See discussions, stats, and author profiles for this publication at: [Link]

net/publication/5284369

The Emergence of Language

Article in Annual Review of Psychology · February 1998

DOI: 10.1146/[Link].49.1.199 · Source: PubMed

CITATIONS READS
436 7,382

1 author:

Brian Macwhinney
Carnegie Mellon University
369 PUBLICATIONS 31,161 CITATIONS

SEE PROFILE

All content following this page was uploaded by Brian Macwhinney on 05 November 2014.

The user has requested enhancement of the downloaded file.

Carnegie Mellon University
Research Showcase
Department of Psychology College of Humanities and Social Sciences

2-1-1998

Models of the Emergence of Language

Brian MacWhinney
Carnegie Mellon University, macw@[Link]

Recommended Citation
MacWhinney, Brian, "Models of the Emergence of Language" (1998). Department of Psychology. Paper 169.
[Link]

This Article is brought to you for free and open access by the College of Humanities and Social Sciences at Research Showcase. It has been accepted
for inclusion in Department of Psychology by an authorized administrator of Research Showcase. For more information, please contact
kbehrman@[Link].
 Models of the Emergence of Language

MODELS OF THE EMERGENCE OF LANGUAGE

Brian MacWhinney
Department of Psychology, Carnegie Mellon University, Pittsburgh, PA 15213,
macw@[Link]
Appeared in: Annual Review of Psychology 1998 49: 199-227

Language is a uniquely human achievement. All of the major social achievements of

human culture -- architecture, literature, law, science, art, and even warfare -- rely on the use
of language. Although there have been attempts to teach language to primates (Allen &
Gardner, 1969; Savage-Rumbaugh, Sevcik & Hopkins, 1988), the successful learning of
human language seems to be a tightly copyrighted component of our basic human nature.
This view of language as a Special Gift has led some researchers (Bickerton, 1990) to
hypothesize that some small set of evolutionary events may have triggered the emergence of
language in the human species. Others (Chomsky, 1980; Fodor, 1983) have argued that the
capacity to learn language is a unique property of the human mind that is represented
neurologically in a separate cognitive module. These scholars believe that this modular
architecture allows the shape and form of human language to be largely independent of
other aspects of cognitive processing or social functioning. Studies of language learning
stimulated by this nativist perspective have tended to focus attention onto a small set of
syntactic structures that are thought to constitute the core of Chomsky’s Universal
Grammar (Chomsky, 1965). According to the “principles and parameters” model of
language structure (Hyams & Wexler, 1993), the learning of particular languages occurs
through a process called parameter-setting. During parameter setting, children identify the
exact shape of their mother tongue by choosing the proper settings on a small set of binary
oppositions. For example, a positive setting on the pronoun omission parameter will select
for languages like Italian or Chinese, whereas a negative setting will select for English.
Recent studies of the neural basis of communication systems in organisms such as
crickets (Wyttenback, May & Hoy, 1996), quail, and song birds (Marler, 1991) have
emphasized the extent to which species-specific communication patterns are stored in highly
localized hard-wired neurological structures. However, even these lower organisms display
some developmental plasticity in the ways in which communication is supported by the
brain. When we look at human language learning we see that children learn language
gradually and inductively, rather than abruptly and deductively. There is little evidence for a
tight biological timetable of developments of the type that we see in other species. In fact,
children can learn language even when they have been isolated until an age of even 6 years
(Davis, 1947). Throughout the protracted period of human language learning, it is
impossible to find evidence for some discrete moment at which a child sets some crucial
parameter (Hyams, 1995; MacWhinney & Bates, 1989) that can determine the shape of the
native language. Moreover, it is very difficult to use standard experimental methods to
prove that children have acquired some of the more abstract categories and structures
required by Universal Grammar, such as argument chains, empty categories, landing sites,
or dominance relations (Gopnik, 1990; van der Lely, 1994). Despite these empirical
problems, the nativist approach remains dominant for studies that investigate the acquisition
of formal linguistic structures. For a comprehensive survey of nativist approaches to the
acquisition of grammar, the reader may wish to consult Atkinson (1992). Similarly,
Markman (1989) summarizes evidence supporting a nativist approach to the acquisition of
the lexicon.

1
 Models of the Emergence of Language

NATIVISM AND EMERGENTISM

The inability of nativist accounts to provide accurate or testable accounts of the details of
language acquisition has led many language development researchers to explore alternatives
to genetically-wired modules. These alternative frameworks emphasize the ways in which
the formal structures of language emerge from the interaction of social patterns, patterns
implicit in the input, and pressures arising from the biology of the cognitive system. The
emergentist approach to language acquisition views language as a structure arising from
interacting constraints, much as the shape of the coastline arises from pressures exerted by
ocean currents, underlying geology, weather patterns, and human construction. The
formalisms that are used to express these nonlinear patterns of interaction include neural
network modelling (Fausett, 1994), dynamic systems theory (Port & van Gelder, 1995), and
structured approaches such as Optimality Theory {Tesar, in press #7502}. In this chapter, I
examine the extent to which neural network models can account for what we currently know
about the early states of language development.

THE EMERGENCE OF AUDITORY PATTERNS

During the first year of life, the child goes through a complex set of experiences that
lays down an extensive perceptual and motor framework for the learning of the first words.
On the perceptual side, the child actively encodes the raw sound patterns of her native
language, organizing these patterns into types and sequences. At one time, researchers
thought that the learning of perceptual contrasts, such as the ones that allow us to
distinguish between “pin” vs “bin”, occurred during the second year of life when words
are being learned (Jakobson, 1968; Shvachkin, 1948). This picture changed radically when
Eimas, Siqueland, Juszcyk, and Vigorito (1971) showed that the ability to detect the contrast
between /b/ and /p/ is present soon after birth. Initially, it was thought that these abilities
were innate components of a species-specific language gift. However, researchers soon
showed that these abilities are shared with other mammals, such as chinchillas (Kuhl &
Miller, 1975; Kuhl & Miller, 1978) and monkeys (Kuhl & Padden, 1982; Kuhl & Padden,
1983). It now appears that the ability to discriminate the sounds of language is grounded
on raw perceptual abilities of the mammalian auditory system. The sharpness and accuracy
of this ability declines during the first year, as children learn to lump together sounds that
their language treats as equivalent (Polka & Werker, 1994). In effect, children spend much
of the first year of life losing the ability to make contrasts that are not used in the speech
they hear about them. Kuhl (1991) has interpreted these findings as evidence for a
“preceptual magnet” effect. This effect can be understood by imagining that there is a
magnet at the center of each phonemic category that tends to draw in the edges of the
category, thereby shortening the distance and leading to an inability to make fine
distinctions within this compressed region.
Given the fact that children do not yet understand the words they are hearing, their
attentiveness to sound is all the more remarkable. Recent research shows that they are
attending not just to the individual phonemes they hear, but even to longer range patterns,
such as syllabic sequences. For example, Saffran, Aslin, and Newport (1996) have shown
that, when eight-month-old children listen to long sound sequences such as
“dabigogatanagotidabigo”, they tend to pull out repeated sequences such as “dabigo”.
As a result, they tend to listen to these familiar sequences more than to similar new
sequences.
Infants also demonstrate an early attentiveness to the prosodic characteristics of the
language they are hearing. Soon after birth, infants tend to prefer sounds produced by their
own mothers to those produced by other women (DeCasper & Fifer, 1980). They also
prefer their native languages to other languages (Moon, Cooper & Fifer, 1993). These
preferences are probably dependent both on the infant’s ability to detect speaker-specific
vocal characteristics and on the detection of language-specific prosodic patterns. Infants
seem to be sensitive early on to the presence of intonational organization in the language

2
 Models of the Emergence of Language

they listen to. Using the sucking habituation technique, Mandel et al. (1994) showed that
2-month-olds tend to remember word strings better when they are presented with normal
sentence intonation, than when they are presented as unintegrated lists of words with flat
prosody. It appears that stressed intonation may have a particularly important role in
picking up auditory strings. Jusczyk and Pisoni (1995) have shown that children tend to
pick up and learn stressed syllables above unstressed syllables. However, it also appears
that syllables which directly follow after a stressed syllable are also well encoded (Aslin,
Jusczyk & Pisoni, 1997). As a result, many of the first sound sequences recorded by the
child consist of a stressed peak followed by one or two further weak syllables. This pattern
of sound learning has been discussed as a “trochaic bias”. However, it can also be viewed
as emerging from the combination of a bias to track stressed syllables together with a linear
sequence recorder that fires when a stressed syllable is detected.

THE EMERGENCE OF ARTICULATORY PATTERNS

During the first year of life, the infant’s articulatory abilities also progress through
radical transformations. The basic shape of these changes has been documented since the
beginning of the century. We know that children’s first vocalizations include the birth cry,
the pain cry, the hunger cry, and the pleasure cry. These cries are tightly linked to clear
emotional states (Lewis, 1936). By the age of 3 months, children begin a type of social
vocalization known as cooing. Around the age of 6 months, children begin a form of
sound play that we call babbling. At first, babbling involves the sporadic production of a
few simple sounds. These sounds include some strange sounds like clicks that are not
found in the input. However, it is not true that each child babbles all the sounds of all the
world’s languages. Nor is there much evidence for any tight linkage before nine months
between the form of the child’s babbling and the shape of the input language (Atkinson,
MacWhinney & Stoel, 1970; Boysson-Bardies & Vihman, 1991). However, around 11
months, there is increasing evidence for a drift toward the segments and prosody of the
target language (Levitt, Utman & Aydelott, 1993), as the child begins to move into the
period of the first words.
Initially, it appears that auditory and articulatory development proceed as if largely
decoupled. The fact that deaf children babble normally at the age of six months is
particularly strong evidence for this conclusion. Given the fact that the brain areas
subserving audition (inferior parietal, superior temporal) and articulation (motor cortex) are
distant neurologically, this initial decoupling is not too surprising. By the age of nine
months, evidence starts to emerge of a connection between babbling and audition. By this
age, deaf children, who are not receiving adequate auditory feedback, cease babbling.
Normal children start to show the first movement in the prosodic shape of their babbling
toward the forms of the input language.

THE EMERGENCE OF THE FIRST WORDS

One of the most active areas of current research in the child language is the study of
early word learning. Philosophers like Quine (1960) have emphasized the extent to which
word learning needs to be guided by ideas about what might constitute a possible word.
For example, if the child were to allow for the possibility that word meanings might include
disjunctive Boolean predicates (Hunt, 1962), then it might be the case that the word “grue”
would have the meaning “green before the year 2000 and blue thereafter”. Similarly, it
might be the case that the name for any object would refer not to the object itself, but to its
various undetached parts. When one thinks about the word learning task in this abstract
way, it appears to be impossibly hard.

3
 Models of the Emergence of Language

Lexical principles
Markman (1989) and Golinkoff, Mervis, and Hirsh-Pasek (1994) have proposed that
Quine’s problem can be solved by imagining that the child’s search for word meanings is
guided by lexical principles. For example, children assume that words refer to whole
objects, rather than parts of objects. Thus, a child would assume that the word “rabbit”
refers to the whole rabbit and not just some parts of the rabbit. However, there is reason to
believe that such principles are themselves emergent properties of the cognitive system. For
example, Merriman and Stevenson (1997) have argued that the tendency to avoid learning
two names for the same object emerges naturally from the competition (MacWhinney,
1989) between closely-related lexical items.
Another proposed lexical principle is the tendency to focus on object names and
nominal categories over other parts of speech. Gentner (1982) compared the relative use of
nominal terms, predicative terms, and expressive terms in English, German, Japanese, Kaluli,
and Turkish. She found that, in all five languages, words for objects constituted the largest
group of words learned by the child. Like Gentner, Tomasello (1992) has argued that
nouns are easier to “package” cognitively than verbs. Nouns refer to objects that can be
repeatedly touched and located in space, whereas verbs refer to transitory actions that are
often hard to repeat and whose contour varies markedly for different agents. However,
Gopnik and Choi (1990) and Choi and Bowerman (1991) have reported that the first words
of Korean-speaking children include far more verbs than do those of English-speaking
children. Findings of this type indicate that the nominal bias emerges only in languages
that tend to emphasize nouns.
Even in English, we know that children will often treat a new word as a verb or an
adjective (Hall, Waxman & Hurwitz, 1993), because words like “run”, “want”, “hot” and
“good” are included in some of the child’s first words. Children are also quick to pick up
socially-oriented words such as “hi” and “please”. As Bloom, Tinker, and Margulis
(1993) and Vihman and McCune (1994) have argued, the nominal bias is far from a
predominant force, even in English.

Social support
The idea that early word learning depends heavily on the spatio-temporal contiguity
of a novel object and a new name can be traced back to Aristotle, Plato, and Augustine.
Recently, Baldwin (1991; 1989) has shown that children try to acquire names for the objects
that adults are attending to. Similarly, Akhtar, Carpenter, and Tomasello (1996) and
Tomasello and Akhtar (1995) have emphasized the crucial role of mutual gaze between
mother and child in the support of early word learning. Moreover, Tomasello has argued
that human mothers differ significantly from primate mothers in the ways that they
encourage mutual attention during language. While not rejecting the role of social support
in language learning, Samuelson and Smith (in press) have noted that one can also interpret
the findings of Akhtar, Carpenter, and Tomasello in terms of low-level perceptual and
attentional matches that help focus the child’s attention to novel objects to match up with
new words.

Child-based meanings
Several researchers have emphasized the extent to which the shape of the meanings of
the first words is governed by a “child-based agenda” (Mervis, 1984; Slobin, 1985).
Children seem to be particuarly interested in finding ways of talking about their favorite
toys, friends, and foods (Dromi, 1997). They also like to learn words to discuss social
activities and functions. In fact, Ninio and Snow (1988) have argued that the basic
orientation of the child’s first words and early grammar is not towards some objective,
nominal, cognitive reality, but towards the interpersonal world involving people and social
roles.

4
 Models of the Emergence of Language

Overgeneralization and undergeneralization

We can refer to the formation of a link between a particular referent and a new name as
“initial mapping”. This initial mapping is typically fast, sketchy, and tentative. Most
lexical learning occurs after the formation of this initial mapping. As the child is exposed
repeatedly to new instances of an old word, the semantic range of the referent slowly
widens. Barrett (1995), Huttenlocher (1974) and others have viewed this aspect of meaning
growth as “decontextualization”. Harris, Barrett, Jones, and Brookes (1988) have shown
that the initial representations of words contain components that are linked to the first few
contacts with the word in specific episodes or specific contexts. Gradually, the process of
generalization leads to a freeing of the word from irrelevant aspects of the context.
Over time, words develop a separation between a “confirmed core” (1984; 1989) and a
peripheral area of potential generalization. As long as the child sticks closely to attested
instances of the category inside the confirmed core, she will tend to undergeneralize the
word “car”. Anglin (1977) and Dromi (1987) have argued that the frequency of such
undergeneralizations is typically underestimated, because undergeneralizations never lead to
errors. If one does a careful analysis of the range of uses of new words, it appears that
undergeneralization is closer to the rule than the exception. As the confirmed core of the
meaning of a word widens and as irrelevant contextual features are pruned out, the word
begins to take on a radial or prototype form (Lakoff, 1987; Rosch & Mervis, 1975). In the
center of the category, we find the best instances that display the maximum category match.
At the periphery of the category, we find instances whose category membership is unclear
and which compete with neighboring categories (MacWhinney, 1989).
According to the core-periphery model of lexical structure, overgeneralizations arise
from the pressures that force the child to communicate about objects that are not inside any
confirmed core. Frequently enough, children’s overgeneralizations are corrected when the
parent provides the correct name for the object (Brown & Hanlon, 1970). The fact that
feedback is so consistently available for word learning increases our willingness to believe
that the major determinants of word learning are social feedback, rather than innate
constraints or even word learning biases.

The shape of vocabulary growth

Researchers have often noted that the growth of the overall size of the lexicon does not
follow a smooth linear trend. After the child has acquired an initial vocabulary of about 100
words, the learning of new words seems to progress more and more rapidly. This rapid rise
in the size of the vocabulary, which has been called the “vocabulary spurt” (Bates &
Carnevale, 1993; Bloom, 1993), is more evident in some children than in others. However,
Mervis and Bertrand (1994) and Dromi (1997) have shown that accurate detection of the
timing of the vocabulary spurt may require following children well past the first 100 words.
Mervis and Bertrand (1995) argue that the timing of the vocabulary spurt is dependent on
the rate of cognitive development, with slower developers having a later spurt. They further
claim that, before the beginning of the vocabulary spurt, children cannot pick up words
through a few brief exposures. However, recent experimental work by Woodward,
Markman, and Fitzsimmons (1994) and Schafer and Plunkett (in press) has indicated that
infants who have not yet gone through the vocabulary shift are still capable of quick
learning of new words in an experimental context.
Three accounts have been offered for the timing of the vocabulary burst and the causes
of the burst. One account attributes the burst to the development of control over articulatory
representations. Schwartz (1988) and Schwartz and Leonard (1981) have shown that young
children tend to avoid producing difficult phonological forms. Once these output
limitations are surmounted, the child is free to produce words that had been difficult to
produce during earlier periods.
A second account (MacWhinney, 1982) focuses on the role of syntactic patterns in the
learning of new words. Often parents make extensive use of stable syntactic frames such

5
 Models of the Emergence of Language

as “Here’s the nice (toy name)” or “Show me your (body part name)”. Having learned
these frames, children can quickly pick up a large quantity of new words in the context of
each frame. In this way, the vocabulary spurt could be dependent upon syntactic
development. In fact, Bates et al, (1988) reported a correlation of between .70 and .84
between lexical size at 20 months and syntactic abilities at 28 months. This level of
correlation is exactly what is predicted by a model that views lexical learning as facilitated
by the appearance of words in the context of well-understood syntactic frames.
In accord with the Piagetian emphasis on cognitive determination of developmental
stages, a third group of authors has attributed the vocabulary spurt to the underlying growth
in those cognitive capacities (Bloom, 1970; Gopnik & Meltzoff, 1987) that allow children to
understand the meanings of new words. For example, one could argue that 14-month-olds
are not yet ready conceptually to acquire the meanings of comparative adjectives,
conjunctions, abstract nouns, speech act verbs, and superordinates. To be sure, very young
children have not yet acquired complex relational concepts, such as the ones required to
support the learning of form like “nonetheless”, “preamble”, or “next Thursday”
(Kenyeres, 1926). However, attempts to relate overall aspects of linguistic development to
fundamental changes in cognitive development have seldom demonstrated strong linkages
(Corrigan, 1978; Corrigan, 1979). Instead, it appears that the links between cognitive and
lexical development are fragmentary and specific to particular lexical fields (Gopnik &
Meltzoff, 1986).
Each of these three accounts is compatible with attempts (Bates & Carnevale, 1993; van
Geert, 1991) to model vocabulary growth as a dynamic system using logistic growth
functions. The nonlinear effects that emerge during the vocabulary spurt can be viewed as
arising from the dynamic coupling of the lexical system with a quickly developing system
of syntactic patterns, phonological advances, or cognitive advances. As these various
patterns develop, they feed into vocabulary growth in a nonlinear and interactive fashion, as
growth in vocabulary leads to further growth in syntactic structures, at least during the
several months of the vocabulary spurt.

Components of a model of word learning

We are now ready to explore ways in which these facts about lexical development can
be captured in an emergentist model based on neural network theory. The preceding
sections indicate that a neural network model of lexical learning will need six components.
First, it must provide a system for representing auditory contrasts. Second it must be able
to use this system to store frequently heard auditory sequences. Third, the model must be
able to account for the development from unconstrained babbling to the controlled
articulation of real words. Fourth, the model must be able to account for both social and
child-based influences on the meanings underlying the first words. Fifth, the model has to
account for the ways in which parents can provide social scaffolding that focuses children’s
attention on referents. Sixth, the model must be able to account for both fast initial learning
and slow subsequent tuning of the meaning of new words. Seventh, the model must be able
to capture facts about the induction of word meanings from syntactic frames.
Neural network models are systems based on the use of a common language of units,
connections, weights, and learning rules. Within this common language of connectionism,
architectures differ markedly both in their detailed patterns of connectivity and in the
specific rules used for activation and learning. There are now many excellent readable
introductions to the theory and practice of neural network modeling. The reader who is
interested in learning more about the mechanics of this framework may wish to consult
Bechtel and Abrahamsen (1991) or Fausett (1994).

Lexical learning as self-organization

One emergentist framework that allows us to model many of these forces is the self-
organizing feature map (SOFM) architecture of Kohonen (1982) and Miikkulainen and

6
 Models of the Emergence of Language

Dyer (1990; 1991). These self-organizing networks treat word learning as occurring in
maps of connected neurons in small areas of the cortex. Three local maps are involved in
word learning: an auditory map, a concept map, and articulatory maps. Emergent self-
organization on each of these three maps uses the same learning algorithm. Word learning
involves the association of elements between these three maps. What makes this mapping
process self-organizing is the fact that there is no pre-established pattern for these mappings
and no preordained relation between particular nodes and particular feature patterns.
Evidence regarding the importance of syllables in early child language (Bijeljac,
Bertoncini & Mehler, 1993; Jusczyk, Jusczyk, Kennedy, Schomberg & Koenig, 1995)
suggests that the nodes on the auditory map may best be viewed as corresponding to full
syllabic units, rather than separate consonant and vowel phonemes. The recent
demonstration by Saffran et al. (1996) of memory for auditory patterns in four-month-old
infants indicates that children are not only encoding individual syllables, but are also
remembering sequences of syllables. In effect, prelinguistic children are capable of
establishing complete representations of the auditory forms of words. Within the self-
organizing framework, these capabilities can be represented in two alternative ways. One
method uses a slot-and-frame featural notation from MacWhinney, Leinbach, Taraban, and
McDonald (1989). An alternative approach views the encoding as a temporal pattern that
repeatedly accesses a basic syllable map. A lexical learning model developed by Gupta and
MacWhinney (1997) uses serial processes to control word learning. This model couples a
serial order mechanism known as an “avalanche” (Grossberg, 1978) with a lexical feature
map model. The avalanche controls the order of syllables within the word. Each new word
is learned as a new avalanche.
The initial mapping process involves the association of auditory units to conceptual
units. Initially, this learning links concepts to auditory images (Naigles & Gelman, 1995;
Reznick, 1990). For example, the 14-month-old who has not yet produced the first word,
may demonstrate an understanding of the word “dog” by turning to a picture of a dog,
rather than a picture of a cat, when hearing the word “dog”. It is difficult to measure the
exact size of this comprehension vocabulary in the weeks preceding the first productive
word, but it is probably at least 20 words in size.
In the self-organizing framework, the learning of a word is viewed as the emergence of
an association between a pattern on the auditory map and a pattern on the concept map
through Hebbian learning (Hebb, 1949; Kandel & Hawkins, 1992). When the child hears a
given auditory form and sees an object at the same time, the coactivation of the neurons that
respond to the sound and the neurons that respond to the visual form produces an
association across a third pattern of connections which maps auditory forms to conceptual
forms. Initially, the pattern of these interconnections is unknown, because the relation
between sounds and meanings is arbitrary (de Saussure, 1966). This means that the vast
majority of the many potential connections between the auditory and conceptual maps will
never be used, making it a very sparse matrix (Kanerva, 1993). In fact, it is unlikely that all
units in the two maps are fully interconnected (Shrager & Johnson, 1995). In order to
support the initial mapping, some researchers (Schmajuk & DiCarlo, 1992) have suggested
that the hippocampus may provide a means of maintaining the association until additional
cortical connections have been established. As a result, a single exposure to a new word is
enough to lead to one trial learning. However, if this initial association is not supported by
later repeated exposure to the word in relevant social contexts, the child will no longer
remember the word.

Word learning and working memory

The account of word learning we have been examining so far has focused on the
learning of the auditory form of the word. In the infant, learning of the articulatory form is
typically more delayed. For adults, the task of articulating a newly perceived word is a
simple one. However, for the child in the second year of life, matching up articulations to

7
 Models of the Emergence of Language

auditions is a major challenge. The simple control of the articulatory system is still a major
challenge for the two-year-old. Apart from this, the child have must acquire a mapping from
individual auditory features to articulatory gestures, and must also encode the sequence and
prosodic contour of each of the syllables in the word. Just like the learning of auditory
sequences requires the mediation of memory systems, the learning of articulatory sequences
may involve support from rehearsal loops or hippocampal systems.
Models of word learning in adults (Burgess & Hitch, 1992; Grossberg, 1978;
Grossberg, 1987; Houghton, 1990)} have tended to emphasize the role of working memory.
Gupta and MacWhinney (1997) have shown that a model based on the encoding of syllable
strings for output phonology in avalanches does a good job of accounting for a wide variety
of well-researched phenomena in the literature on word learning, immediate serial recall,
interference effects, and rehearsal in both adults and children (Gathercole & Baddeley,
1993).

The organization of semantic fields

Parallel with the growth of the auditory map, the child is working on the
development of an extensive system for conceptual coding. As we have noted, studies of
concept development in the preverbal infant (Piaget, 1954; Stiles-Davis, Sugarman & Nass,
1985; Sugarman, 1982) indicate that the child comes to the language learning task already
possessing a fairly well-structured coding of the basic objects in the immediate
environment. Children treat objects such as dogs, plates, chairs, cars, baby food, water, balls,
and shoes as fully structured separate categories (Mervis, 1984). They also show good
awareness of the nature of particular activities such as falling, bathing, eating, kissing, and
sleeping.
Like auditory categories, these basic conceptual categories can be represented in self-
organizing feature maps. Schyns (1991) applied a self-organizing feature map to the task
of learning three competing categories with prototype structures. The individual exemplars
of each category were derived from geometric patterns that were blurred by noise to create a
prototype structure, although the actual prototypes were never displayed. The simulations
showed that the network could acquire human-like use of the categories. When presented
with a fourth new word that overlapped with one of the first three words, the system broke
off some of the territory of the old referent to match up with the new name. This
competitive behavior seems to reflect the process of competition between old words and new
words discussed for children’s word learning by Markman (1989), Clark (1987), and
MacWhinney (1989).
Another simulation of meaning development by Li and MacWhinney (1996) used a
standard backpropagation architecture to model the learning of reversive verbs that used the
prefix “un-” as in “untie” or “dis-” as in “disavow”. The model succeeded in
capturing the basic developmental stages for reversives reported by Bowerman (1982) and
Clark, Carpenter, and Deutsch (1995). In particular, the model was able to produce
overgeneralization errors such as “*unbreak” or “*disbend”. The network’s
performance was based on its internalization of what Whorf (1938; 1941) called the
“cryptotype” for the reversive which involved a “covering, enclosing, and surface-attaching
meaning” that is present in a word like “untangle”, but absent in a form such as
“*unbreak”. Whorf viewed this category as a prime example of the ways in which
language reflects and possibly shapes thought.
A similar neural network model of the learning of fine differences in the meaning of the
word “over” was developed by Harris (1990; 1994). The Harris model is capable of
taking new input test sentences of the type “the pin rolled over the table” and deciding on
the basis of past learning that the meaning involved is “across”, rather than “covering” or
“above”. It does this only on the basis of the cooccurrence patterns of the words involved,
rather than on information from their individual semantics. Thus, it learns that combinations
like “ball”, “roll”, and “table” tend to activate “across” without regard to facts such as

8
 Models of the Emergence of Language

knowing that balls are round and can roll or knowing that tables are flat and that rolling
involves movement.

THE EMERGENCE OF INFLECTIONAL MARKING

One of the most active areas in recent work on language acquisition has been the study
of the child’s learning of inflectional marking. In English, inflections are short suffixes that
occur at the ends of words. For example, the word “dogs” has a final /s/ suffix that marks
the fact that it is plural. There are now well over 30 empirical studies and simulations
investigating the learning of inflectional marking. The majority of work on this topic has
examined the learning of English verb morphology with a particular focus on the English
past tense. These models are designed to learn irregular forms such as “went” or “fell”,
as well as regular past tense forms such as “wanted” and “jumped”. Other areas of
current interest include German noun declension, Dutch stress placement, and German
participle formation. Although the learning of inflectional markings is a relatively minor
aspect of language learning, our ability to quantify this process has made it an important
testing ground not only for the study of child language, but for developmental psychology
and cognitive science more generally.

A sample model for inflectional learning

To illustrate how connectionist networks can be used to study the learning of
inflectional morphology, let us take as an example the model of German gender learning
developed by MacWhinney, Leinbach, Taraban, and McDonald (1989). This model was
designed to explain how German children learn to select one of the six different forms of
the German definite article. In English we have a single word “the” that serves as the
definite article. In German, the article can take the form “der”, “die”, “das”, “des”,
“dem”, or “den”. Which of the six forms of the article should be used to modify a given
noun in German depends on three additional features of the noun: its gender (masculine,
feminine, or neuter), its number (singular or plural), and its role within the sentence (subject,
possessor, direct object, prepositional object, or indirect object). To make matters worse,
assignment of nouns to gender categories is often quite nonintuitive. For example, the word
for “fork” is feminine, the word for “spoon” is masculine, and the word for “knife” is
neuter. Acquiring this system of arbitrary gender assignments is particularly difficult for
adult second language learners. Mark Twain expressed his consternation at this aspect of
German in a treatise entitled “The aweful German language” (Twain, 1935) in which he
accuses the language of unfairness and capriciousness in its treatment of young girls as
neuter, the sun as feminine, and the moon as masculine. Along a similar vein, Maratsos and
Chalkley (1980) argued that, because neither semantic nor phonological cues can predict
which article accompanies a given noun in German, children could not learn the language by
relying on simple surface cues.
Although these relations are indeed complex, MacWhinney et al. show that it is possible
to construct a connectionist network that learns the German system from the available cues.
The MacWhinney et al. model, like most current connectionist models, involves a level of
input units, a level of hidden units, and a level of output units (Figure 1). Each of these
levels or layers contains a number of discrete units or nodes. For example, in the
MacWhinney et al. model, the 35 units within the input level represent features of the noun
that is to be modified by the article. Each of the two hidden unit levels includes multiple
units that represent combinations of these input-level features. The six output units
represent the six forms of the German article.

9
 Models of the Emergence of Language

OUTPUT UNITS der die das des dem den

• • • • • •

200 hidden
7 units

HIDDEN UNITS

20200
gender/
hidden
10 case units
number units

INPUT UNITS 143 phonological 5 meaning 17 case cues 11 phono

Figure 1: A network model of the acquisition of German declensional marking

As noted above, a central feature of such connectionist models is the very large number
of connections among processing units. As shown in Figure 1, each input-level unit is
connected to first-level hidden units; each first-level hidden unit is connected to second-level
hidden units; and each second-level hidden unit is connected to each of the six output units.
None of these hundreds of individual node-to-node connections is illustrated in Figure 1,
since graphing each individual connection would lead to a blurred pattern of connecting
lines. Instead a single line is used to stand in place of a fully interconnected pattern between
levels. Learning is achieved by repetitive cycling through three steps. First, the system is
presented with an input pattern that turns on some, but not all of the input units. In this
case, the pattern is a set of sound features for the noun being used. Second, the activations
of these units send activations through the hidden units and on to the output units. Third,
the state of the output units is compared to the correct target and, if it does not match the
target, the weights in the network are adjusted so that connections that suggested the correct
answer are strengthened and connections that suggested the wrong answer are weakened.
MacWhinney et al. tested this system’s ability to master the German article system by
repeatedly presenting 102 common German nouns to the system. Frequency of presentation
of each noun was proportional to the frequency with which the nouns are used in German.
The job of the network was to choose which article to use with each noun in each particular
context. After it did this, the correct answer was presented, and the simulation adjusted
connection strengths so as to optimize its accuracy in the future. After training was
finished, the network was able to choose the correct article for 98 percent of the nouns in the
original set.
To test its generalization abilities, we presented the network with old nouns in new case
roles. In these tests, the network chose the correct article on 92 percent of trials. This type
of cross-paradigm generalization is clear evidence that the network went far beyond rote
memorization during the training phase. In fact, the network quickly succeeded in learning
the whole of the basic formal paradigm for the marking of German case, number, and
gender on the noun. In addition, the simulation was able to generalize its internalized
knowledge to solve the problem that had so perplexed Mark Twain -- guessing at the gender
of entirely novel nouns. The 48 most frequent nouns in German that had not been included

10
 Models of the Emergence of Language

in the original input set were presented in a variety of sentence contexts. On this completely
novel set, the simulation chose the correct article from the six possibilities on 61 percent of
trials, versus 17 percent expected by chance. Thus, the system’s learning mechanism,
together with its representation of the noun's phonological and semantic properties and the
context, produced a good guess about what article would accompany a given noun, even
when the noun was entirely unfamiliar.
The network’s learning paralleled children’s learning in a number of ways. Like real
German-speaking children, the network tended to overuse the articles that accompany
feminine nouns. The reason for this is that the feminine forms of the article have a high
frequency, because they are used both for feminines and for plurals of all genders. The
simulation also showed the same type of overgeneralization patterns that are often
interpreted as reflecting rule use when they occur in children’s language. For example,
although the noun Kleid (which means clothing) is neuter, the simulation used the initial
“kl” sound of the noun to conclude that it was masculine. Because of this, it invariably
chose the article that would accompany the noun if it were masculine. Interestingly, the same
article-noun combinations that are the most difficult for children proved to be the most
difficult for the simulation to learn and to generalize to on the basis of previously learned
examples.
How was the simulation able to produce such generalization and rule-like behavior
without any specific rules? The basic mechanism involved adjusting connection strengths
between input, hidden, and output units to reflect the frequency with which combinations of
features of nouns were associated with each article. Although no single feature can predict
which article would be used, various complex combinations of phonological, semantic, and
contextual cues allow quite accurate prediction of which articles should be chosen. This
ability to extract complex, interacting patterns of cues is a characteristic of the particular
connectionist algorithm, known as back-propagation, that was used in the MacWhinney et
al. simulations. What makes the connectionist account for problems of this type
particularly appealing is the fact that an equally powerful set of production system rules for
German article selection would be quite complex (Mugdan, 1977) and learning of this
complex set of rules would be a challenge in itself.

Cues vs. rules

The central issue being addressed in the study of the learning of inflectional markings is
whether one can model this process without using formal rules. Rumelhart and McClelland
(1986) were the first to provide a demonstration of how rules could emerge from the
behavior of neural networks without being explicitly learned. Conceding that irregular forms
are produced by connectionist networks, Pinker (1991) nonetheless argues that regular
forms are produced by a regular rule. This dual-route model echoes an earlier account by
MacWhinney (1978) and related dual-route models in the study of reading by Coltheart,
Curtis, Atkins, and Waller (1993)
These attempts to preserve a role for rules in human cognition have run into problems
with the fact that even the most regular patterns or “rules” display phonological
conditioning and patterns of gradience (Bybee, in press) of the type that are well captured in
a connectionist network. Moreover, the existence of differences between regular and
irregular processing does not, in itself, provide strong evidence for the existence of rules.
Kawamoto (1994) has shown that regular and irregular forms display quite different
activation patterns, even within a homogeneous neural network. Therefore, differences in
the processing of regular and irregular verbs that have recently been demonstrated through
neural imaging work (Jaeger, Lockwood, Kemmerer, Van Valin & Murphy, 1996; Weyerts,
Penke, Dohrn, Clahsen & Münte, 1996) do not provide strong evidence for the separate
existence of rule system.

11
 Models of the Emergence of Language

U-shaped learning
A major shortcoming of nearly all connectionist models of inflectional learning has been
their inability to capture the patterns of overgeneralization and recovery from
overgeneralization that have been called “u-shaped” learning. In u-shaped learning, the
child begins by correctly producing an irregularly inflected form such as “went”. Next,
under the pressure of the general pattern, the child produces the overgeneralized form
“goed”. Finally, the child recovers from overgeneralization and returns to saying “went”.
Some writers have mistakenly assumed that this type of u-shaped learning applies across all
verbs to create three major periods in language learning. However, empirical work by
Marcus, Ullman, Pinker, Hollander, Rosen, and Xu (1992) has shown that strong u-shaped
learning patterns occur only for some verbs and only for some children.
The modelling of even these weaker u-shaped patterns has proven difficult for neural
networks. In order to correctly model the child’s learning of inflectional morphology,
models must go through a period of virtually error free learning of irregulars, followed by a
period of learning of regulars accompanied by the first overregularizations (Marcus et al.,
1992). No current model consistently displays all of these features in exactly the right
combination. MacWhinney (1997) has argued that models that rely exclusively on
backpropagation will never be able to display the correct combination of developmental
patterns and that a two-process connectionist approach may be needed (Kawamoto, 1994;
Stone, 1994). The basic process is one that learns new inflectional formations, both regular
and irregular, as items in self-organizing feature maps. The secondary process is a network
that generalizes the information inherent in feature maps to extract secondary productive
generalizations. Unlike Pinker’s dual-route account, this proposed account works on a
uniform underlying connectionist architecture without relying on formal, symbolic linguistic
rules.

The role of semantic factors

The first attempts to model morphological learning focused exclusively on the use of
phonological features as both input and output. However, it is clear that the formation of
past tense forms must also involve semantic factors. In English, the use of semantic
information is associated with the irregular patterns of inflection. The idea is that, because
we cannot access “went” by combining “go” and “-ed”, it might be that we can access it
directly by a semantic route. Of course, this idea is much like that underlying the dual-route
theory. In German gender, the role of semantic information is much clearer. Köpcke and
Zubin (Köpcke, 1994; Köpcke & Zubin, 1983; Köpcke & Zubin, 1984; Zubin & Köpcke,
1981; Zubin & Köpcke, 1986) have shown that a wide variety of both phonological and
semantic factors are used in predicting the gender of German nouns and their plural. Some
of the features involved include: alcoholic beverages, superordinates, inherent biological
gender, gem stones, body parts, rivers inside Germany, and light vs heavy breezes.
Simulations by Cottrell and Plunkett (1991) and Gupta and MacWhinney (1992) have
integrated semantic and phonological information in various ways. However, a better
understanding of the ways in which semantic factors interact during word formation will
require a more extensive modeling of lexical items and semantic features.

Extensions of irregular patterns to new words

Extending earlier work by Bybee and Slobin (1982) with older children, Prasada and
Pinker (1993) examined the abilities of adult native English speakers to form the past tense
for nonsense words like “plink”, “plup”, or “ploth”. They found that, the further the
word diverged from the standard phonotactic rules for English verbs, the more likely the
subjects were to form the past tense by just attaching the regular “-ed” suffix. Ling and
Marinov (1993) noted that the original verb-learning model developed by Rumelhart and
McClelland (1987) failed to match these new empirical data, largely because of its tendency
to overapply irregular patterns. To correct this problem, Ling and Marinov created a non-

12
 Models of the Emergence of Language

connectionist symbolic pattern associator which did a better job modeling the Prasada and
Pinker data. However, MacWhinney (1993a) found that the network model of
MacWhinney and Leinbach (1991) worked as well as Ling and Marinov’s symbolic model
in terms of matching up to the Prasada and Pinker generalization data.

Inflections and the logical problem of language acquisition

In the network we have been discussing, a single lexical feature map can produce both a
rote form like “went” and a productive form like “*goed”. The fact that both can be
produced in the same lexical feature map allows us to develop a general solution to the
“logical problem of language acquisition” (Baker & McCarthy, 1981; Gleitman, 1990;
Gleitman, Newport & Gleitman, 1984; Morgan & Travis, 1989; Pinker, 1984; Pinker, 1989;
Wexler & Culicover, 1980). The logical problem of language acquisition arises from the
(incorrect) assumption that recovery from overgeneralization must depend on corrective
feedback. Because corrective feedback is seldom available for grammatical patterns (as
opposed to lexical and semantic patterns) it can be shown that language learning from input
data is impossible. Therefore, it is argued, the acquisition of grammar constitutes a logical
problem and requires the postulation of innate constraints on the form of language. The
solution to this problem proposed by MacWhinney (1993b) focuses on the competition
between regular and irregular forms. In the case of the competition between “went” and
“*goed”, we expect “went” to become solidified over time because of its repeated
occurrence in the input. The form “*goed”, on the other hand, is supported only by the
presence of the -ed form. Figure 2 illustrates this competition.

go + PAST

went competition go + ed

emergent
episodic lexical
support properties

Figure 2: Competition between episodic and combinatorial knowledge

This particular competition is an example of what Baker (1979) calls a “benign

exception to the logical problem”. The exception is considered benign because the child
can learn to block overgeneralization by assuming that there is basically only one way of
saying “went”. This Uniqueness Constraint is thought to distinguish benign and non-
benign exceptions to the logical problem. However, from the viewpoint of the Competition
Model account we are constructing here, all exceptions are benign.
The basic idea here is that, when a child overgeneralizes and produces “*goed”, the
system itself contains a mechanism that eventually forces recovery. Thus, the solution to the
logical problem of language acquisition emerges from the competition between alternative
competing expressions. One of these forms receives episodic support from the actual
linguistic input. This episodic support grows slowly over time. The other form arises
productively from the operation of analogistic pressures. When episodic support does not
agree with these analogistic pressures, the episodic support eventually comes to dominate

13
 Models of the Emergence of Language

and the child recovers from the overgeneralization. This is done without negative evidence,
solely on the basic of positive support for the form receiving episodic confirmation.

THE EMERGENCE OF SYNTACTIC PATTERNS

Induction from syntactic frames
Many aspects of word meaning can be acquired from individual words without
relying on the role that the word plays in sentences. However, other aspects of meaning
require close attention to the ways in which words are combined. In an early demonstration
of these effects, Katz, Baker, and Macnamara (1974) gave children a human and a non-
human figure and asked them either to “Show me the zav” or to “Show me Zav.” When
“zav” was treated syntactically as a proper noun by omitting the definite article, two-year-
olds tended to hand the experimenter the figure of a doll. When “zav” was treated
syntactically as a common noun by use of the definite article, children tended to hand the
experimenter the non-human figure. In this way, even children as young as 20 months of
age showed how syntactic context can serve as a powerful guide to word learning.
Similar effects have now been demonstrated for a wide variety of syntactic
constructions. Brown (1957) found that children could use a sentence frame like “in this
picture you see sibbing” to infer that “sib” is a verb. Carey (1978) and Landau, Smith,
and Jones (1992) found that when asked to choose “not the red one, but the xerillium one”,
children would assume that “xerillium” was a color name. Golinkoff, Hirsh-Pasek,
Cauley, and Gordon (1987) showed children movies in which Big Bird and Cookie
Monster were either turning separately or turning each other. When 27-month-old children
heard “Big Bird is gorping with Cookie Monster”, they tended to look at the video with
both characters turning separately. However, when they heard “Big Bird is gorping Cookie
Monster”, they tended to look at the video with Big Bird turning Cookie Monster. These
results indicate that children can use the transitive syntactic frame to induce some aspects of
the meaning of the new word “gorp”. Similarly, if we ask children to “please repulsate
Big Bird the banana”, they will assume that “repulsate” is a verb of transfer that permits a
double-object construction. However, if we tell children to “please repulsate the tub with
water”, they will assume that “repulsate” is a verb like “fill” that takes a goal as direct
object and a transferred object in an instrumental phrase.
Gleitman (1990) has argued that the meanings of words can be induced largely on the
basis of this syntactic information. In addition, she has argued that certain aspects of
argument structure can only be reliably induced from syntactic frames. However, P. Bloom
(1994) has argued that representations acquired in this way would be incomplete. Since
children have access to both semantic and syntactic information, it seems likely that both
types of information are used whenever they are reliable. In a detailed computational model
of verb argument frame induction, Siskind (1996) has shown that, if the child has access to
a basic situational representation along with surface cooccurence information, the argument
frames of verbs, which are in fact the backbone of the language (Goldberg, 1995;
MacWhinney, 1988; Pinker, 1989), can be learned easily even from fairly noisy input data.

The emergence of parts of speech

Psycholinguists working in the standard symbolic tradition (Chomsky, 1965; Fodor &
Pylyshyn, 1988; Lachter & Bever, 1988) have pointed to the learning of syntax as a
quintessential problem for connectionist approaches. One of the key abilities involved in
the learning of syntax is the abstraction of syntactic classes or “parts of speech”, such as
nouns, verbs, or prepositions. In the theory of universal grammar, these categories are
innately given. However, their actual realization differs so much from language to language
that it makes sense to explore accounts that induce these categories from the input data.
Bates and MacWhinney (1982) and MacWhinney (1988) emphasize the extent to
which the assignment of words to syntactic classes is heavily dependent upon semantic

14
 Models of the Emergence of Language

category structure. Although not all nouns are objects, the best or most prototypical nouns
all share this feature. As the category of “noun” radiates out (Lakoff, 1987), non-central
members start to share fewer of the core features of the prototype. Maratsos and Chalkley
(1980) point out that words like “justice” and “lightning” are so clearly non-objects that
their membership in the class of nouns cannot be predicted from their semantic status and
can only be inferred from the fact that the language treats them as nouns. Although Bates
and MacWhinney (1982) and Maratsos and Chalkley (1980) staked out strongly
contrasting positions on this issue, each of the approaches granted the possibility that both
cooccurence and semantic factors play a major role in the emergence of the parts of speech.
At this point, language researchers are primarily interested in exploring detailed models
that show exactly how the parts of speech and argument frames can be induced. Elman
(1993) has presented a connectionist model that does just this. The model relies on a
recurrent architecture of the type presented in Figure 3. This model takes the standard
three-layer architecture of pools A, B, and C and adds a fourth input pool D of context units
which has recurrent connections to pool B. Because of the recurrent or bidirectional
connections between B and D, this architecture is know as “recurrent backpropagation”.

A pr edict category

B int er nal st at e

C D

cur rent wor d pr evious st at e

Figure 3: A recurrent network

A recurrent backpropagation network encodes changes over time by storing information

regarding previous states in the pool of units labeled as D. Consider how the network deals
with the processing of a sentence such as “Mommy loves Daddy”. When the first word
comes in, pool C is activated and this activation is passed on to pool B and then pools A and
D. The complete state of pool B at Time 1 is stored in pool D. The activation levels in
pool D are preserved, while pools A, B, and C are set back to zero. At time 2 the networks
hears the word “love” and a new pattern of activations is established on pool C. These
activations are passed on to pools B, C, and D. However, because pool D has stored
activations from the previous word, the new state is blended with the old state and pool C
comes to represent aspects of both “Mommy” and “love”.
Processing in a network of this type involves more than just storage of a superficial
sequence of words or sounds. For example, in the simulations of sentence processing
developed by Elman (1993), the output units are trained to predict the identity of the next
word. In order to perform in this task, the network needs to implicitly extract part-of-
speech information from syntactic cooccurrence patterns. Alternatively, the output units can
be used to represent comprehension decisions, as in the model of MacWhinney (1997). In

15
 Models of the Emergence of Language

that model, part-of-speech information is assumed and the goal of the model is to select the
agent and the patient using a variety of grammatical and pragmatic cues.
The training set for the model consists of dozens of simple English sentences such as
“The big dog chased the girl.” By examining the weight patterns on the hidden units in the
fully trained model, Elman showed that the model was conducting implicit learning of the
parts of speech. For example, after the word “big” in our example sentence, the model
would be expecting to activate a noun. The model was also able to distinguish between
subject and object relative structures, as in “the dog the cat chased ran” and “the dog that
chased the cat ran”.

The emergence of argument structures

The machinery governing word combinations depends not only on part of speech
information, but also on information regarding detailed aspects of argument structure.
Consider the use of verbs like “pour” and “fill”. Bowerman (1988) discusses cases in
which the child says “I poured the tub with water” instead of “I pouredwater into the tub”
and “I filled water into the tub” instead of “I filled the tub with water.” We can describe
these errors by saying that the child has overgeneralized the “pour” pattern to the word
“fill” or overgeneralized the “fill” pattern to the word “pour”. In order to avoid these
overgeneralizations and to recover from them once they are made, children have to organize
verbs into semantic fields. Extending the self-organizing topological network approach we
examined earlier, we can model this process by building a network in which the use of the
pattern “V N with N” is correlated with the semantic features of words like “fill”, “stuff”,
and “load”, and in which the use of the pattern “V N into N” is correlated with the
semantic features of words like “fill”, “paint”, “cover”, and “load”. Because of
network of this type uses semantic features to achieve a separation on the argument frame
map, it is able to implement both the semantic proposals of Bates and MacWhinney (1982)
and the cooccurrence proposals of Maratsos and Chalkley (1980).

CONCLUSION
In this chapter, we have seen how neural network models can help us organize our
growing understanding of auditory, articulatory, lexical, inflectional, and syntactic
development. There are many aspects of language development to which these models have
not yet been applied. We do not yet have models that can learn to control sociolinguistic
relations, conversational patterns, narrative structures, intonational contours, and gestural
markings. Even in the areas to which they have been applied, emergentist models are
limited in many ways. The treatment of the more complex aspects of syntax remains
unclear, the modelling of lexical extensions is still quite primitive, and the development of
the auditory and articulatory systems is not yet sufficiently grounded in physiological and
neurological facts. Despite these limitations, we can see that, by treating language learning
as an emergent process, these models have succeeded in providing an exciting new
perspective on questions about language learning that have intrigued scholars for centuries.

16
 Models of the Emergence of Language

LITERATURE CITED

Akhtar, N., Carpenter, M., & Tomasello, M. (1996). The role of discourse novelty in early
word learning. Child Development, 62, 635-645.
Allen, R., & Gardner, B. (1969). Teaching sign language to a chimpanzee. Science, 165,
664-672.
Anglin, J. M. (Ed.). (1977). Word, object, and conceptual development. New York: Norton.
Aslin, R., Jusczyk, P., & Pisoni, D. (1997). Speech and auditory processing during infancy:
Constraints on and precursors to language. In D. Kuhn & R. Siegler (Eds.), Handbook
of child psychology. Volume 2, . New York: Wiley.
Atkinson, K., MacWhinney, B., & Stoel, C. (1970). An experiment on the recognition of
babbling. Papers and Reports on Child Language Development, 5, 1-8.
Atkinson, M. (1992). Children's syntax. Oxford: Blackwells.
Baker, C. L. (1979). Syntactic theory and the projection problem. Linguistic Inquiry, 10,
533-581.
Baker, C. L., & McCarthy, J. J. (Eds.). (1981). The logical problem of language acquisition.
Cambridge: MIT Press.
Baldwin, D. A. (1991). Infants' contribution to the achievement of joint reference. Child
Development, 62, 875-890.
Baldwin, D. A., & Markman, E. M. (1989). Establishing word-object relations: A first step.
Child Development, 60, 381-398.
Barrett, M. (1995). Early lexical development. In P. Fletcher & B. MacWhinney (Eds.),
Handbook of Child Language, . Oxford: Basil Blackwell.
Bates, E., Bretherton, I., & Snyder, L. (1988). From first words to grammar: Individual
differences and dissociable mechanisms. Cambridge, MA: Cambridge University Press.
Bates, E., & Carnevale, G. (1993). New directions in research on language development.
Developmental Review, 13, 436-470.
Bates, E., & MacWhinney, B. (1982). Functionalist approaches to grammar. In E. Wanner
& L. Gleitman (Eds.), Language acquisition: The state of the art, (pp. 173-218). New
York: Cambridge University Press.
Bechtel, W., & Abrahamsen, A. (1991). Connectionism and the mind: An introduction to
parallel processing in networks. Cambridge, MA: Basil Blackwell.
Bickerton, D. (1990). Language and species. Chicago: Chicago University Press.
Bijeljac, B., R., Bertoncini, J., & Mehler, J. (1993). How do four-day-old infants categorize
multisyllabic utterances? Developmental Psychology, 29, 711-721.
Bloom, L. (1970). Language development: Form and function in emerging grammars.
Cambridge, MA: MIT Press.
Bloom, L. (1993). The transition from infancy to language: Acquiring the power of
expression. Cambridge: Cambridge University Press.
Bloom, L., Tinker, E., & Margulis, C. (1993). The words children learn: Evidence against a
noun bias in early vocabularies. Cognitive Psychology, 8, 431-450.
Bloom, P. (1994). Overview: Controversies in language acquisition. In P. Bloom (Ed.),
Language acquisition: Core readings, (pp. 5-48). Cambridge, MA: MIT Press.
Bowerman, M. (1982). Reorganizational processes in lexical and syntactic development. In
E. Wanner & L. Gleitman (Eds.), Language acquisition: The state of the art, (pp. 319-
346). New York: Cambridge University Press.
Bowerman, M. (1988). The "no negative evidence" problem. In J. Hawkins (Ed.),
Explaining language universals, (pp. 73-104). London: Blackwell.

17
 Models of the Emergence of Language

Boysson-Bardies, B., & Vihman, M. M. (1991). Adaption to language: Evidence from

babbling and first words in four languages. Language, 67, 297-320.
Brown, R. (1957). Linguistic determinism and the part of speech. Journal of Abnormal and
Social Psychology, 55, 1-5.
Brown, R., & Hanlon, C. (1970). Derivational complexity and order of acquisition in child
speech. In J. R. Hayes (Ed.), Cognition and the development of language, (pp. 11-54).
New York: Wiley.
Burgess, N., & Hitch, G. (1992). Toward a network model of the articulatory loop. Journal
of Memory and Language, 31, 429-460.
Bybee, J. (in press). Regular morphology and the lexicon. Cognition.
Bybee, J. L., & Slobin, D. I. (1982). Rules and schemas in the development and use of the
English past. Language, 58, 265-289.
Carey, S. (1978). The child as word learner. In M. Halle, J. Bresnan, & G. Miller (Eds.),
Linguistic theory and psychological reality, (pp. 264-293). Cambridge, MA: MIT Press.
Choi, S., & Bowerman, M. (1991). Learning to express motion events in English and
Korean: The influence of language-specific lexicalization patterns. Cognition, 41, 83-
121.
Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press.
Chomsky, N. (1980). Rules and Representations. New York: Columbia University Press.
Clark, E. (1987). The Principle of Contrast: A constraint on language acquisition. In B.
MacWhinney (Ed.), Mechanisms of Language Acquisition, (pp. 1-34). Hillsdale, NJ:
Lawrence Erlbaum Associates.
Clark, E., Carpenter, K., & Deutsch, W. (1995). Reference states and reversals: Undoing
actions with verbs. Journal of Child Language, 22, 633-652.
Coltheart, M., Curtis, B., Atkins, P., & Haller, M. (1993). Models of reading aloud: Dual-
route and parallel distributed processing approaches. Psychological Review, 100, 589-
608.
Corrigan, R. (1978). Language development as related to stage 6 object permanence
development. Journal of Child Language, 5, 173-189.
Corrigan, R. (1979). Cognitive correlates of language: Differential criteria yield differential
results. Child Development, 50, 617-631.
Cottrell, G., & Plunkett, K. (1991). Learning the past tense in a recurrent network:
Acquiring the mapping from meaning to sounds. In K. Hammond & D. Gentner (Eds.),
Proceedings of the Thirteenth Annual Conference of the Cognitive Science Society, (pp.
109-115). Hillsdale, NJ: Lawrence Erlbaum.
Davis, K. (1947). Final note on a case of extreme social isolation. American Journal of
Sociology, 52, 432-437.
de Saussure, F. (1966). Course in general linguistics. New York: McGraw-Hill.
DeCasper, A. J., & Fifer, W. P. (1980). Of human bonding: Newborns prefer their mothers'
voices. Science, 208, 1174-1176.
Dromi, E. (1987). Early lexical development. New York: Cambridge University Press.
Dromi, E. (1997). Early lexical development. In M. Barrett (Ed.), The development of
language, . London: UCL Press.
Eimas, P. D., Siqueland, E. R., Jusczyk, P., & Vigorito, J. (1971). Speech perception in
infants. Science, 171, 303-306.
Elman, J. (1993). Incremental learning, or the importance of starting small. Cognition, 48,
71-99.
Fausett, L. (1994). Fundamentals of neural networks. Englewood Cliffs, NJ: Prentice Hall.
Fodor, J. (1983). The modularity of mind: An essay on faculty psychology. Cambridge,
Mass.: M. I. T. Press.
Fodor, J., & Pylyshyn, Z. (1988). Connectionism and cognitive architecture: A critical
analysis. Cognition, 28, 3-71.
Gathercole, V., & Baddeley, A. (1993). Working memory and language. Hillsdale, NJ:
Lawrence Erlbaum Associates.

18
 Models of the Emergence of Language

Gentner, D. (1982). Why nouns are learned before verbs: Linguistic relativity versus natural
partitioning. In S. Kuczaj (Ed.), Language development: Language, culture, and
cognition, (pp. 301-334). Hillsdale, NJ: Lawrence Erlbaum.
Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3-
55.
Gleitman, L. R., Newport, E. L., & Gleitman, H. (1984). The current status of the motherese
hypothesis. Journal of Child Language, 11, 43-79.
Goldberg, A. (1995). Constructions. Chicago: University of Chicago Press.
Golinkoff, R., Hirsh-Pasek, K., Cauley, K., & Gordon, L. (1987). The eyes have it: lexical
and syntactic comprehension in a new paradigm. Journal of Child Language, 14, 23-46.
Golinkoff, R. M., Mervis, C. B., & Hirsh-Pasek, K. (1994). Early object labels: The case
for a developmental lexical principles framework. Journal of Child Language, 21, 125-
155.
Gopnik, A., & Choi, S. (1990). Do linguistic differences lead to cognitive differences? A
crosslinguistic study of semantic and cognitive development. First Language, 10, 199-
215.
Gopnik, A., & Meltzoff, A. (1987). The development of categorization in the second year
and its relation to the other cognitive and linguistic developments. Child Development,
58, 1523-1531.
Gopnik, A., & Meltzoff, A. N. (1986). Relations between semantic and cognitive
development in the one-word stage: The Specificity Hypothesis. Child Development, 57,
1040-1053.
Gopnik, M. (1990). Feature blindness: A case study. Language Acquisition, 1, 139-164.
Grossberg, S. (1978). A theory of human memory: Self-organization and performance of
sensory-motor codes, maps, and plans. Progress in Theoretical Biology, 5, 233-374.
Grossberg, S. (1987). Competitive learning: From interactive activation to adaptive
resonance. Cognitive Science, 11, 23-63.
Gupta, P., & MacWhinney, B. (1992). Integrating category acquisition with inflectional
marking: A model of the German nominal system, Proceedings of the Fourteenth
Annual Conference of the Cognitive Science Society, (pp. 253-258). Hillsdale, NJ:
Lawrence Erlbaum Associates.
Gupta, P., & MacWhinney, B. (1997). Vocabulary acquisition and verbal short-term
memory: Computational and neural bases. Brain and Language, 59, 267-333.
Hall, D., Waxman, S., & Hurwitz, W. (1993). How two- and four-year-old children
interpret adjectives and count nouns. Child Development, 64, 1651-1664.
Harris, C. (1990). Connectionism and cognitive linguistics. Connection Science, 2, 7-33.
Harris, C. L. (1994). Back-propagation representations for the rule-analogy continuum. In
J. Barnden & K. Holyoak (Eds.), Analogical connections, (pp. 282-326). Norwood, NJ:
Ablex.
Harris, M., Barrett, M. D., Jones, D., & Brookers, S. (1988). Linguistic input and early
word meaning. Journal of Child Language, 15, 77-94.
Hebb, D. (1949). The organization of behavior. New York: Wiley.
Houghton, G. (1990). The problem of serial order: A neural network model of sequence
learning and recall. In R. Dale, C. Mellish, & M. Zock (Eds.), Current research in
natural language generation, (pp. 287-319). London: Academic.
Hunt, E. (1962). Concept learning: an information processing approach. New York: Wiley.
Huttenlocher, J. (1974). The origins of language comprehension. In R. Solso (Ed.),
Theories in cognitive psychology: The Loyola symposium, (pp. 331-388). Potomac,
Maryland: Lawrence Erlbaum.
Hyams, N. (1995). Nondiscreteness and variation in child language: Implications for
Principle and Parameter modesl of language development. In Y. Levy (Ed.), Other
children, other languages, (pp. 11-40). Hillsdale, NJ: Lawrence Erlbaum.
Hyams, N., & Wexler, K. (1993). On the grammatical basis of null subjects in child
language. Linguistic Inquiry, 24(3), 421-459.

19
 Models of the Emergence of Language

Jaeger, J. J., Lockwood, A. H., Kemmerer, D. L., Van Valin, R. D., & Murphy, B. W.
(1996). A positron emission tomographic study of regular and irregular verb
morphology in English. Language, 72, 451-497.
Jakobson, R. (1968). Child language, aphasia and phonological universals. The Hague:
Mouton.
Jusczyk, P., & Aslin, R. (1995). Infants' detection of the sound patterns of words in fluent
speech. Cognitive Psychology, 29, 1-23.
Jusczyk, P. W., Jusczyk, A. M., Kennedy, L. J., Schomberg, T., & Koenig, N. (1995).
Young infants' retention of information about bisyllabic utterances. Journal of
Experimental Psychology: Human Perception and Performance, 21, 822-836.
Kandel, E. R., & Hawkins, R. D. (1992). The biological basis of learning and individuality.
Scientific American, 266, 40-53.
Kanerva, P. (1993). Sparse distributed memory and related models. In M. Hassoun (Ed.),
Associative neural memories: Theory and implementation, . New York: Oxford
University Press.
Katz, N., Baker, E., & Macnamara, J. (1974). What's in a name? A study of how children
learn common and proper names. Child Development, 45, 469-473.
Kawamoto, A. (1994). One system or two to handle regulars and exceptions: How time-
course of processing can inform this debate. In S. D. Lima, R. L. Corrigan, & G. K.
Iverson (Eds.), The reality of linguistic rules, (pp. 389-416). Amsterdam: John
Benjamins.
Kenyeres, E. (1926). A gyermek elsö szavai es a szófajók föllépése. Budapest:
Kisdednevelés.
Kohonen, T. (1982). Self-organized formation of topologically correct feature maps.
Biological Cybernetics, 43, 59-69.
Köpcke, K.-M. (1994). Funktionale Untersuchungen zur deutschen Nominal- und
Verbalmorphologie. Linguistiche Arbeiten, 319, 81-95.
Köpcke, K. M., & Zubin, D. A. (1983). Die kognitive Organisation der Genuszuweisung zu
den einsilbigen Nomen der deutschen Gegenwartssprache. Zeitschrift für
germanistische Linguistik, 11, 166-182.
Köpcke, K. M., & Zubin, D. A. (1984). Sechs Prinzipien fur die Genuszuweisung im
Deutschen: ein Beitrag zur natürlichen Klassifikation. Linguistische Berichte, 93, 26-50.
Kuhl, P. K. (1991). Human adults and human infants show a "perceptual magnet effect" for
the prototypes of speech categories, monkeys do not. Perception and Psychophysics,
50, 93-107.
Kuhl, P. K., & Miller, J. D. (1975). Speech perception by the chinchilla: Voiced-voiceless
distinction in alveolar plosive consonsants. Science, 190, 69-72.
Kuhl, P. K., & Miller, J. D. (1978). Speech perception by the chinchilla: Identification
functions for synthetic VOT stimuli. Journal of the Acoustical Society of America, 63,
905-917.
Kuhl, P. K., & Padden, D. M. (1982). Enhanced discriminability at the phonetic boundaries
for the voicing feature in macaques. Perception and Psychophysics, 32, 542-550.
Kuhl, P. K., & Padden, D. M. (1983). Enhanced discriminability at the phonetic boundaries
for the voicing feature in macaques. Journal of the Acoustical Society of America, 73,
1003-1010.
Lachter, J., & Bever, T. (1988). The relation between linguistic structure and associative
theories of language learning: A constructive critique of some connectionist learning
models. Cognition, 28, 195-247.
Lakoff, G. (1987). Women, fire, and dangerous things. Chicago: Chicago University Press.
Landau, B., Smith, L., & Jones, S. (1992). Syntactic context and the shape bias in children's
and adults' lexical learning. Journal of Memory and Language, 31, 807-825.
Levitt, A. G., Utman, J., & Aydelott, J. (1993). From babbling towards the sound systems of
English and French: A longitudinal two-case study. Journal of Child Language, 19, 19-
49.

20
 Models of the Emergence of Language

Lewis, M. M. (1936). Infant speech: A study of the beginnings of language. New York:
Harcourt, Brace and Co.
Li, P., & MacWhinney, B. (1996). Cryptotype, overgeneralization, and competition: A
connectionist model of the learning of English reversive prefixes. Connection Science, 8,
3-30.
Ling, C., & Marinov, M. (1993). Answering the connectionist challenge. Cognition, 49,
267-290.
MacWhinney, B. (1978). The acquisition of morphophonology. Monographs of the
Society for Research in Child Development, 43, Whole no. 1, pp. 1-123.
MacWhinney, B. (1982). Basic syntactic processes. In S. Kuczaj (Ed.), Language
acquisition: vol 1. Syntax and semantics, (pp. 73-136). Hillsdale, NJ: Lawrence
Erlbaum.
MacWhinney, B. (1984). Where do categories come from? In C. Sophian (Ed.), Child
categorization, (pp. 407-418). Hillsdale, N.J.: Lawrence Erlbaum.
MacWhinney, B. (1988). Competition and teachability. In R. Schiefelbusch & M. Rice
(Eds.), The teachability of language, (pp. 63-104). New York: Cambridge University
Press.
MacWhinney, B. (1989). Competition and lexical categorization. In R. Corrigan, F.
Eckman, & M. Noonan (Eds.), Linguistic categorization, (pp. 195-242). New York:
Benjamins.
MacWhinney, B. (1993a). Connections and symbols: Closing the gap. Cognition, 49, 291-
296.
MacWhinney, B. (1993b). The (il)logical problem of language acquisition, Proceedings of
the Fifteenth Annual Conference of the Cognitive Science Society, (pp. 61-70).
Hillsdale, NJ: Lawrence Erlbaum Associates.
MacWhinney, B. (1997). Lexical connectionism. In P. Broeder & J. Murre (Eds.), Models
of language acquisition: Inductive and deductive approaches, . Cambridge, MA: MIT
Press.
MacWhinney, B., & Bates, E. (Eds.). (1989). The crosslinguistic study of sentence
processing. New York: Cambridge University Press.
MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations:
Revising the verb learning model. Cognition, 29, 121-157.
MacWhinney, B. J., Leinbach, J., Taraban, R., & McDonald, J. L. (1989). Language
learning: Cues or rules? Journal of Memory and Language, 28, 255-277.
Mandel, D. R., Jusczyk, P. W., & Kemler Nelson, D. G. (1994). Does sentence prosody
help infants to organize and remember speech information? Cognition, 53, 155-180.
Maratsos, M., & Chalkley, M. (1980). The internal language of children's syntax: The
ontogenesis and representation of syntactic categories. In K. Nelson (Ed.), Children's
language: Volume 2, (pp. 127-214). New York: Gardner.
Marcus, G., Ullman, M., Pinker, S., Hollander, M., Rosen, T., & Xu, F. (1992).
Overregularization in language acquisition. Monographs of the Society for Research in
Child Development, 57(4), 1-182.
Markman, E. (1989). Categorization and naming in children: Problems of induction.
Cambrdige, MA: MIT Press.
Marler, P. (1991). Song-learning behavior: the interface with neuroethology. Trends in
Neuroscience, 14, 199-206.
Merriman, W. E., & Stevenson, C. M. (1997). Restricting a familiar name in response to
learning a new one: Evidence for the mutual exclusivity bias in young 2-year-olds. Child
Development, 68, 211-258.
Mervis, C. (1984). Early lexical development: The contributions of mother and child. In C.
Sophian (Ed.), Origins of cognitive skills, (pp. 339-370). Hillsdale, N.J.: Lawrence
Erlbaum.
Mervis, C., & Bertrand, J. (1994). Acquisition of the novel name-nameless category (NC3)
principle. Child Development, 65, 1646-1662.

21
 Models of the Emergence of Language

Mervis, C., & Bertrand, J. (1995). Early lexical acquisition and the vocabulary spurt: a
response to Goldfield and Reznick. Journal of Child Language, 22, 461-468.
Miikkulainen, R. (1990). A distributed feature map model of the lexicon, Proceedings of the
12th Annual Conference of the Cognitive Science Society, . Hillsdale, NJ: Lawrence
Erlbaum Associates.
Miikkulainen, R., & Dyer, M. (1991). Natural language processing with modular neural
networks and distributed lexicon. Cognitive Science, 15, 343-399.
Moon, C., Cooper, R. P., & Fifer, W. P. (1993). Two-day infants prefer their native
language. Infant Behavior and Development, 16, 495-500.
Morgan, J., & Travis, L. (1989). Limits on negative information in language input. Journal
of Child Language, 16, 531-552.
Mugdan, J. (1977). Flexionsmorphologie und Psycholinguistik. Tübingen: Gunter Narr.
Naigles, L. G., & Gelman, S. A. (1995). Overextensions in comprehension and production
revisited: Preferential looking in a study of dog, cat, and cow. Journal of Child
Language, 22, 19-46.
Ninio, A., & Snow, C. (1988). Language acquisition through language use: The functional
sources of children's early utterances. In Y. Levy, I. Schlesinger, & M. Braine (Eds.),
Categories and processes in language acquisition, (pp. 11-30). Hillsdale, NJ: Lawrence
Erlbaum.
Piaget, J. (1954). The construction of reality in the child. New York: Basic Books.
Pinker, S. (1984). Language learnability and language development. Cambridge, Mass:
Harvard University Press.
Pinker, S. (1989). Learnability and cognition: the acquisition of argument structure.
Cambridge: MIT Press.
Pinker, S. (1991). Rules of Language. Science, 253, 530-535.
Polka, L., & Werker, J. F. (1994). Developmental changes in perception of non-native
vowel contrasts. Journal of Experimental Psychology: Human Perception and
Performance, 20, 421-435.
Port, R. F., & van Gelder, T. (Eds.). (1995). Mind as motion. Cambridge, MA: MIT Press.
Prasada, S., & Pinker, S. (1993). Generalisation of regular and irregular morphological
patterns. Language and Cognitive Processes, 8, 1-56.
Quine, W. V. O. (1960). Word and object. Cambridge, MA: MIT Press.
Reznick, S. (1990). Visual preference as a test of infant word comprehension. Applied
Psycholinguistics, 11, 145-166.
Rosch, E., & Mervis, C. B. (1975). Family resemblances: Studies in the internal structure of
categories. Cognitive Psychology, 7, 573-605.
Rumelhart, D. E., & McClelland, J. L. (1986). On learning the past tense of English verbs.
In J. L. McClelland & D. E. Rumelhart (Eds.), Parallel distributed processing:
Explorations in the microstructure of cognition, (pp. 216-271). Cambridge: MIT Press.
Rumelhart, D. E., & McClelland, J. L. (1987). Learning the past tenses of English verbs:
Implicit rules or parallel distributed processes? In B. MacWhinney (Ed.), Mechanisms
of Language Acquisition, (pp. 195-248). Hillsdale, N.J.: Lawrence Erlbaum.
Saffran, J., Aslin, R., & Newport, E. (1996). Statistical learning by 8-month-old infants.
Science, 274, 1926-1928.
Samuelson, L. K., & Smith, L. B. (in press). Memory and attention make smart word
learning: An alternative account of Akhtar, Carpenter, and Tomasello. Child
Development, xx, xx.
Savage-Rumbaugh, S., Sevcik, R. A., & Hopkins, W. D. (1988). Symbolic cross-modal
transfer in two species of chimpanzees. Child Development, 59, 617-625.
Schafer, G., & Plunkett, K. (in press). Rapid word learning by 15-month-olds under tightly
controlled conditions. Child Development, xx, xx-xx.
Schmajuk, N., & DiCarlo, J. (1992). Stimulus configuration, classical conditioning, and
hippocampal function. Psychological Review, 99, 268-305.

22
 Models of the Emergence of Language

Schwartz, R. G. (1988). Phonological factors in early lexical acquisition. In M. D. Smith &

J. L. Locke (Eds.), The emergent lexicon: The child's development of a linguistic
vocabulary, . New York: Academic Press.
Schwartz, R. G., & Leonard, L. B. (1981). Do children pick and choose? An examination of
phonological selection and avoidance in early lexical acquisition. Journal of Child
Language, 9, 319-336.
Schyns, P. (1991). A modular neural network model of concept acquisition. Cognitive
Science, 15, 461-508.
Shrager, J. F., & Johnson, M. H. (1995). Waves of growth in the development of cortical
function: A computational model. In B. Julesz & I. Kovacs (Eds.), Maturational
windows and adult cortical plasticity, (pp. 31-44). New York: Addison-Wesley.
Shvachkin, N. (1948). Razvitiye fonematicheskogo vospriyatiya rechi v rannem vozraste.
Izvestiva Akademii Pedagogicheskikh Nauk RSFSR, 13, 101-132.
Siskind, J. M. (1996). A computational study of cross-situational techniques for learning
word-to-meaning mappings. Cognition, 61, 39-91.
Slobin, D. (1985). Crosslinguistic evidence for the language-making capacity. In D. Slobin
(Ed.), The crosslinguistic study of language acquisition. Volume 2: Theoretical issues,
(pp. 1157-1256). Hillsdale, N. J.: Lawrence Erlbaum.
Stiles-Davis, J., Sugarman, S., & Nass, R. (1985). The development of spatial and class
relations in four young children with right-cerebral-hemisphere damage: Evidence for an
early spatial constructive deficit. Brain and Cognition, 4(4), 388-412.
Stone, G. (1994). Combining connectionist and symbolic properties in a single process. In
S. D. Lima, R. L. Corrigan, & G. K. Iverson (Eds.), The reality of linguistic rules, (pp.
417-444). Amsterdam: John Benjamins.
Sugarman, S. (1982). Developmental change in early representational intelligence: Evidence
from spatial classification strategies and related verbal expressions. Cognitive
Psychology, 14, 410-449.
Tomasello, M. (1992). First verbs: A case study of early grammatical development.
Cambridge: Cambridge University Press.
Tomasello, M., & Akhtar, N. (1995). Two-year-olds use pragmatic cues to differentiate
reference to objects and actions. Cognitive Development, 10, 201-224.
Twain, M. (1935). The aweful German language, The family Mark Twain, . New York:
Harper & Brothers.
van der Lely, H. (1994). Canonical linking rules: Forward vs. reverse linking in normally
developing and Specifically Language Impaired children. Cognition, 51, 29-72.
van Geert, P. (1991). A dynamic systems model of cognitive and language growth.
Psychological Review, 98, 3-53.
Vihman, M. M., & McCune, L. (1994). When is a word a word? Journal of Child
Language, 21, 517-542.
Wexler, K., & Culicover, P. (1980). Formal principles of language acquisition. Cambridge,
Mass.: MIT Press.
Weyerts, H., Penke, M., Dohrn, U., Clahsen, H., & Münte, T. (1996). Brain potentials
indicate differences between regular and irregular German noun plurals. Essex Research
Reports in Linguistics, 13, 54-67.
Whorf, B. (1938). Some verbal categories of Hopi. Language, 14, 275-286.
Whorf, B. (1941). The relation of habitual thought and behaviour to language. In L. Spier
(Ed.), Language, culture, and personality: Essays in memory of Edward Sapir, (pp. 75-
93). Ogden, Utah: University of Utah Press.
Woodward, A. L., Markman, E. M., & Fitzsimmons, C. M. (1994). Rapid word learning in
13- and 18-month-olds. Developmental Psychology, 30, 553-566.
Wyttenback, M., May, M., & Hoy, D. (1996). Categorical perception of sound frequency
by crickets. Science, 273, 1542-1544.

23
 Models of the Emergence of Language

Zubin, D. A., & Köpcke, K. M. (1981). Gender: A less than arbitrary grammatical category.
In R. Hendrick, C. Masek, & M. Miller (Eds.), Papers from the Seventeenth Regional
Meeting, (pp. 439-449). Chicago: Chicago Linguistic Society.
Zubin, D. A., & Köpcke, K. M. (1986). Gender and folk taxonomy: The indexical relation
between grammatical and lexical categorization. In C. Craig (Ed.), Noun classes and
categorization, (pp. 139-180). Amsterdam: John Benjamins.

24

View publication stats