Language in Primates - University of Arkansas

Language in Primates
• Do our closest evolutionary relatives have
the ability to learn and use language?
• If so, then the differences between our
respective species may be less than we have
previously believed.
– Chimpanzees (Washoe, Loulis, Nim)
– Bonobos (Kanzi and Matata)
– Gorillas (Koko)
Chimps: Washoe
• Adopted by Drs. Beatrix
and Allen Gardner
• Studied by Drs. Roger &
Deborah Fouts since 1980
• First chimp to be taught
ASL
• Was able to learn about
200 signs and combine
them 2 or 3 at a time
• Taught signs to her
adopted son Loulis
Chimps: Nim Chimpsky
• Taught ASL by Herbert
Terrace.
• Skeptical of reports about
Washoe
• Was able to teach Nim
numerous signs
• Never saw evidence that
Nim could combine signs
except when promted to
do so by experimenters
• Nim died last year
Bonobos: Kanzi
• Studied by Dr. Sue
Savage-Rumbaugh
• Was present when his
mother (Matata) was
being taught to use
lexigrams
• Though Matata didn’t
learn, Kanzi did
• Spontaneously learned and
combined many lexigrams
• Shows evidence for
comprehending word
order
Bonobos: Is Kanzi typical?
• In a word, no.
• Though other bonobos have learned some lexigrams, only
one other has done so to the degree of Kanzi. (His mother
learned 6 lexigrams in 5 years, his sibling learned about 15
in 3 years).
• Was it because Kanzi is somehow unique, or were the
conditions of his rearing unique?
• Probably the latter. Learned the symbols through natural
exposure rather than rigid training.
• Best evidence for language acuqisition in non-humans to this
date
Gorillas: Koko
• Dr. Francine Patterson
taught Koko ASL
• Knows several
hundred signs and
combines them
• Uses signs for emotion
(e.g. Sad kitten gone)
• Chats on the internet
(no, really)
• Would Koko like to have a kitten, a dog, or another Gorilla as a friend?
• LiveKOKO:dog
• DrPPatrsn:She actually has two dog friends right now one kitty and
two gorillas.
• HaloMyBaby:SBM87 asks, What are the names of your kittens? (and
dogs?)
• LiveKOKO:foot
• DrPPatrsn:Foot isn't the name of your kitty
• HaloMyBaby:Koko, what's the name of your cat?
• LiveKOKO:no
• Koko tell us what you look like in your words?
• LiveKOKO:flower
• DrPPatrsn:One of the scrunchies has a big flower on it.
• LiveKOKO:eat now
• DrPPatrsn:She wants some more of the snack, apparently.
• LiveKOKO:sleep, red red
• DrPPatrsn:She's indicating the red scrunchie.
How about computers?
• Many attempts to get computational systems to
learn language a la the child
• Most have used some sort of connectionist, or
parallel distributed processing, network to
accomplish this feat (artificial neural networks)
• Try to model various aspects of language
acquisition
– Vocabulary learning (Elman)
– Acquisition of the past tense (Rumelhart &
McClelland)
Rumelhart & McClelland (1986,
1996)
• Past-tense learning
• Works on phonological pattern recognition
and prior experience with regular and
irregular verbs
• Attempts to generalize rules and apply them
to novel stems the system has never
encountered before (just like a child)
How does it do?
• Correctly generalizes to about 70% of unfamiliar
word stems
• Makes many errors of types that children never
make
–
–
–
–
–
–
mailmembled
winkwok
satisfysedderded
smairfsprurice
frilgfreezled
smeejleefloag
What to conclude
• Pinker and others take these sorts of results as
proof that a general AI learning system could not
learn language
• Others see these as good approximations of what
actually occurs, and simply that the model and/or
parameters and/or the input needs to be specified
differently
• Will we ever have a language learning machine?
• Nick Lacey’s gonna tell us