symbolic gestures, because both
types
of
sequences,
semantic
and
indicative,
are
references
to
given objects.
To
examine
the
level
of
referential understanding
of the
manual
pointing gesture, responses
to
complex relational
sequences, composed
in
whole
or in
part
of
indicative
gestures, were compared with responses
to
wholly symbolic
relational constructions
(SI
+ S2 + R). In
Part
A,
only
direct
points were
used
in
constructing complex indicative
sequences, consisting
of two
direct points
or a
direct point
together with
a
symbolic object gesture.
In
Part
B,
direct
points, cross-body points,
or
both were used
in
complex
indicative relational sequences. Again, there could
be two
pointing
gestures
or a
pointing gesture
in
combination
with
a
symbolic
object gesture.
If Ake
were able
to
correctly
interpret these complex indicative sequences,
it
would
indicate
forcefully
that
she
treats indicative gestures referen-
tially,
as
calling
her
attention
to
specific
objects.
Part
A
We
tested
Ake's
understanding
of
three
different
complex
indicative
relational
sequences:
(a) Pd + S + R, in
which
the
destination object
was
referred
to by a
direct point
and the
transport
object
by a
symbolic gesture;
(b) S + Pd + R, in
which
the
destination object
was
referred
to by a
symbolic
gesture
and the
transport object
by a
direct pointing gesture;
and
(c) Pdl + Pd2 + R, in
which both
the
destination
and
transport
objects were referred
to by
directly pointing
at
them. Note that
an
inverse grammar
was
used,
as was the
case
for
wholly symbolic relational constructions,
SI
+
S2
+ R.
Figure
4
shows
the
dolphin responding successfully
to
the Px + S + R
sequence
"that
[surfboard],
pipe fetch"
(meaning "bring
the
pipe
to the
surfboard").
The
surfboard,
located
to the
dolphin's
right, was
referred
to by a
cross-
point,
and the
pipe, located
to the
dolphin's
left,
was
referred
to
by a
symbolic gesture.
Figure
5
shows
the
dolphin
responding
successfully
to the
Pdl + Pd2 + R
sequence
"that
[hoop], that [surfboard]
fetch" (meaning
"bring
the
surfboard
to the
hoop").
The
hoop
was to the
dolphin's
right and the
surfboard
was to the
left.
Both were referred
to by
direct points.
Method
Ten
sessions, ranging
in
length
from
21 to 36
trials
(M
= 31
trials), were run. Each
session
consisted
of
mixed trials
of
simple
direct-pointing sequences
(Pd + A),
simple cross-pointing
se-
quences
(Px + A),
simple symbolic sequences
(S +
A),
complex
symbolic
sequences
(SI
+ S2 + R),
plus
the
three complex
pointing sequences
described
above
(Pd + S + R, S + Pd + R,
and
Pdl + Pd2 + R).
There were also
filler
trials,
as in the
previous experiments.
There
were
six
possible
unique instructions that could
be
given
for
each
of the
three complex indicative relational
sequences.
The
dolphin could
be
instructed
to
take
the
left
object
to
either
the right
object
or the
object behind,
to
take
the
object behind
to
either
the
left
or the right
object,
or to
take
the right
object
to
either
the
object
behind
or the
left
object.
A
replication consisted
of all six
possible
combinations. During
the
course
of the 10
sessions, three complete
replications were given,
for a
total
of 18
trials
for
each
of the
three
types
of
complex pointing sequences. Also,
a
total
of 69
wholly
symbolic sequences
(SI
+ S2 + R)
were given.
The
other types
of
trials were given
in
various numbers during
the 10
sessions
to
provide variability
in
trial type,
to
avoid expectations
for any
particular trial type,
and to add to the
database
for the
types
of
trials
studied
in
Experiments
1 and 2.
Results
and
Discussion
Table
1
summarizes
the findings for the
wholly symbolic
relational sequences
and for
each type
of
complex indicative
relational sequence. Performance
on all
four
sequence types
was
well above chance levels. Inasmuch
as
there were three
objects
to
choose among,
the
probability
of
choosing
the
correct transport object (the second element
in the
sequence)
was
1 out of 3, and
given that
the
correct transport object
had
been
selected,
the
probability
of
arriving
at the
correct
destination
object (the second element)
was 1 out of 2. The
joint chance probability
of
success
was
therefore
1 out of 6.
All
percentage
correct
values
in
Table
1 are
thus significant
atp
<
.0001,
by the
cumulative binomial
test.
Ake
responded correctly overall
on 72% of the
S1
+ S2 +
R
sequences. Noteworthy, however,
is
that
the
error rate
was
significantly
greater
for
selection
of the
destination object

DOLPHINS
COMPREHEND HUMAN POINTING
355
Figure
4.
Five video
frames
of the Px + S + R
sequence "that
[surfboard]
pipe fetch" (meaning
"bring
the
pipe
to the
surfboard").
A: The
experimenter uses
a
cross-body point
to
refer
to the
surfboard
to the
dolphin's right.
Ake
leans slightly
to her
right.
B: The
experimenter
is
giving
the
symbolic object sign
lor
pipe
as Ake
watches.
C: The
experimenter
is
giving
the
symbolic sign
for
fetch.
Ake is
beginning
her
swim
to her
left
toward
the
pipe.
D: Ake has
arrived
at the
pipe
and is
beginning
to
push
it in a
counterclockwise rotation back toward
the
surfboard.
E: Ake has
deposited
the
pipe
at the
side
of the
surfboard. Note that
the
experimenter
is
wearing
sunglasses
and
also
holds
a
whistle
in his
mouth.
Px =
cross-point;
S =
symbolic gesture;
R =
relational term
(fetch).
(SI)
than
for
selection
of the
transport object (S2),
x2(l >
N
=
138)=
11.74,
p <
.01.
The
greater error rate
to
S1
was
consistent with
earlier
results
for Ake
when
she was
responding
to
SI
+ S2 + R
sequences within
her
inverse
grammar (Herman, 1986; Herman
et
al.,
1984).
In
Table
1,
the
trend
for
errors
to
increase when
SI
was
referred
to
symbolically
is
evident also
in the
form
S + Pd + R, but in
this
case,
the
difference between
S and Pd
errors
was not
significant,
x2(l,
N = 36) =
2.20,
p >
.05.
In
contrast,
referring
to the
destination object
by
pointing
(Pd + S + A
and
Pdl
+ Pd2 + A)
resulted
in a
slightly lower error rate
to
the
destination object than
to the
transport object, although
again
the
differences
were
not
significant.
The
superiority
of
indicative versus symbolic references
to the
destination
object
is
best illustrated
by
comparing
the
forms
S + Pd + R
and
Pd + S + R.
Errors
to the
destination object were
significantly
greater
for
symbolic than
for
indicative refer-
ences,
x2(l,
N = 36) =
5.79,
p <
.05.
At the
same time,
the
difference
in
errors
to the
transport object between these
two
forms
of
referral
was not
significant,
x2(l>
N
= 36) =
0.23,

356
HERMAN
ET
AL.
B
Figure
5.
Five
video
frames
of the
Pdl
+ Pd2 + R
sequence
"that
[hoop]
that [surfboard] fetch"
(meaning
"bring
the
surfboard
to the
hoop").
A: The
experimenter
points
directly
at the
hoop,
located
to
Ake's
right. Ake is
orienting toward
her right. B: The
experimenter
points
directly
at the
surfboard,
located
to
Ake's
left.
Ake now has
turned
rapidly
to her
left.
C: The
experimenter
is
giving
the
symbolic
sign
for
fetch.
Ake is
swimming toward
the
surfboard.
D: Ake is
pushing
the
surfboard
back
to the right. She
will actually
pass
over
the
pipe,
which
was
behind her,
on her way to the
hoop,
displacing
the
pipe
slightly
to the right. E: Ake has
deposited
the
surfboard
at the
hoop
and
slightly
on
top
of it. Pd =
direct
point;
R =
relational
term
(fetch).
p
>
.05.
In the
context
of the
inverse grammar used
for
relational sequences,
the
findings
in
Table
1 can be
inter-
preted
as
underscoring
the
advantages
of
spatial references
to
destination objects over purely symbolic references.
No
such
superiority
was
seen, however,
for
references
to
transport objects. Inasmuch
as the
destination object
was
referred
to first in the
sequence
but was the
last object
arrived
at,
there
was a
requirement
for the
dolphin
to
remember this initial reference while organizing
and
carry-
ing
out a
response
to the
transport object. Hence,
with
spatial
encoding available,
the
dolphin
may
benefit
by
representing
in
memory
not
only
the
object referred
to
(what)
but
also
its
location (where). This also
may
reflect, theoretically,
a
superiority
of
prospective coding over retrospective coding
(Honig
&
Thompson,
1982).
The
lack
of any
advantage
to
spatial encoding
of the
transport object
may
simply
reflect
that
there
is an
almost immediate response
to it, and
thus
memory load
or
retroactive interference
after
receipt
of
transport information
is
minimal
and
predictable
in
content
(it is
always
the
gestural
symbol
fetch).

DOLPHINS COMPREHEND HUMAN POINTING
357
Table
1
Experiment
3,
Part
A:
Correct
Responses
to
Purely
Symbolic
Relational Sequences
and to
Each
of
the
Three
Complex
Indicative
Relational Sequences
Involving
Direct Points
Sequence type
SI
+ S2 + R
S + Pd + R
Pd
+ S + R
Pdl
+ Pd2 + R
No. of
trials
69
18
18
18
No. of
wholly
correct
responses
(%)
50
(72.5)
11(61.1)
16
(88.9)
13
(72.2)
No.
of
correct
responses
per
element
(%)
Destination
object
50
(72.5)
11(61.1)
17
(94.4)
16
(88.9)
Transport
object
65
(94.2)
15
(83.3)
16(88.9)
14
(77.7)
Note.
In
each sequence type,
the first
element
is the
destination object,
and the
second element
is
the
transport object. Elements
in
boldface
are
those referred
to by
pointing.
All
percentages
are
significant
atp
<
.0001
by
the
cumulative binomial
test.
S =
symbolic gesture;
R =
relational term
(fetch);
Pd =
indicative direct point.
Most
important,
the
results illustrate
the
dolphin's
under-
standing
of the
indicative gesture when used
in
these
complex
relational
forms.
The
indicative gesture
was
under-
stood spontaneously whether
it was a
single element
of the
sequence, appearing
as
either
the
destination object
or the
transport
object,
or
whether
it
involved both elements.
The
spontaneity
with
which
the
indicative gesture
was
under-
stood
in
these contexts
can be
evaluated
by
examining
performance
on the first 6
unique
occurrences
of
each
sequence type. These occurrences represent
the six
possible
unique
combinations
of
taking
one of
three objects
to one of
the
remaining
two
(e.g., taking
the
left
object
to the
object
behind,
the
object behind
to the
left
object,
and the
left
object
to the right
object). Table
2
shows that across
the
three
sequence types containing pointing gestures,
the
object
referred
to by
pointing
was
responded
to
correctly
in 22 of
24
initial cases
(first
occurrences).
In
contrast,
for the
three
sequences
containing symbolic gestures,
the
object referred
to
symbolically
was
responded
to
correctly
in
only
15 of 24
cases. This
difference
was
significant,
x2(l,
N
=
48) =
5.78,
p
<
.025. Nonetheless, wholly correct performance (all
elements responded
to
correctly)
was
significantly above
chance levels
for all
four
types. Using
the
cumulative
binomial
test with chance probability equal
to 1 out of 6,
three wholly correct responses
out of six was
significant
at
p
< .05 and
four
and five out of six
were significant
at
p
<
.005
and p <
.0001,
respectively. Again,
as in the
overall results,
the
major benefit derived
from
pointing
at
objects rather than referring
to
them symbolically seemed
to
be the
establishment
of
correct responding
to the
destination
object.
It is
noteworthy that
the
duration
of
pointing
gestures,
as
measured
by the
videotape record,
was no
longer than
the
duration
of
symbolic gestures, averaging
about
1 s in
length.
Figure
6A
summarizes
the
results
for the
simple refer-
ences
to
objects using direct points
(Pd + A),
symbolic
references
(S + A), or
cross-points
(Px + A).
Performance
on
all
sequence types ranged
from
80% to
100%
correct.
As
previously noted, these performance levels
did not
include
errors
to
actions.
There were only
eight
such
errors
on
159
trials. Also, there were
no
errors
on the 28 filler
trials.
Direct-pointing references
to the
object behind were still
Table
2
Experiment
3,
Part
A:
Accuracy
in
Responding
to the
First
6
Unique
Relational
Sequences
for
Purely
Symbolic
Relational Sequences
and
for
Each
of
the
Three
Complex
Indicative
Sequences
Involving
Direct Points
Object
location
Destination Transport
object object
SI
+ S2 + R S + Pd + R Pd + S + R
Pdl
+ Pd2 + R
SI
S2
Pd
Pd
Pdl
Pd2
Right
Right
Behind
Behind
Left
Left
Behind
Left
Right
Left
Behind
Right
0
1
1
1
0"
0
0
1
1
1
0"
0
1
1
1
0
0°
1
1
1
1
1
0"
1
1 1
1 1
1 1
1
0
1 1
1 1
1
1
1
1
0
1
1
1
1
1
1
1
Total correct
Note.
In
each sequence,
the first
element
is the
destination object,
and the
second
element
is the
transport object.
S =
symbolic
gesture;
R =
relational term
(fetch);
Pd =
direct
point;
0 =
incorrect
response;
I =
correct
response.
"The
first and
second
elements
were
reversed.

358
HERMAN
ETAL.
Left
Behind
Right
Left
Behind Right
Figure
6.
Ake's
response
accuracy
in
Experiment
3,
Part
A (A)
and
Experiment
3,
Part
B (B) to
direct-pointing
references (Pd),
symbolic references (S),
and
cross-pointing references (Px)
to
objects
to her
left
(3.1
m),
to her
right
(3.1
m),
or
behind
her
(2.7
m). The
number
of
trials
is
shown
in
each bar.
The
horizontal
lines
indicate chance performance level
(33%).
Note that
there
were
no
Px
+ A
sequences
when
the
object
was
behind
Ake.
A =
action
command.
more
difficult
for Ake
than were references
to
laterally
located objects,
but
performance
was
well above chance
(p
<
.0001,
cumulative binomial test)
and was not
signifi-
cantly
different
from
performance
after
symbolic references
to
the
object behind,
x2(l,
N
= 61) =
1.90,
p >
.05. Also,
the
observed level
of
performance
in
response
to
pointing
references
to the
object behind
was
only slightly below that
observed
in
Part
B of
Experiment
2,
x2(l,
N = 86) =
1.15,
p
>
.05.
PartB
In
this
final
part,
we
tested
for
understanding
of five
different
complex indicative relational sequences involving
cross-body points alone
or in
combination with direct points
or
symbolic gestures. These sequences were
(a) Px + S + R,
in
which
the
destination object
was
referred
to by a
cross-point
and the
transport object
by a
symbolic gesture;
(b)
S + Px + R, in
which
the
destination object
was
referred
to by a
symbolic gesture
and the
transport object
by a
cross-point;
(c) Px + Pd + R, in
which
the
destination object
was
referred
to by a
cross-point
and the
transport object
by a
direct point;
(d) Pd + Px + R, in
which
the
destination
object
was
referred
to by a
direct point
and the
transport
object
by a
cross-point;
and (e)
Pxl
+ Px2 + R, in
which
both
the
destination
and
transport objects were referred
to by
cross-body points. Note that once again
the
inverse grammati-
cal
structure
was
used (Herman
et
al.,
1984). Also, recall that
cross-body points could
be
used only
to
refer
to the
objects
to the
dolphin's
left
and
right
and not the
object behind her.
Direct points,
of
course, could
be
used
to
refer
to any of the
three positions.
The
sequence
Pd + Px + R is
illustrated
in
Figure
7,
showing
the
dolphin successfully responding
to the
instruc-
tion
"that [pipe] that [hoop] fetch."
The
pipe
was
behind
the
dolphin
and was
referred
to by a
direct point;
the
hoop
was to
the
dolphin's right
and was
referred
to by a
cross-body point.
Figure
8
shows
the
dolphin responding correctly
to the
Pxl + Px2 + R
sequence "that [pipe] that [surfboard]
fetch." Both objects were referred
to by
cross-body points.
Method
Eight
sessions
were given
over
a
2-week
period.
Sessions
ranged
from
20 to 25
trials
in
length
(M
= 23
trials).
Sessions
consisted
of
mixed trials
of
simple direct-pointing
sequences
(Pd + A),
simple
cross-pointing
sequences
(Px + A),
simple symbolic
sequences
(S
+ A),
complex symbolic
sequences
(SI
+ S2 + R),
plus each
of
the five
types
of
complex pointing
sequences
described
above
(Types
a
through
e).
There
were
also
filler
trials,
as in the
previous
experiments.
There
were four
possible
unique instructions that could
be
given
for
each
of the
four complex indicative
relational
sequences
involving
a
cross-point
in
combination with either
a
direct
point
or
a
symbolic reference (Types
a, b, c, and d,
above).
For
example,
for
Pd
+ Px + R, the
dolphin could
be
instructed
to
take
the
left
object,
referred
to by a
cross-point,
to
either
the right
object
or the
object
behind, both referred
to by a
direct
point.
Or, the
dolphin could
be
instructed
to
take
the right
object
(cross-point)
to
either
the
left
object
or the
object behind (direct points).
Two
replications
(8
trials)
of
each
of the
four types were given over
the
eight
sessions.
At
each
session,
there were either
one or two
exemplars
of
each
type given.
There
were only
two
possible unique combinations
for the
type
Pxl + Px2 + R:
Take
the
left
object
to the right
object
or
take
the
right
object
to the
left
object, both referred
to
only
by
cross-points.
Two
complete
replications
of
this type
(4
trials) were given.
One
exemplar
of
each type
was
given
at
four
of the
eight
sessions
(Sessions
1,3,
5,
and
8).
Also given during
the
eight
sessions
were
50
SI
+ S2 + R
sequences
(6 per
session
except
for one
session
of 8), 24 S + A
sequences
(3 per
session),
32 Pd + A
sequences
(4 per
session),
and
16
Px + A
sequences
(2 per
session). Finally,
26 filler
trials,
2 to 4
per
session, were given.
All
other procedures were
the
same
as
those used
in
Part
A.
Results
and
Discussion
Table
3
gives
the
results
for the
wholly symbolic complex
relational sequence
(SI
+ S2 + R) and for the
remaining
five
different
types
of
complex indicative relational
se-
quences,
all
involving cross-body points. Performance
on
the
wholly symbolic relational sequence (82%
wholly

DOLPHINS
COMPREHEND
HUMAN
POINTING
359
D
Figure
7.
Five video
frames
of the Pd + Px + R
sequence "that [pipe] that [hoop] fetch" (meaning
"bring
the
hoop
to the
pipe").
A: The
experimenter uses
a
direct point
to
refer
to the
pipe behind
Ake.
Ake
begins
a
clockwise rotation toward
the
pipe.
B: The
experimenter uses
a
cross-body point
to
refer
to the
hoop
to
Ake's
right. Ake is now
beginning
to
move
to her right. C: The
experimenter
is
giving
the
symbolic sign
for
fetch.
Ake
continues
to her right. D: Ake has
arrived
at the
hoop
and is
beginning
to
push
it
toward
the
pipe.
E: Ake has
deposited
the
hoop
at the
side
of the
pipe.
Pd =
direct point;
Px =
cross-point;
R =
relational term
(fetch).
correct responses)
was
improved over that
in
Part
A
(72%
correct), although
the
difference
was not
significant,
x2( l >
N
=
119)
=
1.46,
p >
.05.
The
trend
for
errors
to be
greater
to the
destination object than
to the
transport object
was
still
significant,
however,
x2(l,
N =
100)
=
7.11,
p <
.01.
As
we
found
in
Part
A, the
greater error rate
to the
destination object
was no
longer present when
it was
referred
to by an
indicative gesture, either
a
direct point
or a
cross-point. Indeed, errors
to
each sequence element were
few
and
about equal
in
number.
The
overall performance
level
on
each
of the
complex indicative relational types,
except
for the
last type listed
in
Table
3, was
well above
chance
(p <
.005, cumulative binomial test).
For the
last
type,
Pxl
+ Px2 + R,
only
4
trials were given (two
replications),
too few to
allow
for a
confident assessment
of
performance
ability. Recall that with
the
cross-body point,
one
could refer only
to
objects
to the
left
and right of the
dolphin,
not to the
object behind.
In
Table
4, we
analyzed
the
spontaneity
of
understanding
these
new
complex indicative relational forms
by
examining
performance
on the 1st
unique trials
of
each type. There
were
4
unique trials
for all
types shown, except
for the
last,

360
HERMAN
ET AL.
(meaning
"bring
the
surfboard
to the
pipe").
A: The
experimenter
uses
a
cross-body
point
to
refer
to
the
pipe
to the
dolphin's
left.
Ake
leans
slightly
to her
left.
B: The
experimenter
uses
a
second
cross-body
point
to
refer
to the
surfboard
to the
dolphin's
right. Ake
watches.
C: The
experimenter
is
giving
the
symbolic
sign
tor
fetch.
Ake is
beginning
to
swim
to her right
toward
the
surfboard.
D: Ake
has
arrived
at the
surfboard
and is
beginning
to
push
it
toward
the
left
to the
pipe.
E: Ake has
deposited
the
surfboard
at the
pipe.
Px =
cross-point;
R =
relational
term
(fetch).
for
which there were only
2.
Spontaneity
of
understanding
may
be
inferred
from
the
high performance levels
on
these
initial
trials:
14
wholly correct responses
out of the
initial
18
unique
trials
(p <
.001, cumulative binomial
test).
In
fact,
there
was
virtually
no
difference
in
performance
on
these
18
initial trials
as
compared with
the
subsequent
18
trials.
The
high level
of
immediate performance
on the first
replication
and
the
maintenance
of
that level during
the
second replica-
tion
argue that learning
was not a
factor governing
the
dolphin's understanding
of
these complex indicative rela-
tional forms.
Initial performance
on the
cross-body points contained
within
these complex
forms
was
impressive, with only
one
error occurring
in the 20
unique initial cross-point trials
across
the five
sequence types shown
in
Table
4.
Thus,
the
findings are
clear that
the
dolphin
was
able
to
incorporate
cross-body points into
an
understanding
of
these complex
indicative relational forms, whether
the
cross-points were
used
alone
or
occurred
in
conjunction with symbolic
or
direct-pointing references
to
objects.
Figure
6B
shows results
for the
simple sequences
Pd + A,
S + A, and Px + A
given
in
this part
of the
experiment.
All

DOLPHINS
COMPREHEND
HUMAN
POINTING
361
Table
3
Experiment
3,
Part
B:
Correct Responses
to
Purely
Symbolic
Relational Sequences
and to
Each
of
the
Three
Complex
Indicative Relational Sequences Involving Cross-Points,
Direct
Points,
or
Both
Sequence type
SI
+ S2 + R
S + Px + R
Px
+ S + R
Pd
+ Px + R
Px
+ Pd + R
Pxl
+ Px2 + R
No. of
trials
50
8
8
8
8
4
No. of
wholly
correct
responses
(%)
41
(82.0)
5
(62.5)
8(100)
6
(75.0)
6
(75.0)
2
(50.0)
No.
of
correct responses
per
element
(%)
Destination
object
41
(82.0)
5(61.1)
8(100)
6
(83.3)
7
(87.5)
3
(75.0)
Transport
object
49
(98.0)
6
(83.3)
8(100)
7
(87.5)
6
(83.3)
2
(50.0)
Note.
In
each sequence type,
the first
element
is the
destination object,
and the
second element
is
the
transport object. Elements
in
boldface
are
those referred
to by
pointing.
S =
symbolic gesture;
R
=
relational term
(fetch);
Px =
indicative cross-point;
Pd =
indicative direct point.
performance levels were significant
at p < .02
(cumulative
binomial test; chance probability
=
.33). However,
re-
sponses
to
direct points
to the
object behind were
signifi-
cantly
below
the
level observed
for
symbolic references,
X2(l,
N = 49) =
5.80,
p <
.025, although
not
significantly
different
from
that
observed
in
Part
A,
x2(l,
N = 60) =
1.82,
p
>
.05. Again, action
errors
were rare (one error
in 72
trials)
and
were
not
accounted
for in
Figure
6.
Also, there
was
only
one
error
in 28 filler
trials.
General
Discussion
What
did
Ake
understand about
the
human manual
pointing gesture? Clearly,
in the
studies reported here,
she
treated
the
pointing
gesture
as a
reference
to an
object
in her
habitat. Most telling
in
this regard
was her
spontaneous
incorporation
of the
pointing gesture into
the
framework
of
the
inverse grammar
of her
learned symbolic gestural
receptive language (Herman
et
al,
1984). Within that
language, abstract (arbitrary) gestures referred
to
objects
in
her
habitat. When pointing references
to
objects were
substituted
for
symbolic references,
Ake
responded
in the
same manner
and
about
as
efficiently
as she did to
sequences
composed
of
wholly symbolic gestures. That these pointing
gestures were spontaneously treated
similarly
to the
abstract
symbolic gestures suggests that
the
points, like
the
symbolic
gestures, were treated
as
abstract references
to
objects.
No
comparable level
of
referential understanding
of
human
manual pointing
has
been demonstrated convincingly
for
apes. Povinelli
et al.
(1997) concluded that
the
adolescent
chimpanzees they studied failed
to
understand that
the
pointing gesture
of the
human "referred
to (or was
about)
a
particular object
or
location
in
space"
(p.
455).
A
key
aspect
of the
disclaimer
of
Povinelli
et al.
(1997)
was
the
failure
of
their chimpanzees
to
understand pointing
references
unless
the
human's
finger was
almost touching
the
object. Povinelli
et al.
(1997) also noted
the
same
Table
4
Experiment
3,
Part
B:
Accuracy
in
Taking
the
Transport
Object
(T) to the
Destination
Object
(D) on the
First
Trial
of
Each Sequence
Type
Shown
as a
Function
of
the
Initial
Location
of
Each
Object
to the
Left
of,
to the
Right
of,
or
Behind
the
Dolphin
Object's initial
location
S + Px + R Px + S + R Pd + Px + R Px + Pd + R Pxl + Px2 + R
D
Right
Right
Behind
Behind
Left
Left
T
Behind
Left
Right
Left
Behind
Right
D
—
1
1
0
—
1
T
1
1
1
1
D
1
1
—
—
1
1
T
1
1
1
1
D
—
1
1
0
—
1
T
1
1
1
1
D
1
1
—
—
Oa
1
T
0
1
0"
1
D
—
1
—
—
—
1
T
1
1
Total correct
Note.
Dashes indicate that
the
sequence type could
not be
given
if a
cross-point
was to be
used
and
the
object
was
behind
the
dolphin.
S =
symbolic gesture;
Px =
cross-point;
R =
relational term
(fetch);
Pd =
direct point;
1 =
correct response;
0 =
incorrect response.
aThe
first and
second elements were reversed.

362
HERMAN
ET
AL.
limitation
in
understanding
of
distal points
in
other studies
using
great
ape
species
(Call
&
Tomasello, 1994;
Menzel,
1974; Povinelli
et
al.,
1992;
see
also Tomasello
&
Camaioni,
1997).
Human infants younger than 12-15 months show
similar limitations
in
their understanding
of
points
to
distant
objects
(Butterworth,
1991;
Butterworth
&
Graver,
1988,
1990;
Desrochers
et
al.,
1995; Lempers, 1979;
Morissette
et
al.,
1995).
In
contrast
to
these
findings
with apes
and
human
infants,
both
of the
dolphins studied
in the
present experiments
immediately responded correctly
to
laterally displaced
ob-
jects, even though their distances were well away
from
the
signer
and the
dolphins,
8.7
m
in
Experiment
1 and 3.1 m in
subsequent experiments. Spontaneous understanding
of in-
dicative gestures toward laterally displaced objects occurred
in
response
to not
only direct points
but
also
the
cross-body
points
given
to
Ake
that were never previously experienced
by
her.
The
human pointing gesture
functions
as a
one-term
spatial deictic system (Anderson
&
Keenan,
1985)
in
that
there
is no
separate gesture
for
near objects versus
far
objects.
In
spoken English,
in
contrast,
the
demonstrative
pronouns
this
and
that
function
as a
two-term deictic system
to
identify,
at
least
grossly, relative distance
from
the
speaker.
In
this sense, pointing
for the
dolphin
had
broader
application than
it did for the
apes
or the
human
infants
in
that
it was
understood
as
referring
to
objects relatively
far
laterally
as
well
as
relatively near.
Ake had
initial
difficulty
in
responding reliably
to
points
to
objects behind her. Nonetheless,
the
introduction
of
exaggerated motions, characterized
by
leaning
the
body
toward
the
object behind while also extending
the arm
forward,
succeeded
in
more reliably orienting
Ake to
that
object.
After
some experience
with
this method
of
referral,
further
use of
exaggeration
to
indicate
the
object behind
was
no
longer necessary
to
yield performance levels well above
chance
to
objects behind, although still
not at the
level
observed
for
references
to
laterally displaced objects.
The
dolphin Phoenix, tested
in
Experiment
1
with
the
same
configuration
of
objects that
was
used with Ake, also
had
relatively
greater
difficulty
in
responding
to the
object
behind
her
than
to
laterally displaced objects, although
unlike
Ake, Phoenix responded immediately
at
above chance
levels
to the
object behind her. Infants
at 12
months
of age
can
follow
an
adult's manual pointing gesture
to an
object
within
their
field of
view,
but at
that age, they still
are
unable
to
understand
a
pointing reference
to an
object behind them
(Butterworth
&
Grover, 1988). Even
at 18
months,
infants
do not
search
for
objects behind them unless there
are no
targets
in
their immediate
field of
view (Butterworth
&
Grover, 1988). Clearly,
a
reference
to an
object that
is out of
sight
is a
more
difficult
conceptual task than
is a
reference
to
an
object that
is
within
the field of
view
and is
seemingly
linked
to
cognitive development (Morissette
et
al.,
1995).
In
the
case
of Ake and
Phoenix, both
of
whom were adult
dolphins, there was,
of
course,
no
question
of a
developmen-
tal
limitation
but
rather only
a
transient conceptual limitation
that,
for
Ake,
was
largely overcome through some simple
training
techniques.
The
strongest support
for
granting
to Ake
referential
understanding
of the
human's pointing gesture comes
from
the
results with more complex pointing sequences, consist-
ing
of
combinations
of
direct
and
cross-body points
as
well
as
combinations
of
each
of
these with symbolically refer-
enced objects.
To our
knowledge, spontaneity
in
understand-
ing
comparable sequential points
has not
been investigated
in
other animals
or in
infants. Consider
the
implicit task
within
these complex forms,
all
requiring
Ake to
transport
one
indicated object
to
another indicated object. Inasmuch
as
an
inverse grammar governs these transport instructions,
the
dolphin must retain
in
memory
a
representation
of the
first
object
referred
to
(the destination object) while
process-
ing the
gestural indication
for the
second object referred
to
(the transport object) and,
finally,
transporting that second
referenced object
to the first.
Also consider that
the
gestures,
whether
indicative
or
symbolic,
are
highly transitory physi-
cally, lasting
on the
order
of 1-2 s.
Hence, there
is no
permanent
finger-pointing
"trail"
for the
dolphin
to
follow,
unlike
the
case
for the
chimpanzees
or
infants tested
by
Povinelli
et al.
(1997),
for
whom
the
pointing gesture toward
an
object remained
in
place until
a
response
was
made.
Furthermore, although
Ake
(and Phoenix)
had
prior informal
exposure
to
simple direct points, both dolphins spontane-
ously
responded
to
sequences
of
direct points
at
above
chance levels (for Phoenix,
see
Herman
&
Uyeyama,
1999).
The
spontaneity
of
these
responses
of the
dolphins suggests
that
they
had
developed
a
generalized conceptual understand-
ing
of
pointing
as an
abstract reference
to
objects.
Ake
responded accurately
to
cross-body pointing during
its
initial
use as
well
as
thereafter, within both
the
simple
and
the
complex sequences.
It is
clear, therefore, that
her
understanding
of the
cross-point
was not
dependent
on
specific
learning
or on
specific prior experiences with
cross-points. Rather,
it
appears that
Ake's
broad understand-
ing
of
direct points
as
object references readily generalized
to
related physical components
of the
cross-body point, most
likely
the
direction
of
movement
of the arm and
hand (e.g.,
toward
the
left
for
both direct
and
cross-body points when
referring
to an
object
to the
left).
Furthermore,
the
spontane-
ous
comprehension
of
cross-points argues that, during
the
prior direct-pointing tests,
Ake was not
simply responding
to
learned associations between
a
particular
arm and the
left
or
right
side
at
which
an
object
was
located.
That
is,
when
referring
to the
left
object
by a
direct point,
the
left
arm was
used,
but
when referring
to it by a
cross-body point,
the right
arm
was
used.
Additional evidence that
Ake
understood
the
require-
ments
of
these complex indicative tasks comes
from
her
ancillary behaviors.
For
example,
after
receiving
a
relational
instruction (e.g.,
Pdl
+ Pd2 + R), Ake
deposited
the
trans-
port object (Pd2) next
to the
destination object (Pdl)
and
then,
typically, moved slightly away
from
the
pair
with
her
head
out of the
water, apparently seeking confirmation
from
the
experimenter
by
waiting
for a
whistle sound, hand
clapping,
or
even excited vocalizations. This feedback
of
information
from
the
experimenter,
in
turn, engendered
excited vocalizations
from
the
dolphin. Hence, these tasks
involved
a
large social component between
the
human
and
the
dolphin.
The
basis
for
these
skills
of Ake
(and Phoenix)

DOLPHINS
COMPREHEND
HUMAN
POINTING
363
may
reflect,
in
part,
the
long history
of
informal
use of
pointing
by
trainers when asking
a
dolphin
to
bring them
an
object.
Often,
objects pointed
to by
trainers might
be
located
distantly (e.g.,
from
3 to 15
m).
Thus,
the
dolphins could
be
said
to
have
had
prior
experience
with
distal indicative
references. Nonetheless,
a
comparable prolonged history
of
informal
exposure
to
pointing, including pointing
to
distal
objects,
did not
help
the
chimpanzees studied
by
Povinelli
et
al.
(1997)
to
referentially
understand manual pointing when
they
subsequently were tested
in a
formal situation. Overall,
these chimpanzees failed
to
respond correctly when
the
distance between
the
experimenter's
finger
and the
object
was
anything more than almost touching.
No
comparable
constraint occurred
for
Ake
or
Phoenix,
in
that laterally
displaced
objects
at
approximately
3-m or
9-m
distances
were responded
to
reliably
from
the
beginning.
Clearly,
Ake and
Phoenix
can be
considered
"encultur-
ated,"
in
that they have
had a
long history
of
daily
interactions
with
humans. They also have been taught
a
language system that includes extensive symbolic references
to
objects.
The
facility
of
these dolphins
in
comprehending
human
manual pointing
may
largely
reflect
these experi-
ences.
Again, however, comparable experiences
with
hu-
mans
and
with language systems
by the
orangutan
Chantek
(Miles, 1990)
did not
result
in
substantial comprehension
of
human
manual pointing
to
objects (Call
&
Tomasello, 1994;
Povinelli
et
al.,
1997; Tomasello
&
Camaioni,
1997).
The
same
is
true
for the
language-trained common chimpanzees
studied
by
Savage-Rumbaugh
(1986),
which
also
showed
no
spontaneous understanding
of
human manual pointing.
Alternatively,
as
suggested
by
Herman
et al.
(1993),
attending
to
another's distal interrogation
of an
object
through sound (i.e., through echolocation)
may be a
normal
part
of the
dolphin's behavior
in its
natural world. This
natural
trait
may
then
be
generalized
to
other types
of
functionally
equivalent distal referencing.
The
sonar emis-
sions
of
dolphins
are
projected
in a
relatively narrow beam,
with
the
longitudinal axis
of the
body oriented
in the
direction
of the
beam
and
with
occasional scanning head
movements
(Au, 1993).
The
whole body
of the
dolphin
as
well
as the
emitted sonar
field are
thus,
in a
sense, pointing
toward
an
object.
An
observing dolphin, therefore,
may
gain
some sense
of
where another dolphin
is
searching with
its
sonar
beam,
or
even what
it is
"looking"
at.
Recently,
Xitco
and
Roitblat
(1996)
showed that
an
observing dolphin that
was
itself
not
echolocating could, through passive listening
alone,
identify
the
target being interrogated
by an
echolocat-
ing
dolphin positioned nearby. Also, recent research
at our
laboratory
has
shown that echolocation
may in
fact
yield
a
representation
of the
shape
or the
appearance
of the
target
object
(Herman, Pack,
&
Hoffmann-Kuhnt,
1998; Pack
&
Herman, 1995).
If
echolocation information
can be
shared,
in
the
least
by
drawing
the
attention
of
another dolphin
to a
distal target
of
interest,
it may
provide
an
underlying
basis
for
the
ready conceptual understanding
of the
human
pointing reference
to
distal objects. Spanning such
a
wide
phylogenetic
gap in
social communication,
from
human
to
dolphin,
may be
possible because
of the
dolphin's demon-
strated ability
to
understand
a
variety
of
abstract relation-
ships
(see, e.g., Herman, 1986; Herman
et
al.,
1993)
and
because
it is
prepared
by its own
biology
to
engage
in
complex
social
transactions
(Pryor
&
Norris,
1991).
Finally, although
the
dolphin understands very well what
to do in
response
to the
points,
in an
imperative sense,
and
also understands
the
outcome contingencies,
we
cannot
yet
conclude that
the
dolphin views
the
task
as one of
joint
attention. Does
the
dolphin understand that
the
experimenter
is
himself aware
of the
object? Additional studies
are
necessary
to
test
for
indicants
of an
appreciation
by the
dolphin that attention
is
shared with
the
human
or for
other
features
that might bear
on a
theory
of
mind.
Likewise,
the
ability
of the
dolphin
to use a
pointing
gesture (e.g.,
with
its
rostrum
and
body
or
with
its
echoloca-
tion beam)
to
intentionally
direct
the
attention
of
another
to a
location
or an
object
is an
open question. Although
encultur-
ated chimpanzees have
not
been shown
to
easily referen-
tially understand human manual pointing (Povinelli
et
al.,
1997), they,
as
well
as
chimpanzees
not
exposed
to
language
training
but
having
had
extensive human contact (Leavens
et
al.,
1996), apparently
can
learn
to use
manual pointing
to
direct attention
of
humans
to
distal objects
or
locations (see
also Greenfield
&
Savage-Rumbaugh, 1990; Miles, 1990;
Woodruff
&
Premack,
1979).
As
noted
by
several research-
ers
(e.g., Call
&
Tomasello, 1994; Herman, 1987; Herman
&
Morrel-Samuels,
1990; Savage-Rumbaugh, 1986),
the
rela-
tionship
between comprehension
and
production
is
complex,
and
the
presence
of one
does
not
necessarily imply compe-
tency
in the
other. Therefore,
a
study
of the
dolphin's ability
to use
indicative signals
to
direct
the
attention
of a
human
or
another dolphin
is of
considerable comparative interest,
given
its
comprehension
of the
referential character
of the
human
manual pointing gesture.
References
Anderson,
S. R., &
Keenan,
E. L.
(1985).
Deixis.
In T.
Shopen
(Ed.),
Language
typology
and
syntactic description (Vol.
3, pp.
259-308).
New
York:
Cambridge
University
Press.
Au, W. W. L.
(1993).
The
sonar
of
dolphins.
New
York:
Springer-
Verlag.
Baldwin,
D.
A.,
&
Moses,
L. J.
(1994).
Early
understanding
of
referential intent
and
attentional
focus:
Evidence
from
language
and
emotion.
In C.
Lewis
& P.
Mitchell
(Eds.),
Children's early
understanding
of
mind (pp.
133-156).
Hillsdale,
NJ:
Erlbaum.
Butterworth,
G. E.
(1991).
The
ontogeny
and
phylogeny
of
joint
visual
attention.
In A.
Whiten
(Ed.),
Natural theories
of
mind:
Evolution, development
and
simulation
of
everyday mind read-
ing
(pp.
223-232).
Oxford, England:
Basil
Blackwell.
Butterworth,
G.
E.,
&
Grover,
L.
(1988).
The
origins
of
referential
communication
in
human infancy.
In L.
Weiskrantz
(Ed.),
Thought without language (pp.
5-25).
Oxford,
England:
Oxford
University
Press.
Butterworth,
G.
E.,
&
Grover,
L.
(1990).
Joint visual
attention,
manual
pointing
and
preverbal
communication
in
human
in-
fancy.
In M.
Jeannerod
(Ed.),
Attention
and
performance
XIII
(pp.
605-624).
Hillsdale,
NJ:
Erlbaum.
Call,
J.,
&
Tomasello,
M.
(1994).
Production
and
comprehension
of
referential
pointing
by
orangutans
(Pongo
pygmaeus).
Journal
of
Comparative
Psychology,
108,
307-317.
Desrochers,
S.,
Morissette,
P., &
Ricard,
M.
(1995).
Two
perspec-

364
HERMAN
ET
AL.
lives
on
pointing
in
infancy.
In C.
Moore
& P. J.
Dunham
(Eds.),
Joint
attention:
Its
origins
and
roles
in
development
(pp.
85-101).
Hillsdale,
NJ:
Erlbaum.
Goodall,
J.
(1986).
The
chimpanzees
of
Gombe:
Patterns
of
behavior.
Cambridge,
MA:
Belknap
Press.
Greenfield,
P. M., &
Savage-Rumbaugh,
E. S.
(1990).
Grammatical
combination
in Pan
paniscus:
Processes
of
learning
and
inven-
tion
in the
evolution
and
development
of
language.
In S. T.
Parker
& K. R.
Gibson
(Eds.),
"Language"
and
intelligence
in
monkeys
and
apes
(pp.
540-578).
New
York:
Cambridge Univer-
sity Press.
Herman,
L. M.
(1986). Cognition
and
language competencies
of
bottlenosed
dolphins.
In R. J.
Schusterman,
J.
Thomas,
& F. G.
Wood
(Eds.),
Dolphin
cognition
and
behavior:
A
comparative
approach
(pp.
221-251).
Hillsdale,
NJ:
Erlbaum.
Herman,
L. M.
(1987).
Receptive competencies
of
language trained
animals.
In J. S.
Rosenblatt,
C.
Beer,
M. C.
Busnel,
& P. J. B.
Slater
(Eds.),
Advances
in the
study
of
behavior (Vol.
17, pp.
1-60).
Petaluma,
CA:
Academic Press.
Herman,
L.
M.,
&
Morrel-Samuels,
P.
(1990).
Knowledge
acquisi-
tion
and
asymmetries between language comprehension
and
production: Dolphins
and
apes
as a
general model
for
animals.
In
M.
Bekoff
& D.
Jamieson
(Eds.),
Interpretation
and
explanation
in the
study
of
behavior.
Vol.
I:
Interpretation,
intentionality,
and
communication
(pp.
283-312).
Boulder,
CO:
Westview Press.
Herman,
L.
M.,
Pack,
A.
A.,
&
Hoffmann-Kuhnt,
M.
(1998).
Seeing through sound: Dolphins
(Tursiops
truncatus)
perceive
the
spatial structure
of
objects through
echolocation.
Journal
of
Comparative
Psychology,
112,
292-305.
Herman,
L.
M.,
Pack,
A.
A.,
&
Morrel-Samuels,
P.
(1993).
Representational
and
conceptual skills
of
dolphins.
In H. L.
Roitblat,
L. M.
Herman,
& P. E.
Nachtigall
(Eds.),
Language
and
communication:
Comparative perspectives (pp.
273-298).
Hillsdale,
NJ:
Erlbaum.
Herman,
L.
M.,
Richards,
D.
G.,
&
Wolz,
J. P.
(1984). Comprehen-
sion
of
sentences
by
bottlenosed dolphins. Cognition,
16,
129-219.
Herman,
L.
M.,
&
Uyeyama,
R. K.
(1999).
The
dolphin's
grammati-
cal
competency: Comments
on
Kako
(1999). Animal Learning
and
Behavior,
27,
18-23.
Honig,
W.
K.,
&
Thompson,
R. K. R.
(1982). Retrospective
and
prospective processing
in
animal working memory.
In G. H.
Bower
(Ed.),
The
psychology
of
learning
and
motivation (Vol.
16,
pp.
239-273).
New
York: Academic
Press.
Leavens,
D.
A.,
Hopkins,
W.
D.,
&
Bard,
K. A.
(1996). Indexical
and
referential pointing
in
chimpanzees (Pan
troglodytes).
Journal
of
Comparative
Psychology,
110,
346-353.
Lempers,
J. D.
(1979).
Young
children's production
and
comprehen-
sion
of
nonverbal deictic behaviors.
Journal
of
Genetic
Psychol-
ogy,
135,
93-102.
Menzel,
E.
W.,
Jr.
(1974).
A
group
of
chimpanzees
in a
one-acre
field.
In
A.
Schrier
& F.
Stollnitz
(Eds.),
Behavior
ofnonhuman
primates:
Modern
research
trends
(Vol.
5, pp.
83-153).
New
York:
Academic Press.
Miles,
H. L. W.
(1990).
The
cognitive foundations
for
reference
in a
signing orangutan.
In S. T.
Parker
& K. R.
Gibson
(Eds.),
"Language"
and
intelligence
in
monkeys
and
apes
(pp.
511-
539).
New
York: Cambridge University Press.
Morissette,
P.,
Ricard,
M.,
&
Decarie,
T. G.
(1995).
Joint visual
attention
and
pointing
in
infancy:
A
longitudinal study
of
comprehension. British
Journal
of
Developmental
Psychology,
13,
163-175.
Pack,
A.
A.,
&
Herman,
L. M.
(1995). Sensory integration
in the
bottlenosed dolphin: Immediate recognition
of
complex shapes
across
the
senses
of
echolocation
and
vision.
Journal
of the
Acoustical
Society
of
America,
98,
722-733.
Povinelli,
D.
J.,
&
Davis,
D. R.
(1994).
Differences
between
chimpanzees (Pan
troglodytes)
and
humans
(Homo
sapiens)
in
the
resting state
of the
index
finger:
Implications
for
pointing.
Journal
of
Comparative
Psychology,
108,
134—139.
Povinelli,
D.
J.,
Nelson,
K.
E.,
&
Boysen,
S. T.
(1992). Comprehen-
sion
of
role
reversal
by
chimpanzees: Evidence
of
empathy?
Animal
Behaviour,
43,
633-640.
Povinelli,
D.
J.,
&
Preuss,
T. M.
(1995). Theory
of
mind:
Evolutionary
history
of a
cognitive specialization.
Trends
in
Neurosciences,
18,
418-424.
Povinelli,
D.
J.,
Reaux,
J.
E.;
Bierschwale,
D.
T.,
Allain,
A.
D.,
&
Simon,
B. B.
(1997).
Exploitation
of
pointing
as a
referential
gesture
in
young
children,
but not
adolescent chimpanzees.
Cognitive
Development,
12,
423-461.
Premack,
D.
(1984).
Pedagogy
and
aesthetics
as
sources
of
culture.
In
M. S.
Gazzaniga
(Ed.),
Handbook
of
cognitive
neuroscience
(pp.
15-36).
New
York: Plenum Press.
Pryor,
K.,
&
Norris,
K. S.
(1991).
Dolphin
societies: Discoveries
and
puzzles.
Los
Angeles: University
of
California
Press.
Savage-Rumbaugh,
E. S.
(1986).
Ape
language:
From
conditioned
response
to
symbol.
New
York:
Columbia University Press.
Tomasello,
M.,
&
Call,
J.
(1994).
The
social cognition
of
monkeys
and
apes.
Yearbook
of
Physical
Anthropology,
37,
273-305.
Tomasello,
M.,
&
Camaioni,
L.
(1997).
A
comparison
of the
gestural communication
of
apes
and
human
infants.
Human
Development,
40,
7-24.
Woodruff,
G.,
&
Premack,
D.
(1979). Intentional communication
in
the
chimpanzee:
The
development
of
deception.
Cognition,
7,
333-362.
Xitco,
M.
J.,
Jr.,
&
Roitblat,
H. L.
(1996). Object recognition
through
eavesdropping: Passive echolocation
in
bottlenose
dol-
phins. Animal
Learning
and
Behavior,
24,
355-365.
Received
November
11,1998
Revision received March
19,
1999
Accepted
March
26,
1999