Transcript Linguistics 001, Fall 2004 - University of Pennsylvania

Animal Communication
Ling 001, Spring 2009
Starting Points
• The idea that humans are biologically
equipped for language…
• How does “general intelligence”
(however you would measure that)
correlate with linguistic ability?
• As we’ve seen at different points,
intelligence and linguistic ability are
dissociable in humans.
Starting Points, 2
• The same kind of question can be raised
when we look across species
• E.g., does general intelligence correlate with
different abilities in different species?
• In particular, do animals have “more
language” to the extent that they are
generally smarter than other animals, with
humans being at the top of the chain?
Basic Answer
• The basic answer is: No.
• Human language has syntax: it makes infinite
use of finite means. This does not occur in
the communication systems of other animals
• In some domains, e.g. association a “sound”
with a “meaning”, there might be some effects
that correlate with general intelligence. In
most cases, though, communication systems
are tailored to very specific needs
• There are some surprises; e.g. bees (not very
sophisticated in terms of general intelligence)
have real skills in communication
Two approaches (roughly)
• Animal Communication: The study of what
different species do under “normal”
circumstances
• Instructing Animals: The attempt to teach
(aspects of) human language to a non-human
species
• We’ll see a nice analogy in the reading,
where the former type of study is more like
anthropology, the latter like missionary
work…
Questions
• Animal Communication studies in
general ask things like
– What is being communicated?
– How does this relate to different properties
in other systems? In human language?
– What are the adaptive functions of
communication in a particular species?
– How does the communicative system
relate to the environment, etc.
Questions, 2
• The teach human language to animals view
is less clear in many ways. Assumptions:
– Maybe animals don’t have language because they
don’t happen to have invented it?
– Maybe language is a facet of general intelligence,
so smarter animals should get “more” language
than animals that have lower general intelligence
• Definitely better for headlines/PR. Cp. “bees
show displaced reference property” vs. “Koko
expresses grief about dead kitten”.
Reference (Signal Set)
• Animal communication systems have a limited signal
set: the inventory of things that are communicated
about is very limited (e.g. food, sex, aggression,
predators/threats)
• Human language, recall, has the ability to incorporate
new basic elements (typically words) in a way that is
unlike what is found in animals
• Moreover there is nothing like syntax for generating
new ‘utterances’ in the animal kingdom (in some
domains linear order is important, however, e.g.
birdsong)
• We’ll illustrate by looking at what animal calls ‘refer’
to
Example: Stickleback
• Male changes color (becomes
red)
• Swims in “zigzag” dance
toward egg-laden female
• Female approaches
• Male heads towards his nest
• Female follows
• Male pokes head in nest
entrance
• Female enters, eventually lays
eggs, which are fertilized
Introductory comments
• Something is being communicated in the
stickleback interaction
• In the broadest sense, the male
coloration/dance indicate the readiness to
mate
• But there’s no need to attribute “thinking” to
the participants
• It could be a stimulus/response chain
• I.e., each action elicits a stimulus response
from the other participant
General Question: Reference
• Consider a simple example of communication: dog
posture. In this case, the dog is communicating that it
is ready to play (the following is a game)
Reference, II
• The dog in the example above is communicating
something, namely, something about its internal state
• With reference, we mean roughly what it is that a
signal in communication refers to
• Most animal communication systems are limited in
what their signals are capable of referring to;
typically, an internal state, or an immediately present
stimulus
Basics: Displays
• Returning to our first example…
• Display: “…a conspicuous stereotyped
movement performed in a special context
with an apparent communicative function”
• Example: Stickleback zig-zag dance
– “normally occurring” behaviors evolve into
exaggerated forms
– Displays of this type are less ambiguous and more
likely to provoke a response of the appropriate
type
Basics: Functions
• Views on the evolutionary roles of
communication
– Mutual benefit: signaling systems exist
because both the signaler and receiver
benefit
• E.g. dog displays, avoiding fights is “more
efficient”
• Sticklebacks: producing healthy, fertilized eggs
is “good”
Functions, Cont.
• The “Selfish” view: animal communication
systems evolve because animals benefit from
manipulating one another
– Signals are evolved to increase the sender’s
fitness (not the receiver’s)
– A conspicuous display that attracts predators
wouldn’t be selected for simply because it benefits
others
– Example: “Deception”; males attract mates not by
being better, but simply by appearing to be
healthier
Aspects of Communication
• We’ll concentrate on what is transmitted, although
how speaker/hearer effects are modeled will enter
into this
• I.e., focus on information transmission with the intent
of informing or changing the cognitive state of
receivers.
• This gives us various things to look for in animal
systems
– Appropriate receiver response when stimulus is absent?
– Communication suited to the audience?
– Receivers generalizing properties of the caller?
Functional Reference
• One property of interest is called
functional reference; two parts
– A signal is given in the presence of an
appropriate stimulus; it is not just an
indication of ‘general excitement’
– Conspecifics hearing the signal (receivers)
behave consistently and appropriately,
even in the absence of the stimulus
triggering the signal.
An Example: The Chicken
• Research using a variety closer to the
ancestral type
• Useful test case because
– Chickens court/produce alarm calls in
controlled conditions (lab)
– Chickens respond to video/audio tape of
predators or other chickens, so that
controlled experiments can be set up
Alarms
• Two kinds of alarm calls:
– Aerial predators: shriek/whistle
– Ground predators: series of pulses
• Behavior to different types is matches as well
(cover, looking at sky vs. standing and
searching the ground level)
• What about functioning like reference? How
do signalers/receivers react?
Responses
• Chickens seeing hawk/raccoon stimulus
(simulated) behave appropriately
• Importantly for “functional reference”,
audience chickens show behavior appropriate
to the call
• The latter was tested with hens isolated from
the stimulus viewed by the calling chicken; all
they perceived was the alarm; they behaved
according to the alarm type.
Audience Effects
• Alarm calling has a cost for a single
animal: it may attract attention of a
predator
• It might therefore be expected that there
is an audience effect in such systems:
aspects of the alarm calling depend on
the receivers of the signal
Manipulating Audience Effect
• Do chickens behave as if they intend to
communicate to other chickens?
– Roosters alarm call more when they see a live or
video-taped hen
– It’s not just any audience, it’s an audience of
conspecifics (same species); the alarm call when
other chickens are present is more than if e.g. a
quail is present
– This is interesting but it doesn’t mean that
communication in chickens is intentional. Rather, it
shows that both predator and audience play a role
in some aspect of signalling
Some further examples
• Two different communication systems;
bees and vervet monkeys
• Vervet Monkeys: different calls for
different predators, which trigger
appropriate behavior
• Bees: Communicate distance and
location (relative to the hive) of food
sources
Vervet Monkeys
•
(multimedia from Hauser’s webpage,
http://www.wjh.harvard.edu/~mnkylab/media/vervetcalls.html)
• Snake Alarm: Hearers stand on hind legs
and look on the ground
• Leopard Alarm: Hearers run to the top of the
nearest tree (where leopards can’t go)
• Eagle Alarm: Vervets run under a bush/cover
of tree branches
Function
• The organization of the alarm system makes
sense given the living conditions of the
vervets
• The predators are categorized in different
ways: in particular, ways that require different
responses
• Categorization is acquired in the first four
years of life; young vervets miscategorize,
but eventually learn the correct association of
calls with predators (without, apparently, any
correction)
Vervets: Details
• Further question: what kind of
information is represented in the
receiver?
• (In particular…) Is something about the
nature of the alarm caller represented
by the listening vervets? I.e., do vervets
show awareness of properties of the
individual who is calling?
Outline
• Habituation: Behavioral response
decreases with repeated exposure to
the same stimulus
• Two additional calls (intergroup):
– Wrrr: Approach of another group
– Chutter: Approach of another group, with a
more aggressive aspect
Set up
• Vervets look toward a calling vervet
• The situation can be filmed, and then the
amount of time a listener spends looking at a
“caller” can be measured
• Recorded calls from one vervet; then
– Baseline response
– Eight habituating calls (30 min. apart)
– Test call (30 min. later)
Tests
Question : Would an unreliable wrrr-er be
treated as unreliable for chuttering?
• The test involves two cases:
– One in which the baseline/test are identical to the
habituating call:
A
A A
chut. Wrr chut.
– Another in which the habituating call is from a
different vervet:
A
B
A
chut. Wrr chut.
Call Results (Habituation)
• With habituation, the response goes down
• The question is then for the last (“test”) call, is
the habituation retained?
– Habituation does not transfer from one individual’s
wrrr’s to another’s chutters
– Habituation DOES transfer from the same
individual’s wrrr’s to that individual’s chutters
• This seems to suggest that the unreliability of
the individual can be extracted from the
context independent of the specific call
That is…
• Habituation transferred between two
calls from the same individual
• It appears that the vervets hearing the
habituation sequence learned that the
individual making the call was
“unreliable”
• What is learned is not “call specific”…
General Comments
• What do the different alarm calls mean?
• Note that there are several ways in which to interpret
what the different calls might ‘mean’, if we approach it
that way
– ‘Leopard!’
– ‘Head for the trees!’
– Etc.
• Even in systems of this type, which show some
complexity because of the categorizations involved,
we still seem to have calls made in the presence of
some stimulus
Displacement
• What the reading refers to as situational
freedom or displacement involves
referring to things that are not
immediately present
• This is an obvious property of human
language; but with limited exceptions,
this is clearly not the norm in animal
communication
Bee Dancing
• Honeybees forage for food sources and,
upon returning from a successful trip (i.e.
after finding a food source) they perform a
dance
• This dance is called the waggle dance
because it involves this particular motion
– The waggle dance conveys aspects of the journey the bee
has completed– in particular, the location and distance of the
food source
– Other bees that have witnessed the dance then go to the
food source (or a lot do, in any case)
The dance
• The dance proceeds
in a figure 8 pattern
• The orientation of the
dance with respect to
either vertical or the
position of a light
indicates the
direction (other
factors indicate
approximate
distance)
Orientation
• Dances
oriented
directly to
vertical
indicate that
food is in line
with the sun
• Otherwise, the
orientation of
the dance
indicates the
angle of the
food from the
sun
Competing Hypotheses
• One possibility with the bee dance is that it
does not actually encode information
• I.e. other factors, such as scent (which bees
also use) were thought to be responsible for
the fact that bees could find sources after a
dance
• An experiment designed to test this
hypothesis versus the communication one
involves making a bee ‘lie’ to others
Further facts
• The experiment involves a further fact about
the bee dance
– When it is dark in the hive, the dance is oriented
towards vertical
– When there is a light source visible in the hive, the
dance is oriented towards the light source
• This provides the basis for a way to test what
the bees witnessing a dance are actually
doing
Making bees lie
• Bees detect overall light with their ocelli, a set
of photoreceptors on top of the head
• Experiments can paint the ocelli of bees, so
that bees with painted ocelli behave as if it is
dark, whereas bees with unpainted ocelli act
as if it is light
• The experiment involves bees with painted
ocelli dancing about the location of a food
source; the dance is witnessed by bees with
unpainted ocelli
Ocelli
The idea
• Dancing Bees: The dancing bees orient with
respect to vertical
• Witnessing Bees: The witnessing bees
interpret the dance with respect to a light
source, which is not at vertical
Result: Witnessing bees (the majority) went to the
source that was indicated by the dance, not the
source actually visited by the dancing bees (who
were giving false information)
This is consistent with the communication
hypothesis, but not the odor-alone hypothesis.
Human language in nonhumans
• Another aspect of research on language in
animals involves a different methodology than
studying communications systems that are
natural to non-human species
• This involves trying to teach non-human
animals (typically chimpanzees, bonobos, or
e.g. gorillas in the case of Koko) human
language
On the methodology
• To some extent, studies of this type capture
the popular imagination, often concentrating
on the question of whether animals have ‘true
language’
• The bulkpack reading quotes a paper by
Snowdon (1993) with a perspective on this:
“…ethologists who study natural communication differ
from psychologists who try to teach human language
to other species in the same way as anthropologists
differ from missionaries. Anthropologists try to
understand the natives whereas missionaries try to
civilize them.”
Justification
• One view (Rumbaugh/Savage-Rumbaugh),
discussed in reading: results of language
teaching to primates must be relevant to
humans because chipmanzees and humans
are clsoe (they are our closest relatives).
• Counterpoint (Pinker and others): “closest
relative” has no important status; if
chimpanzees disappeared, some other
animal would be our “closest relativ”
Some facets of the program
• In most of these cases, the primates
were taught some version of sign
language, such as with Washoe
• In a few cases, the primates were
taught visual symbols
• There are some differences as well in
the training situation (home,
laboratory, etc.)
Constraints
• Using sign language (e.g. ASL) is practical:
chimpanzees don’t have the vocal system
etc. of humans
• In some early experiments, such as with
Washoe, the experiment was “naturalistic”;
I.e., the chimp lived with humans who were
signing
• The idea was that in this context, the
chimpanzee would develop language like a
human child does
Basic Results
• In the basic case, it seems clear that the
primates are able to associate signs (whether
visual or manual) with a ‘meaning’
• This is unsurprising given that many species
can associate a symbol of some type with
something else (consider e.g. dogs)
• Whether this amounts to the primates having
a ‘vocabulary’ is a more difficult question
Harder questions
• An additional question is whether the
primates who have been studies are able to
use new symbols in new combinations
• This is, of course, a central facet of human
language
• In many cases, claims for new use of symbols
or ‘proto syntax’ are taken from long
sessions, and are not always representative
• For instance, it has been claimed that Lana,
upon encountering an orange for the first
time, signed: “apple which-is orange-color”
More context
• Tim: What color of this ?
Lana: Color of this orange
Tim: Yes
Lana: Tim give cup which-is red
Tim: Yes
Lana: Tim give which-is shut ?
Shelley give ?
Tim: No Shelley
Lana: Eye // Tim give which-is orange?
Tim: What which-is orange
Lana: Tim give apple which-is green ?
Tim: No apple which-is green
Lana: Tim give apple which-is orange ?
Tim: Yes
Lana
Note the use of
fixed
expressions and
the fact that the
trainer introduced
color into the
conversation
More Success?
• Kanzi the bonobo
• Involved in language experiments from his
early days; noteworthy claims:
– Learned symbols from watching his foster mother
being taught (I.e. without explicit training of him)
– Supposedly has comprehension abilities of a twoyear old, although under controlled circumstances
– On some performance tests, like 1.5 year old
• No explosion (vocabulary or utterance length)
beyond this.
Synopsis
• The idea that human language should be
taught to non-humans involves several
components
– One the one hand, it is asking primates to do
something that they do not by nature do
– This seems to relate to a perspective on human
language which we have found evidence against
before: if language is just general intelligence,
then primates and other ‘intelligent’ species should
be able to learn some of it
– Of course, we might learn something about
primate intelligence in such projects, but what we
learn about human language is limited
Conclusions
• In complex systems of animal
communication, there are questions about
– What animal calls, etc. ‘refer to’ (if they refer)
– Whether or not they are used to influence the
behavior of others, or are simply responses to
stimuli
– Some evidence exists for the encoding of
information about stimuli that are not present
(bees)
• It is unclear what can be learned about
language per se from experiments that try to
teach human language to non-humans