Plunkett
-
Early
language
acquisition
Theories of early
language acquisition
Kim Plunkett
p&y a crucial role in
I studies of developmental
change in children
promise to contribute
riding of how the brain becomes wired up for language.
disciplinary
psychoiiiistics
perspectives
of cognitive
and neural network
arms in the study of language
word learning
acquisition:
and the acquisition
mechanisms,
highly
a richly structured
sensitive
environment
tic primlpleg
V
cry few of the major landmarks
guiding
=
ahe mrgence
of Ii
Early speech perception
in language develop-
born infants can discriminate
the speech contrasts
their way of cat-
egorizing speech sounds is universal, so that a child born to
role in language development
Japanese-speaking parents has the same phonemic category
has been discussed’-3, in most
remains unclear. Language development
neuroimaging
and children
changes
criminate remains intact until fairly late in the first year. For
of cortical representations.
Recent
ing exrensive plasticity
depends critically
speaking background
in Nthlakapmx,
language system
indigenous
display-
(see Box l), and that its final layout
of which
detailed picture
implementations
Copyright
0 1997, Elrevier
in
Cogmtlve
of the speech input. Com-
Sciences
-
Vol.
reserved.
1,
No
July
processing
voice-
For English-learning
develop$.
sensitivity
appear that sensitivuy
clines earlier
months as
For other con-
can decline even earlier.
for vowels
1997
PII: 51364.6613(97)01039-5
by 6-8 months of age. It
to non-native
contrasts de-
than for consonants*.
ingly, however, not all sensitivities
1364.6613/97/$17.00
4.
in English.
German vowel contrasts [Yl-(U]
would
researcher with important
Science Ltd All rights
Pacific, and the Hindi
For example, Polka and Werker- have shown that English
infants have already lost the ability to discriminate the
clues as to how the brain becomes wired up for language.
Trends
are exploited
trasts, discrimination
of these learning mechanisms
provide the language acquisition
to the Northwest
their phonological
the need for learning mechanisms that are finely tuned to
statistical properties
stop contrast [k’i]-[q’i]
infants, this ability declines after the age of lo-12
of the sophrsticated linguistic propensities of the young infant. In this article, I propose that these findings indicate
putational
velar/uvular
a language spoken by people in a tribe
less aspirated versus breathy voiced contrast [t”]-[dh], neither
on experience. At rhe same time, experi-
extracting
to dis-
have been shown to be able to dis-
tinguish the glottalized
research on infants
mentalists have uncovered an increasingly
speech contrasts, the ability
example, infants of 6-8 months of age from an English-
may reflect
struc-
shows that the developing
parents.
For some non-native
generically-determined
and neuropsychological
as a child born to Spanish-speaking
may de-
and neurophysiological
language development
boundaries
and be-
undergoes striking changes in neural organization,
I
thei
occur during the language learning years and their potential
tures and a consolidation
’
regulariti~:fn
2i .
Newly
a fine tuning of prefashioned,
@
statistical
the necessary ~~~~~
of all human languages *S Furthermore,
in the brain. Conversely,
wford.ac.uk
can be driven
which provides
many critical events in neural development
pend on neuroanatomical
plunkerr@psy.
It is suggested
ment. Although
haviour
r-mad:
that each: ’
i
by the interaction
of
inflections.
to particular
word recognition,
ment have been tagged to specific aspects of brain develop-
cases the causal link between language development
WI: + 4’s 1865 271398
f&Y 144 I865 3 10447
to bear on four distinct
early speech perception,
development
In this r$wi&
experimental
are brought
of grammatical
s how linguistic
I learning
neuroscience,
modelling
to9
to non-native
Interestcontrasts
Box 1. Language
development
in children
Children with early unilateral brain injury to either brain hemisphere typically go on to achieve levels of language performance
that are within the normal range. Earlier reports on recovery of
language in children with focal brain injury led some investigators to conclude that initially the two hemispheres are equipotential for language”. This remarkable plasticity of the developing language system was investigated at different stages of
language acquisition in a recent neuropsychological stndy with
a large sample of subjects. Bates etalb reported on the effects of
focal brain injury during the early development of language in
53 children. All children had suffered a single, unilateral brain
injury to the left or right hemisphere, incurred before six
months of age. The results indicated a striking difference between the critical functional anatomy of the developing and the
mature language system. All of the most consistent hypotheses
regarding language organization in the adult were violated,
namely: left hemisphere dominance, left temporal involvement
in language comprehension and left frontal involvement in ex-
decline in this fashion. Best et al’
English-learning
discriminate
infants
have shown that even
aged 12-14months
are able to
Zulu click contrasts, as are English-speaking
with focal lesions
pressive language functions and grammar. Studies on infants
aged IO-17 months showed that children with right hemisphere injuries are at a greater risk of comprehension impairments. Children with left temporal lesions showed significantly
greater delays in expressive vocabulary and grammar between
the ages of 10 and 44 months. Frontal lesions to either hemisphere disrupted the typical dramatic increase in vocabulary
growth observed around 21 months, independently of motor
impairments. The implication of this important study is that
the localization of language and other cognitive functions in the
adult may reflect the developmental or experience-dependent
status of a behaviour.
References
a
Lenneberg.
Wiley
b
Bates, E.
brain
E.H.
et al.
injury
From first words
tive to correlations
by which
the newly
born
Foundations
to grammar
Dev. Neuropsychol.
infant
all human speech sounds, and the process
of
Language,
in children
with
focal
(in press)
of certain distributional
properties of the
speech stimulus and not others.
ment. Neurons
The mechanisms
Biological
The model is based on vertebrate
adults.
can discriminate
(1967)
are allowed
neuronal
to migrate,
develop-
grow axons and
synapses under the control of genes for various trophic factors. Other genes then control the means by which synapses
by which the child becomes attuned to the parental lan-
are modified
guage are not well understood.
networks is generated and allowed to breed, with a selective
Dehaene’”
Dehaene-Lambertz
have shown that auditory
and
evoked related po-
tentials (ERPs) can be used to unravel the temporal
spatial organization
of the neuronal
phoneme discrimination.
and
processes underlying
They played two-month-old
by experience. A population
pressure for networks
that respond in the desired way to
speech sounds. Network
fitness is calculated
stored output unit activities after the network
in-
function
(e.g. /ba/, lbal, Ibal, Ibal, /pa/) where the first four syllables
the same phoneme
were identical (the standard) and the fifth was either iden-
phonemes as differently
tical or phonetically
erations of this evolutionary
different
(deviant). A significant
uli showed that the infant could discriminate
stimuli.
differ-
ERPs between standard and deviant stim-
Christophe
and Morton”
the deviant
suggest that this tech-
nique might be used to study the developmental
of responses to native and non-native
shedding light on whether
profile
contrasts, thereby
the brain is still sensitive to
favoured networks that represented occurrences of
networks
as similarly
as possible and different
as possible. When, after many genprocess, one of these neural
is exposed to speech spectra from
human languages (including
Farsi, Czech, Hindi,
Hungarian,
Slovak, Spanish, Ukrainian
its connections
and Urdu),
it has been exposed. Furthermore,
tations of speech in the network
Auditory
of speech
sounds that is the same regardless of the language to which
the ability
contrasts is truly lost.
it rapidly modifies
and creates a representation
speech contrasts, but ignores the information,
speech
Swahili,
Korean, Polish, Russian,
or whether
non-native
any of 14
English, Cantonese,
non-native
to discriminate
using the
has been ex-
posed to a test set of spoken English sentences. The fitness
fants synthesized speech stimuli as groups of five syllables
ence in auditory
of these neural
the internal
represen-
show the same categorical
boundaries that are observed in adult and infant perception
ERPs offer an important
new tool for study-
(see Box 2). Once a network
ing which aspects of the acoustic stimulus young infants are
by the evolutionary
sensitive to, when they are sensitive to it and even, in some
required to train the network.
architecture
has been selected
process, only two minutes of speech are
cases, the parts of the brain that are the most revealing of
The innately guided learning exhibited by this network
these discriminatory
capacities. However, ERP measurements are unlikely to tell us how the brain actually accom-
enables it to learn very quickly and makes it less dependent
on the ‘correct’ environmental
plishes the task. Nakisa and PlunketP
account ofhow infants from different linguistic environments
neural network
have developed a
model of early phonological
development.
can learn the same featural representation
The model is based roughly on Jusczyk and Bertoncini’sr3
In this sense, innately
proposal that the development
this model
of speech perception
statistics. The model offers an
should
so soon after birth.
guided learning as implemented
is half-way
between
nativism
be viewed as an innately guided learning process: learning
tivism. It shows how genes and the environment
the speech contrasts
act to ensure rapid development
of the native language takes place
rapidly because the system is innately structured to be sensi-
Trends
Cognitive
Sciences
-
can inter-
of a featural representation
of speech on which further linguistic development
in
in
and construc-
Vol.
1,
No.
4,
depends.
July
1997
$3
(,
,,
$
i
:
I
Plunkett
-
Early
language
acquirltion
$~ibiqp-;~i~~~~~~~~.~ i ‘h.:w&:: :sw!a?=ie2&8*3@
There
Box 2. Categorical
perception
in the Nakisa
Plunkett”
model
is also evidence
prosodic organization
that infants
al.‘” report that prelinguistic
and
are sensitive
to the
of their native language. Jusczyk et
infants have identified
a regu-
larity of English wherein disyllabic words tend to adhere to
a trochaic
(strong-weak)
stress pattern”.
Newsome
and
Jusczyk’* show that infants aged seven and a half months
Categorical perception of phonemes is a robust phenomenon observed in both infants and adults. The nerwork was
tested on a series of 11 spectra which fnrmed a linear continuum from the pure ishl to a pure Is/. Individually,
each of the 11 spectra in the continuum were fed into a
network that had been trained on 30 sentences of continuous speech in English. The output feature responses were
stored for each spectrum in the continuum. The distances of
rhese feature vectors from the pure lshi and pure /s/ indicated the categorical nature of the network’s internal representations of the speech spectra, as shown in the figure
below.
All of the human languages tested seemed to be equally
effective for training the network co represent English
speechsounds. To seewhether any sounds could be used for
training, the network was trained on white noise. This resulted in slower learning and a lower final fitness. The fitnessfor a network trained on white noise neyer reached that
of the same network trained on human speech. An even
worse impediment tn learning was to train on low-pass filtered human speech.
can use this knowledge
to segment disyllabic
words from
the main speech stream. Young children are also more likely
to imitate syllables that are stressed or word-final
bles that are both unstressed and nonfinal”.
using a preferential
looking
et aLao,
task, have shown that infants
are more likely to recognize familiar
initial or final position
than sylla-
Fernald
words in utterance-
than when the word occurs in the
middle of the utterance.
Saffran
al.” have focused on the ability of young chil-
et
dren to acquire linguistic
point
structure via statistical cues. They
out that the statistical
words are potentially
properties
of multisyllabic
useful for infant word segmentation.
Over a corpus of speech sounds, there are measurable regularities
that distinguish
those recurring
sound sequences
that comprise words from the more accidental
sound se-
quences which occur across word boundaries. Using the familiarization-preferential
Saffran et A*’
infants
a two minute
the infants’
(see Box 3),
computations.
stream containing
exposure to a synthetic
are
speech
only statistical cues to word boundaries,
listening
had extracted
procedure
the necessary statistical
able to perform
Following
looking
showed that eight-month-old
preferences demonstrated
and remembered
about the familiarization
that they
serial order information
items, distinguishing
‘words’ (re-
current syllable sequences) from syllable strings spanning
word boundaries. This preferential
the infants
computed
behaviour indicates that
the co-occurrence
pairs of sounds across the familiarization
al.*’
report
that nine-month-old
month-olds)
are attentive
phonotactic
infants
3
4
5
6
7
8
910
Sample
/S/
/Sh/
(but
to the frequency
sequences occur within
together with the findings of Saffran
012
frequencies
with
R.C. and
neural
network:
Plunkett.
K. Innately
the care of featural
guided
learning
representation
et
al*‘, suggest that in-
fants have access to a powerful
mechanism for the compu-
tation of statistical
of the language input even
properties
infants may be far better at deriving structure from statisti-
by a
cal information
of speech
Lang. Cogn. Processes (in press)
tion literature.
than has often been assumed in the acquisiIn particular,
certain aspects of language that
are argued to be unlearnable
and thus innately
may be discoverable by appropriately
Word recognition
specified
constrained statistical
learning mechanisms.
During the first year of life, infants become attuned to more
Recent connectionist
and statistical analyses23-*5 of the
than the phonemic contrasts of their native language. They
properties of real language corpora have contributed
pick up knowledge that enables them to identify words and
view that the distributional
other linguistic
very useful to the language learning
units in speech. Jusczyk and A.&t’*
used the familiarization-preferential
Box 3) to demonstrate
speech. Jusczyk
have
looking procedure (see
Christiansen
that even infants aged seven and a
half months have some ability
vided with
et al. ” showed that nine-month-old
infants prefer to listen to word lists
that conform
to the phonetic and phonotactic
Cognltlve
Sciences
-
Vol.
1.
No.
4.
July
between
cues to word boundaries
American
tual word boundaries.
child-directed
1997
child. For example,
about phonemes,
these sources of information
infants showed no preference for lists from either language.
to the
in the input may be
task. The model was explicitly
information
stress and boundaries
structure of
their own language. In contrast, six-month-old
information
et ~1.‘” trained a simple recurrent network on a
phoneme prediction
to detect words in fluent
and Dutch
in
which
English. These results,
between the age of six and nine months. They indicate that
a Nakisa,
Trends
et
not six-
from very brief exposures and that this develops sometime
Reference
American
for
corpus. Jusczyk
utterances.
provide
relative
prolexical
Individually,
relatively
unreliable
and no direct evidence about ac-
After training
on a large corpus of
speech, the model was able to use these cues
Plunkett
The familiarization-preference procedure was developed by
Jusczyk and A.&n’. In this procedure, infants are exposed to auditory material that serves as a potential learning experience.
Subsequently, they are presented with two types of test stimuli:
(a) items that were contained within the familiarization material, and (b) items that were highly similar but are not contained
within the familiarization material. During a series of test trials
that immediately follows familiarization, infants control the
-
Early
language
acquisition
duration of each test trial by their sustained visual fixation of a
blinking light. If infants have extracted the crucial information
about the familiarization items, they may show differential durations of fixation (listening) during the two types of test trials.
Reference
a Jusczyk,
patterns
P.W. and Aslin,
of words
R.N. (1995)
in fluent
Infant’s
speech Cognitive
detection
Psycho/.
of sound
29, 1-23
R
Video camera
Child
8
Flashing light
I
Loudspeaker
Parent
to reliably identify word boundaries. The model shows that
Schafer and PlunketP
have succeeded in replicaring
aspects of linguistic structure that are not overtly marked in
findings of Woodward
et al. under conditions
the input can be derived by efficiently
probabilistic
combining
multiple
require the presence of an instructor,
cues.
vocabulary
the
that do not
suggesting that the pre-
spurt child is already equipped with a powerful
learning mechanism for forming object-label
associations.
Word learning
One of the most dramatic manifestations
of language devel-
Imaging early language development
opment during rhe first two years of life is the rapid increase
Event-related
in the rate of vocabulary
opmental changes in neural processing in normal children.
development
observed at around
18 months of age. This spurt in development
in both comprehension
and production”.
usually occurs
There are three
Mills
potentials
have been used to examine devel-
et aL3’ examined the changes in the organization
brain activity
linked to comprehension
of
of single words in
main families of theories about the mechanism underlying
infants aged from 13 to 20 months. Auditory
this vocabulary spurt. These are linguistic development2’-z9,
potentials were recorded as children listened to both a series
conceptual
development30~3’ and the development
of con-
of words whose meanings they did or did not understand,
straints on word learnin$z-34. All of these theories postulate
and to words pronounced
the triggering
a function
child’s
of a new principle
understanding
Woodward
of
of organization
the object/label
into the
relationship.
et aLj5 argue that these explanations
evoked related
word
of word
onset. At
backwards. The ERPs differed as
comprehension
13-17months
within
200ms
after
of age comprehension-
imply that
related differences
were bilateral
and broadly
learning a new word prior to the vocabulary spurt is likely to
over the anterior
and posterior
cortex.
be a time-consuming
distributed
In contrast,
at
process, requiring
considerable expo-
20 months of age these effects were limited to temporal and
sure to a new word. There is growing
evidence, however,
parietal regions of the left scalp.
that the young (prevocabulary
spurt) child may not be as
These results indicate that the neural organization
for
hampered in learning new words as was thought previously.
word comprehension shifts, and that this shift occurs precisely
Woodward
during the period of development when language acquisition
et a1.35 report that, under favourable
stances, 13-month-old
as few as nine
BaldwiG
circum-
infants can learn novel words from
presentations
of a novel
word
token.
argues thar joint attention between the infant and
its instructor
is necessary for word
learning.
However,
is most pronounced.
reorganization
The implication
is that plasticity
and
may be natural properties of the developing
system and are not restricted to compensatory
changes in
damaged brains. Mills et aL3* suggest that aspects of their
Trends
in
Cognitive
Sciences
-
Vol.
1,
No.
4.
July
1997
Plunkett
-
Early
language
acquisition
Box 4. Multimodal
model of vocabulary
(A) Profile of vocabulary scores typical for many children during
their second year (taken from Ref. a). Each data point indicates
the number of different words usedby the child duringa recording
session. The vocabulary spurt that occurs around 22 months is
observed in many children. It usually consists of an increased
rate of acquisition of nominals, specifically names for object&.
(B) Simplified version of the network architecture used in
Plunkett et al.’ The image is filtered through a retinal preprocessor prior to presentation to the network. Labels and
images are fed into the network through distinct ‘sensory’
channels. The network is trained to reproduce the input patterns
at the output, a process known as autoassociation. Production
corresponds to generating a label at the output when an image
is presented at the input. Comprehension corresponds to genet-
ating an image at the output when a label is presented at the
input. The model exhibits the same non-linear pattern of
vocabulary growth observed in young children, both in comprehension and production. Furthermore, comprehension scnres
are always ahead of production scores. The network model also
produces over- and under-extension errors.
References
a
b
12
.-a’
14
I’
1’
16
I’
I
20
18
K. (1993) Lexical segmentation
language
McShane,
acquisition
J. (1979)
and vocabulary
J. Child Lang.
growth
in
20. 43-60
The development
of naming
Linguistics
17,
879-905
c Plunk&,
Jens’ vocabulary
OJ
Plunkett,
early
A
140
acquisition
K. et al. (1992)
symbols?
Vocabulary
Connect
sci 4.293-312
Symbol
growth
grounding
in children
or the emergence
and a connectionist
of
net
B
Preprocessed image
Label
Preprocessed image
Label
II
22
24
Age (months)
ERP findings are linked to changes in early lexical develop-
that occurs in young children towards the end of their sec-
ment that occurs typically
ond year. Furthermore,
of age. However,
between
13 and 20 months
it is still unclear whether the changes in
ERP reflect qualitative
changes in the underlying
processing of lexical items or a consolidation
lexical representations.
language
infants on novel words. Combining
the observed
hemispheric
that are often observed in young children
task for
multiple representations.
the results observed
to
This modelling work suggests that
in children with focal lesions (see Box
1) and in ERI? studies of word comprehension
specialization
necessarily imply prewired,
for lexical processing arises from prolonged experience with
are entirely
consistent with the non-linear
behaviours
linked
of new cognitive processing
strategies.
the underlying
modelling
has demonstrated
that small and gradual changes in a network,
the maturation
in its behaviour.
For example, Plunkett
have developed a connectionist
that involves an auto-associative
model involved small continuous
strengths within
in
Cognltlve
learning
representations
in
neural system.
Sciences
-
Vol.
1.
4,
block its application.
formation
exceptional
words: a rule-governed
For example, in this view, the plural
of ‘sheeps’ is blocked by the identification
plural
form
whereas plural formation
‘sheep’ in associative
of ‘boys’
of the
memory
is achieved by appli-
cation of the rule (add Is/) to the word ‘boy’. The rule-
spurt
July
mor-
memory attempts to identify the exceptions to the rule and
of the
behaviour of
No.
of inflectional
mechanism for the pro-
process attempts to inflect all words, while an associative
in vocabulary
the well known vocabulary
morphology
accounts of the acquisition
cessing of regular and exceptional
processing modal-
(see Box 4), the linguistic
mimicking
et al”
changes in the connection
and across the different
ities in the network
of multiple
phology40.4’ assume a dual-route
process
the training
the network exhibited dramatic nonlinearities
Trends
Symbolic
model of children’s vocabu-
of relating labels to images. Although
development,
Inflectional
not involving
of new systems, can lead to dramatic non-
lary development
onset of overt
experience-driven
of non-linear word learning
Recent work in connectionist
linearities
to gradual
processes and coordination
need not
dedicated modules. The results
words or from the development
Connectionist mode&g
development,
between comprehension
may be a natural outcome of the child’s attempt to integrate
novel word
learning with ERP measurements offers an opportunity
evaluate whether
and production
have devel-
oped a technique based on the preferential-looking
training
suggesting that the asymmetries
of existing
Schafer and Plunkett”
the model showed clear-cut dissoci-
ations in receptive and expressive vocabulary
governed process acts as a default that applies to any word,
1997
offering
the language user economy in representation
Plunkett and Nakisa5’ trained a neural network on the
(no
need to store information about inflected forms that conform to the default) and creativity (the capaciry to inflect
Arabic plural and evaluated its performance
forms previously not encountered).
In contrast, connectionist accounts of the acquisition
work was superior to the dual-route
encountered
of
on words not
in the training set. They showed that the netmodel at predicting
the
plural class of Arabic words on which it had never been
inflectional morphology4246 assume a single-route mechanism for the processing of both regular and exceptional
trained.
forms. There is no distinction
model. In a similar fashion, Nakisa et ~1.~~have shown that
regular and exceptional
In particular,
prediction
of membership
in the
sound plural class was more accurate in the neural network
in the manner in which
forms are handled in this account.
a connectionist
network trained on a subset of German plu-
They are processed by the same network of connections that
rals predicts accurately
maps an uninflected
plurals that it has never seen before. The network is in much
The network’s
form of the word to its inflected form.
capacity to inflect novel forms is shaped by
the class membership
of German
the same position as the Arabic or German child who may
its experience with the forms on which it has already been
have to guess how to form the plural of a word. These re-
trained.
In English, the inflectional
sults indicate that the distribution
systems ofthe past tense and
the plural are highly regular. Irregular past tense forms and
ship which
irregular noun plurals constitute only 14% and 2% of their
sional linguists.
respective systems47. The dual-roure
morphology
of forms need to be stored in associative
memory and the default rule can deal with the majority
forms. A connectionist
the words.
network stores information
Nevertheless,
the dominance
words in the system results in the network
lar responses to novel words.
route and connectionist
are not obvious even to sophisticated
profes-
account of inflectional
is very efftcient at representing these systems as
only a minority
of nouns in Arabic and
German may provide subtle clues to plural class member-
of
about all
of the regular
producing
Consequently,
Conclusions
As yet, none of the domains of language acquisition
de-
scribed
the
above
are understood
However,
picture of the language learning child is becoming increasingly refined as we uncover the details of what is developing
regu-
both dual-
approaches can explain the prepon-
and when development
and how
linguistic
these systems actually
derance of regular responses to novel words by English
neuropsychological
the young infant
account
occurs, where the neural systems in
the brain for controlling
speakers but for different
reasons: the dual-route
properly.
behaviours
function.
and computational
is richly
are located
Behavioural,
studies reveal that
endowed with
neural systems,
exploits a default rule that attempts to regularize any word
well-adapted
available to the language user; the connectionist
cessing. At the same time, I believe that a multidisciplinary
ploits the skewed distribution
account ex-
in favour of regular words in
approach
the language.
the utility
Minority
with
to the business of linguistic
to the study of language acquisition
of viewing
the interaction
defaults
information
a richly
linguistic
of powerful
structured
development
points to
as driven by
general learning
environment
pro-
mechanisms
that provides
the
There is evidence from speakers of other languages that
necessary ingredients for the emergence of mature linguistic
their ability to produce a default response to novel words or
representations.
overgeneralize the default to exceptional words does not rely
upon a numerical
superiority
of the words that epitomize
Outstanding
the default in the language. For example, Clahsen et al.**
and Marcus et aL.*’ claim that the ‘s’ plural in German is the
default process even though it constitutes a minority
l
of the
plural forms in the language. A similar claim is made for the
default status of the ‘sound’ plural in Arabic. These authors
claim that languages whose speakers conform to a minority
default pattern, appear to present a major challenge to connectionist
accounts of inflectional
morphology
l
as networks
operating on the principle of ‘similar inputs produce similar
outputs’ are unlikely
to produce a default response to novel
forms.
Hare
et alSo have demonstrated
models of inflectional
morphology
that connectionist
can learn a default re-
l
sponse even in the absence of superior numbers for the default class. Two factors contribute
respond in a default-like
to a networks
capacity to
fashion: First, words which look
similar at the input need not have similar internal representations. Second, the distribution
of the words in the lan-
guage influences the ability of the network
fault-like
to act in a de-
fashion (see Box 5). Under appropriate
(see Box 5), it is possible for a network
conditions
to learn a distribu-
tional default.
Trends
l
questions
What are the mechanisms that enable infants to tune in to the speech
contrasts that are specific to their native language? Why do some nonnative speech contrasts remain discriminable
by adults and children
while others are no longer perceived as distinct? Does the brain remain
sensitive to non-native speech contrasts even though discrimination
experiments demonstrate
a lack of such sensitivity?
Are the word-like
linguistic chunks that infants extract from continuous
speech stored in a prelexical mental repository awaiting adequate
conceptual development
to achieve semantic grounding?
How does
conceptual development
influence the process of lexical segmentation?
What developments
underpin the dramatic changes in the statistical
processing of speech that seems to occur between six and nine months
of age?
What is the nature of the label-object
associations shown in recent
demonstrations
of rapid word learning in prevocabulary
spurt infants? Is
the lateralization
of early word representations
to the left hemisphere in
postvocabulary
spurt infants a consequence of prolonged experience
with specific words or is it due to the emergence of new lexical
processing strategies?
What are the facts of acquisition in complex inflectional
systems like the
Arabic and German plural? Are there default processes operating in
these and other languages or are over-regularizations
and
generalizations
better understood as operating through processes
governed by analogy and frequency?
in
Cognitive
Sciences
-
Vol.
1,
No
4.
July
1997
Box5, How
to obtain
a minority
lkett’ trained a neural network
!,
‘I
default
Period of training network
to categorize
elonging
to one of three classes. Each input
a point
on a two-dimensional
points is shown below. The majority
Late
Middle
Early
plane. The
of the
,d in two squares. All the points within
II,
from a neural network
a
L to belong ro the same class. These can be
toI’
/,I
~ exceptional
I
,~Ol,,
of the
these square regions. All the
1,: square regions belong to the same class.
HIC\.
‘8
.lL~l.illll ‘8
,I,( j
I ught
of as the patterns
s,,,
11(work trained
88 I patterns,
I I ,I.,<
’ I,\,18
il.irl/lcl
representing
~rere) class. The question
‘i~ll<liil.
i:v
patterns. The minority
uted outside
on this distribution
I, 10 specie
in the two-
it has never been trained? The
81J\ used by Forrester
,.,I ‘,/I
the
is how
of points
that is, the points
I IL. on which
II ,
of interest
and Plunkett’
contained
the x, y coordinates
in the two-
<II u:,k 1111/ 181‘II’, 20 hidden units that formed
internal
ii’~i,,I.,ll~~’
or IIX input
patterns
a non-linear
I d\\ltik III,
1’ /he input
space and three output
permitting
L1.~t\~t~ I/C III,T\I: patterns. The network
units
I : j 1, !‘,I.<’ plots show the activation
&I UWII .I
il~:t~lmtstages in
training.
I~‘+\~
iiil’b.i, lill The final column
,111!h
<I I\\I’I,:~ lmits
to
was trained with the
,v1111\ ~/ t ‘,,I1 )i low and then tested on every point
}‘I in.
rep-
in the
of the three classi-
Darker regions indicate
(late training)
shows that
do a good job at partitioning
the space
lx :hc irlc: (I/ II Inning. In particular,
most of the points in the
r\~o-~i~rrli.tl\~,~n.~l plane are treated as though
they belong to
+++iL+&--SM
-w+++++m _- oo&
0000000000-- 8wm
ooooooomo
t2fim
0000oooooO
_
-ooocloooooo
oooooooooo
- --
I~I< rhll-,t ~1.a.~.I he so-called default, even though the training
\a Lon~S~un~~
I ,/ nlinority
of forms in this class. This example
~~I:~oI-~~I~~,I~~~
II~\v a neural network
can be trained to produce
A d~i‘:iilr
it has the
IIL
response
provided
U~IIQ:~I~I 111~rn.11 representations
~I~I~~KIC 1st 111~ mput space and provided
‘l.~ngu.~gc’ AI t. .i~lpropriately
resources
that permit a non-linear
to
par-
the forms in the
distributed.
Reference
a Forrester,
N and Plunkett,
the case tor minority
I” Proceedings
K. (1994)
default
of the Sixteenth
Conference.
Atianta,
pp. 319-323.
Erlbaum
Learning
mappings
GA (Vol.
the Arabic
in connectionist
Cognitive
Science Society
16) (Ram,
A. and
.
..
plural.
networks,
Annual
Eiselt,
.
K.. eds),
. . ..
...
..
.
6 Jurczyk,
References
its neural
(Rapin,
2 Elman.
correlates,
J.L. et
3 Quaftz.
al.
(1996)
S.R. and
Rethinking
the
(Vol.
7)
Elsevier
Innateness:
a Connectionist
Sejnowski.
T J. The
neural
Behav.
perception
basis
of
cognitive
Science
171,
12 Nakisa.
P.W. (1992)
speech
signal,
implications
C.A.,
for
Behav.
L. and
from
Models,
the
Research,
Steel-Gammon
13 lusczyk,
C.. eds),
speech
Cross language
reorganiration
during
speech
the
first
perception:
in
year
of life
patterns
Dev. 7.49-63
R.C. and
the
Dehaene,
Morton,
S (1994)
in infants
J. (1994)
Plunk&t,
K. Innately
case of featural
P.W. and
perception
15 Jusczyk,
J.F. (1994) Developmental
contrasts
J. Exp. Psycho/.
changes
Hum.
in perception
Percept.
native
Perform.
16 Jusczyk.
P.W.
click
1. Exp. Psycho/.
Speed
Nature
and
cerebral
370,292-295
Comprehending
guided
baby-think
learning
representatw
Bertoncini,
1. (1988)
as an innately
Sciences
-
Vol.
1.
No.
4,
July
1997
and
of words
P.W.
language
P.W.,
predominant
Cognitive
and infants
of
Zulu
370, 250-251
14 Jusczyk,
20.421-435
Trends
contrasts:
of speech
by a neural
Lang.
Cog”.
Viewing
guided
the
process
development
Lang.
of
Speech
31,
217-238
perceptual
vowel
Examination
Processes (in press)
Press
L. and Werker,
of non-native
categories
Development:
Menn,
J.F. and Tees, R.C. (1984)
evidence
Infant
York
phonological
Phonological
(Ferguson,
pp. 17-62,
6 Werker,
in
Developing
MIT Press
(1988)
speech
adults
discrimination
A. and
Language,
N.M.
14, 345-360
G. and
of syllable
network:
5 Jurczyk,
7 Polka,
by English-speaking
Perform.
Percept.
Nature
303-306
for
discrimination
11 Christophe,
Brain Sci. (in press)
m Infants
reorganization
Hum.
correlates
manifesto
Speech
perceptual
of Spoken
G.W. and Sithole.
10 Dehaene-Lambertz,
MIT Press
a constructiwst
P.D. et al. (1971)
development
of Neuropsychology
S., eds), pp. 669-710,
on Development,
development:
J. (1992) Early language
in Handbook
I. and Segalowitz.
Perspective
4 Eimas,
P.W. (1996) The Discovery
9 Best, C.T.. McRoberts,
1 Bates, E., Thal, D. and Janowsky,
and
+ Exception 2
0 Exception 1
- Minority default
etal.
Aslin,
in fluent
(1993)
words
Cutler,
R.N.
(1995)
speech
Infants’
detection
Psycho/.
sensitivity
to the sound
1. Mem.
Lang.
A. and
Redanz,
stress patterns
Infant’s
Cognih’ve
of English
of
sound
29, l-23
patterns
of
32,402-420
N. (1993)
words
Preference
for
Child Dev. 64,675-687
the
Plunkett
17 Cutler,
A. and Caner,
syllables
D.M.
in the English
18 Newsome,
(1987) The predominance
vocabulary
M. and Jurczyk,
for segmenting
fluent
on Language
P.W. (1995)
speech?
Development
pp. 415426,
Caxadilla
20 Fernald,
word
21 Saffran.
22 Jurczyk,
P.W.,
phonotactic
23 Brent,
ability
Mem.
Conference
McEwen,
S., eds),
in
2, 16%220
C. Effects
to recognize
of prosody
words
in fluent
and
learning
P.A. and
patterns
Lute,
in the native
language
for
Infant’s
J. Mem.
sensitivity
Lang. 33,630645
Distributional
regularity
segmentation
Cognition
constraints
are useful
M.H.,
J. and Seidenberg,
to
93-125
speech
using
Allen,
multiple
cues:
MS. Learning
a connectionist
to segment
model
Lang.
25 Reddington.
M.
approaches
and
Chater.
N.
acquisition:
Connectionist
and
a distributional
statistical
perspective
26 Goldfield.
E. and
development
Reznick,
1.5.
in comprehension
Social Context
Conditions
(1992)
and
Rapid
production
for the acquisition
of Language
(Markova,
20 Lock, A. (1980) The Guided
29 Plunkett,
K. (1993)
early language
30 Corrigan,
change
in
Dev.
Psycho/.
acquisition
permanence
lexical
of
acquisition
and
categorization
cognitive
in the
32 Markman.
E.M.
exclusivity
Perspectives
on
Development
Cambridge
33 Golinkoff.
principles
development
A.
University
second
E.V. (1993)
year
The
Press
growth
verbs,
learning
Center
H.J. Language
13 months
Symbol
by
Res. Lang.
comprehension
to
grounding
20months
or the
in children
and
Overregularization
A. (1988)
On
distributed
Dev.
emergence
of
a connectionist
net
in language
acquisition
language
and
processing
connectionism:
model
of
language
28. 73-193
in Parallel
J.L. (1986) On learning
Distributed
of Cognition
Processing:
(McClelland,
the past tense
Explorations
J.L. and
in the
Rumelhart,
D.E..
MIT Press
B. and
K.
frequency
Leinbach.
revising
J. (1991)
the
verb
Implementations
learning
model
are
not
Cognition
40,
and
its relation
to
and
of
the
object,
taxonomic
and
on word
other
Thought:
Gelman,
eds),
72-106,
(1991)
U-shaped
perceptron:
learning
and
implications
for child
38,43-102
V. (1993)
morphology
From
rote
learning
in children
and
to system
connectionist
48, 2169
Marchman.
V. (1996)
acquisition
G. (1995)
of the
Children’s
analysis
Learning
English
from
past
a connectionist
tense
Cognition
61,
(1992)
of German
M.,
50 Hare.
and
plurals
German
Psycho/.
Elman,
Regular
noun
G. et al. (1995)
the rule Cognitive
overregularization
of English
plurals:
a
J. Child Lang. 22, 447459
H. et al.
generalization
Press
Irregular
Cognition
Inflection
in the
45, 225-255
inflection:
the exception
that
proves
29, 189-256
J.L.
and
in connectionist
Daugherty,
networks
K.G.
Lang.
(1995)
Cogn.
Default
Processes
10,
601630
Young
new nouns
children
Dev. Psycho/.
The Lexicon
A.L., Markman.
verb
quantitative
49 Marcus,
in
and
pp.
acquiring
of the
48 Clahsen,
mutual
Interrelations
S.A..
Marchman,
acquisition
meanings.
Cognition
K. and
K. and
47 Marcus,
development
V.
299-30s
5, 173-189
constraints
and
J.P.
to stage 6 object
Marchman,
in a multi-layered
acquisition
46 Plunk&t,
model
The
and
effects
45 Plunkett.
and
adults
use lexical
51 Plunkett,
28, 99-108
in Acquisition,
plural
Cambridge
University
E.M. and Fitzsimmons,
in 13. and 18.month-olds
Dew Psycho/.
C.M. (1994)
K. and
Raptd
30. 553-566
Nakisa,
system Lang.
52 Nakisa,
single
learning
Prince,
a parallel
nets Cognition
in
Press
35 Woodward,
from
D.E. and McClelland,
building:
Child Deb’. 58, 1523-1531
whole
as initial
R.M. et al. (1992)
to learn
Wiley
Academic
as related
(1987)
developments
Language
(Bymes,
acts, in The
and vocabulary
J. Child Lang.
(1991)
assumptions
of Language,
segmentation
Meltzoff,
and linguistic
S.A. and Neville.
Cognition
44 Plunkett,
28,
J. Child Lang. 20, 43-60
development
A.,
of speech
I., ed.), pp. 87-111,
Reinvention
Lexical
R. (1978) Language
31 Gopnik,
word
5. and
language
1. (1978)
word
conditions
121-157
406413
27 Dare,
Rapid
Sot. Res. Child Dew 57. l-182
conceptualizations:
Cogn. Processes (in press)
34 Clark,
Monogr.
43 MacWhinney,
Lang.
K.
controlled
growth
eds), pp. 216271,
to language
attention
(in press)
Microstructure
Processes (in press)
joint
and Role in Development
sci. 4. 293-312
of English
Cogn.
tightly
G.F. et a/. (1992)
42 Rumelhart,
24 Christiansen,
under
specialization
connection
analysis
61,
Plunkett,
K. et a/. (1992)
41 Pinker,
and
link between
P., edr), Erlbaum
and
Vocabulary
40 Marcus,
J.C. (1994)
T.A. (1996)
cerebral
39 Plunkett,
by
acquisition
(in press)
symbols?
E.L. (1996) Statistical
the
its Origin
38 Mills, D.L., Coffey-Corina,
Neuropsychol.
speech
language
Attention:
C. and Dunham,
G.
Early
Understanding
in Joint
1%month-olds
and
Science 274. 1926-1928
Lute,
(Moore,
37 Schafer,
Cogn. (in press)
M.R. and Cartwright,
phonotactic
Acquisition
R.N. and Newport,
infants
D. and
G.W. and Herrera,
Learn.
J.R., Aslin,
use stress as a cue
University
Ned.
Language
on infants’
B-month-old
Do Infants
in 19th Boston
D.A. (1996)
and language,
E.L. (1992) The role of stress and position
A., McRoberts,
position
36 Baldwin,
mltlal
Press
first words
J. Exp. psycho/.
of strong
Speech Lang. 2, 133-142
(MacLaughlin,
19 Echols. C.H. and Newport,
determining
Comput.
-
R.C.. Plunkett,
and dual-route
Language
Acquisition:
P. and Murre,
R.C. A connectionist
model
of the
Arabic
Cogn. Processes (in press)
K. and Hahn,
models
U. A cross-linguistic
of inflectional
Inductive
comparison
morphology,
and Deductwe
in Models
Approaches
of
of
(Breeder,
J., eds). MIT Press (in press)
In the September issue... .
The Trends Guide to the Internet (1997)
Feeling left behind? Worried about time-wasting?
This free guide will lead you through those ail-important
first
steps towards using the Internet efficiently
and effectively.
Written by an international
panel of experts, the
guide will tell you what to expect and where to find it. Articles include:
The origins of the Internet
Key terms: the jargon explained
Basic Internet facilities and how to connect
The World Wide Web
USENET: setting up and joining newsgroups
Creating your own home page
FTP: how to retrieve files from around the world
What’s new in online journals
Where to go next for help
Also including
For further
information
a poster listing useful sites to visit.
concerning bulk sales, contact Thelma Reid: e-mail, t.reidBelsevier.co.uk;
Tel, +44 1223 311114; Fax, +44 1223 321410.
Trends
in
Cognitive
Sciences
-
Vol.
1.
No.
4.
July
1997