Approaches to Bootstrapping: v. 1 (Language Acquisition & Language Disorders)
By J. Weissenborn, Barbara Hohle
(October 2001)
* Publisher: John Benjamins Publishing Co
* Number Of Pages: 318
* ISBN-10 / ASIN: 9027224919
* ISBN-13 / EAN: 9789027224910
Introduction
Jürgen Weissenborn & Barbara H鰄le
University of Potsdam
There is growing consensus that by the age of three children have acquired the
basic phonological, morpho-syntactic, and semantic regularities of the target
language irrespective of the language or languages to be learned, and the
language modality in which learning takes place, i.e., spoken or signed language.
Evidence is also accumulating that the schedule for the major milestones of
language development in the productive as well as in the receptive mode is
largely identical from language to language (for a detailed overview see Jusczyk
1997).
How is this early learning or bootstrapping into the native language
possible? The notion of bootstrapping implies that the child (on the basis of
already existing knowledge and information processing capacities) can make use
of specific types of information in the linguistic and non-linguistic input in order
to determine the language particular regularities which constitute the grammar
and the lexicon of her native language. Depending on the type of information
which the child makes use of, we can distinguish prosodic, lexico-semantic,
conceptual, morpho-syntactic, and pragmatic bootstrapping. The central assumption
behind the bootstrapping approach is that there is a systematic relationship
between properties of the input at one level of representation, which the child
already has access to, and another level of representation. An example is the
intensively studied parallelism between prosodic and syntactic structure, or
between lexico-semantic and syntactic structure (e.g., Gleitman 1990; Pinker
1994). In other words, the child makes use of the regularities that characterize
the interface, i.e., the interaction between different linguistic and non-linguistic
domains of representation. A problem with this strategy is, as has repeatedly
been pointed out, that this parallelism between levels of representation is only
partial (e.g., Selkirk 1984; Jackendoff 1997). The child must thus use other
means to solve the problems that result from this type of discrepancy. It could be
that the child makes use of different types of information in order to overcome
these difficulties (Hirsh-Pasek & Golinkoff 1996; Mattys, Jusczyk, Luce &
Morgan 1999; Morgan, Shi & Allopenna 1996).
Other questions related to the process of bootstrapping are whether and how
the bootstrapping strategies and their interrelation may change during development.
Such a change is to be expected given the constantly increasing knowledge
of the child in the linguistic and non-linguistic domain. For example, the growing
lexicon of the child, especially in the domain of the closed class, functional
vocabulary which in languages like English, French or German constitute about
50% of the lexical tokens of any given text, should considerably facilitate and
enhance the lexical (e.g., word segmentation and categorization) and syntactic
(e.g., determination of syntactic boundaries) bootstrapping capacities of the child
because of the distributional properties of these items. That is, from a very early
stage, the child should be able to apply — albeit only to a certain extent — an
adult-like top-down parsing strategy (e.g., H鰄le & Weissenborn 2000). We may
thus have to reckon with a constantly changing hierarchy of bootstrapping
strategies. The extent to which these changes may result in the attrition of
bootstrapping capacities that are no longer in use is not clear. The best known
evidence for changes in the sensitivity of the child to distinctions in the input is
the attested restriction of the child’s segmental discrimination capacities to the
phonological contrasts of the target (Werker & Lalonde 1988).
In addition to the dependency on the perceptual and representational
capacities of the child in the different linguistic and non-linguistic domains,
success of bootstrapping strategies will also depend on the availability of
information processing capacities like memory and attention which are necessary
to integrate the information extracted from the input into the learning mechanisms.
Thus, rule formation on the basis of distributional learning probably puts
particular demands on memory because of the necessity to keep track of the
relevant co-occurrence relations. The existence of such frequency effects in prelinguistic
children points to the importance of memory processes (e.g., Jusczyk,
Luce & Charles-Luce 1994). Consequently, changes in the bootstrapping
capacities of the child may also be the result of changes in her information
processing capacities, i.e., changes in memory and attentional resources, like, for
example, changes in (short term) auditory memory or in the capacity of the child
to coordinate her eye gaze with the eye gaze of the caretaker (e.g., Adams &
Gathercole 1995; Baldwin 1995).
In order to understand the acquisition process, it is crucial to ask to which
extent (and how) the child uses the information accessed in the input in her rule
learning mechanisms (via bootstrapping mechanisms). The fact that the child is
sensitive to a certain property of the input which may be relevant from a
theoretical perspective for the acquisition of a particular aspect of linguistic
knowledge does not yet mean that she actually uses this information to acquire
this knowledge.
Last but not least, we also have to reckon with the fact that all the capacities
and the related processes mentioned so far may be affected by changes in
the biological, neurophysiological environment in which they are embedded, and
which in turn will also be affected by the perceptual and cognitive processes
supported by it. Thus, the assumption that certain processes like the processing
of closed class functional elements are subject to an increasing degree of
automatization may be the expression of changes in the underlying brain
structure (e.g., Friederici 1995). Another effect of maturational processes in the
brain may be the existence of critical periods for the acquisition of specific
aspects of linguistic knowledge.
The main aim of the present collection of studies is to contribute to the
clarification and understanding of the questions and issues mentioned above. One
important aspect is the interdisciplinary and cross-linguistic approach taken.
Apart from experimental studies, the study of the acquisition of different
languages, which differ only minimally in some well-defined respect, is a
powerful tool for collecting evidence about the structure and interaction of
bootstrapping mechanisms.
The present studies should both challenge and stimulate the efforts in
related areas of research which are only marginally represented by these two
volumes, like the increasingly active field of modelling of acquisition processes,
the study of the interaction between general cognitive and linguistic development,
the reflection on general models of language development, and especially the
study of developmental language disorders. If, as we mentioned in the beginning,
and as shown pervasively in the research on language acquisition in the last
years, the decisive steps into language are taken during the first two years of life
(made on the basis of the powerful bootstrapping capacities displayed by the
child), it seems more promising to investigate the hypothesis that it is deficiencies
in the bootstrapping capacities that largely contribute to the emergence of
developmental language disorders.
For the study in the origin of language disorders, the importance of getting
a clearer picture of the contribution of the different bootstrapping mechanisms
and their interactions with normal language development becomes more and
more clear. As mentioned before, the relative strength of the contribution of the
different bootstrapping strategies for the extraction of language-specific regularities
from the input seems to change over time. What we do not yet know is how
much development differs across subjects and how much deviance from the
general course is tolerable without constituting a risk for successful language
acquisition. In order to find out where the potential risks for the emergence of
language disorders lie, it is necessary to compare the language development of
unimpaired and language-impaired children over time. Initial results from current
longitudinal studies in impaired and unimpaired language acquisition point to the
fruitfulness of this approach (e.g., Benasich 1998; Lyytinen 1997). Longitudinal
data is also needed to answer the question of which of the child’s early linguistic
and non-linguistic capacities underlying the bootstrapping mechanisms are
innately determined and which are rather the result of epigenetic processes.
The papers contained in the two volumes are organized into five chapters.
Chapter one concentrates on the prerequisites of early word learning. In his paper
Jusczyk discusses the beginnings of word segmentation abilities at around 7 to
8 months of age. He presents evidence that English children use mainly prosodic
cues with a preference for trochaic rhythmical patterns at the beginning but also
benefit from phonotactic constraints, allophonic cues and distributional regularities
from very early on. Furthermore, he reviews findings on the detection of
function words in the input as an aid for the development of syntactic knowledge.
Echols reports further evidence for a trochaic segmentation strategy in
English children. Moreover, she argues that perceptually salient syllables are
those syllables in the speech stream infants are especially sensitive to. Besides
stressed syllables final syllables have a high degree of perceptual saliency. She
presents findings according to which final syllable lengthening is more pronounced
in child-directed speech than in adult-directed speech. This fits with
production patterns where stressed and final syllables are more likely to be
included in the speech of one-word speakers than unstressed nonfinal syllables.
This saliency pattern could also contribute to the tendency to extract trochaic feet
from the input.
Fisher and Church discuss another open question with regard to lexical
processing:namely the question how the initially rather poor word recognition
abilities of young children develop into the efficient and rapid recognition skills
found in adults. Differences in processing as well as in lexical representations
are discussed as potential sources for these differences between children and
adults. In a series of experiments the authors found evidence that basic word
identification processes of preschoolers resemble those of adults. On the basis of
these findings it is argued that the learning mechanisms that children use to
create lexical phonological representations are the same as those mechanisms that
create long-term auditory word priming in adults, i.e., a mechanism that continu
ally updates the representations of the sound of words to reflect ongoing auditory
experience.
Bernstein Ratner and Rooney provide evidence that certain structural
properties of child-directed speech facilitate the early stages of word learning,
especially the segmentation of the speech input in word like units. Their analysis
of 10000 utterances spoken to children between 13 and 20 months of age shows
several features that might assist children in solving the segmentation problem,
namely a high proportion of very short utterances with many repetitions of
lexical items and syntactic frames. Along with the demonstrated abilities of
young children to use input information these specific input characteristics might
support early language acquisition.
With the study by Gleitman and Gleitman the focus of the discussion
changes to the semantic aspects of the acquisition of the lexicon:they ask how
word meanings are learned and how word meanings function in the semantics of
sentences. They argue that one potential source for the learning of word meanings
lies in the child’s capacity to match the occurrence of words with the scenes
and events that accompany the words in adult-to-child interactions. Furthermore,
within the syntactic bootstrapping account language internal contextual information
is assumed to provide another powerful source of information on word
meaning. Some experiments with adults reveal that these different sources of
information might be relevant for the acquisition of the meaning of different
word classes:given only extralinguistic context of a word use by video scenes
without tone adult subjects were much better in identifying the meanings of
nouns as compared to verbs. Verb identification abilities were better giving the
subjects sentence structures in which only the grammatical morphemes appeared
and all lexical morphemes were replaced by nonsense syllables. In language
acquisition these different information sources for different word classes might
be related to the initial dominance of nouns in children’s production.
Fernald, McRoberts and Swingley focus on the developmental changes in
word comprehension during the second year of life. They report findings that the
speed and the accuracy in recognizing familiar words increases significantly
within this period and that children from 18 months on already show the features
of incremental processing which are found also in adults. They argue that these
changes may reflect changes in the nature of lexical representations as well as
changes in general perceptual and cognitive processing abilities.
McKee and Iwasaki argue in a similar direction on the basis of production
data. Within the framework of a model of lemma-driven syntactic processing
they point out that the misuse and the missing of closed-class elements in
children’s production data may have several reasons:it could either result from
incomplete linguistic knowledge or from a deficient processing system that put
this underlying knowledge into actual utterances. A critical feature for distinguishing
between these alternatives is the consistency with which a pattern of
misuse appears:a deficient processing system allows for more variability than
lack of linguistic knowledge. Based on data on the acquisition of Japanese they
show the relevance of this criterion.
Chapter two focuses on the development of early syntactic knowledge. In
the first paper Gerken argues that one of the main tasks of future research is to
build the bridge between input features and the acquired system in the domain of
syntax. She focuses on the question which input cues might help the child to
detect phrase and clause boundaries to find out about syntactic structure and
syntactic categories. Besides mentioning prosodic cues she draws the attention to
the importance of the processing of grammatical morphemes which could signal
phrase and clause boundaries and could also be used to assign a syntactic
category to adjacent words. She points out that the recent findings on the
richness of the signal and the high sensitivity of infants for distributional
properties of the input should shed new light on the discussion of the logical
problem of the acquisition of syntax.
Golinkoff, Hirsh-Pasek and Schweisguth follow the line of Gerken
arguing that the sensitivity to grammatical morphemes may contribute in
important ways to the acquisition of syntax. They report findings of an experiment
that support the assumption of an early sensitivity to grammatical morphemes:
children between 18 to 20 months of age react differently to correctly
inflected verbforms than to verbs with a “wrong” inflectional ending or a
nonsense syllable replacing the inflection.
A slightly different perspective on possible input cues to the acquisition of
syntactic categories is taken in the paper by Durieux and Gillis. They discuss
several phonological features of a word itself that could be used to predict its
syntactic category. They show that the integration of several phonological cues
(stress, length, vowel and consonant quality among them) leads to good predictions
of the syntactic category for English as well as in Dutch words. But it is
still an open question whether infants can benefit from these cues in natural
language acquisition.
Within the framework of the parameter setting model for acquisition of
syntax Guasti, Nespor, Christophe and van Ooyen argue that children use the
correlation of prosodic and syntactic structure — especially the rhythmic pattern
within the intonational phrase — to find out whether their target language is head
initial or head final.
Following this idea H鰄le, Weissenborn, Schmitz and Ischebeck present
the results of a series of studies on the sensitivity of German children to word
order regularities. They found clear prosodic differences between sentences
involving head-complement constructions as compared to head-modifier constructions.
This may help children to discriminate between complements and modifiers.
Furthermore they present evidence, that children of 20 to 21 months of age
may discriminate grammatical vs. ungrammatical word order if the difference in
grammaticality correlates with differences in prosody.
Penner,Wymann andWeissenborn discuss an apparent asymmetry in the
speech of children learning German between systematic violations of the
canonical strong-weak pattern in speech production and target consistent word
order which is assumed to be acquired on the basis of the knowledge of the
stress pattern of the target. They explain the delay at the production level by the
fact that intricate interface data force the child to resort to intermediate underspecified
representations of phonological phrases.
Chapter three focuses on the interaction between prosodic and morphosyntactic
factors in the process of development of linguistic knowledge. Demuth
reports an account of syllable omission and the development of grammatical
morphology in early mono- and multimorphemic utterances of a Spanish child on
the basis of a theory of Prosodic Constraints. She shows that these constraints
are different from those found in English. The main result is that the appearance
of grammatical morphology depends on the level at which grammatical morphemes
are prosodified, with lower level elements being acquired before higher
level elements. She concludes by pointing out possible implications of her
approach for the study of individual differences, for the identification of children
at risk of language delay, and for a more general constraint-based approach to
language acquisition.
In a similar vain Lleó shows in her contribution that the fact that Spanish
determiners are acquired way before their appearance in the language of German
speaking children is explained by the different prosodic structures of the article
in the languages concerned. These prosodic differences explain that the Spanish
article appears already on single nouns whereas in German the article is first
realized within larger structures. These results provide further evidence for the
importance of the prosody-syntax interface for the acquisition of grammatical
knowledge.
This importance is confirmed by the findings of the study by Freitas,
Miguel and Hub Faria on the acquisition of codas in European Portuguese.
They show that the acquisition of elements of syllabic structure like codas may
differ depending on the grammatical features encoded by them in the target
language. Thus, codas with fricatives encoding plural are acquired earlier than
one would expect on the basis of prosodic factors alone. This finding opens up
new perspectives on the intricate interaction of different linguistic levels in
development, and especially draws attention to the fact that from very early on
abstract grammatical features must be taken into account.
Fikkert discusses data from the development of the prosodic structure of
monomorphemic and compound nouns in Dutch. In this domain, contrary to a
widely held view, it is not the case that simple structures are acquired earlier
than complex ones. What she observes instead is that the acquisition of compounds
guides the child in the acquisition of monomorphemic words consisting
of more than one foot. Her analysis is formulated in terms of a parameter setting
approach that assumes that parameters are set from an initial unmarked (default)
value to the marked value when the required evidence is encountered in the
input.
In his paper Lebeaux develops an account of how the properties of
telegraphic speech in children can be explained as the result of a prosodicsyntactic
tree mapping at the phonology-syntax interface. More specifically, he
argues that telegraphic speech is derived as a consequence of the child computing
structure with two representations:the syntactic one and the prosodic one.
The child attempts to find the maximal alignment of these two structures by
factoring out their discrepancies which had been introduced by generalized
transformations operating on identical phonological and syntactic kernel structures.
Peters proposes a model for the development of distinct closed class lexical
elements in English from an initial undifferentiated single protomorpheme
occupying grammatical positions which the child is assumed to discover on the
basis of their prosodic characteristics. The subsequent differentiation of this
protomorpheme into three distinct classes (catenatives, auxiliaries, and modals)
is the result of a gradual process of specification on the basis of growing
information from phonological, semantic, and syntactic properties of the input.
In the last paper of this section Str鰉qvist, Ragnarsdóttir and Richthoff
show on the basis of a particular cross-linguistic approach, namely the withinlanguage
group comparison (Danish, Icelandic, Swedish) that subtle differences
in the configuration of function words in terms of frequency, stress, word order,
and ambiguity have an impact on the course and structure of acquisition. They
provide evidence that the child starts with stressed, more concrete (e.g., deictic)
elements which may serve as templates for the acquisition of unstressed,
functionally different (e.g., expletive) forms instantiating the developmental
principle that new functions are first expressed by old forms.
Chapter four deals with neurophysiological aspects of language acquisition.
Molfese, Narter, van Matre, Ellefson and Modglin give an overview of changes
found in ERP-patterns to linguistic stimuli in infancy and early childhood. In the
domain of sound discrimination ERPs reflect behavioral findings very closely,
including categorical perception and the emergence of the discrimination of
different speech cues at different times. Changes observed during early language
development include changes in temporal as well as in topological features of the
ERPs. If words are used as stimuli ERPs reflect whether the words are rated as
known or as unknown by the child. Furthermore, the paper discusses findings
that ERPs may be used as a predictor for later language development:longitudinal
data suggest that children who differ in their language abilities at three or
five years of age already differ in their ERPs to speech at birth.
Friederici and Hahne focus on ERP components that correlate with the
processing of syntactic information. They report findings that adult-like temporally
different ERP patterns to semantic and syntactic violations can be found
already in children from 6 years on but that especially the component related to
a first-pass syntactic parsing mechanism is slowed down in the children. On the
basis of a three stage model for language comprehension they argue that the
parsing routines of the children are similar to those used by adults but have not
yet reached the highly automatic status found with adults.
St. George and Mills take a closer look at correlations of changes in ERP
patterns and changes in word knowledge. They report that the vocabulary spurt
goes hand in hand with dramatic changes in the topology of the ERP-pattern of
known and unknown words. They recorded ERP responses to open and closed
class items during the second to the fourth year of life linking the acquisition of
lexical knowledge and the acquisition of syntax. While initial responses to open
and closed class items are the same, at around 28 to 30 months of age the ERPs
start to be different for the two classes with a greater lateralization to the left
hemisphere for the closed class than for the open class. This difference is even
bigger for older children. Furthermore, the appearance of these changes seem to
be linked to language abilities and not to chronological age.
Chapter five groups together studies on additional perspectives of language
acquisition addressing questions of methodology, the nature of linguistic primitives,
and the development of bird song as compared to human language
acquisition. Plunkett summarizes the recent contributions of cognitive neuroscience,
experimental psycholinguistics, and neural network modelling for our
understanding of how brain processing, neural development, genetic programmes,
and the environment interact in language acquisition by focussing on the areas of
early speech perception, word recognition and the acquisition of inflectional
morphology. Each area demonstrates how linguistic development can be driven
by the interaction of general learning mechanisms, highly sensitive to particular
statistical regularities in the input, with a richly structured environment.
Bierwisch addresses the question whether the primitives of linguistic
knowledge, i.e., phonetic, semantic, and formal, morpho-syntactic features, are a
prerequisite or a result of the acquisition process. He concludes that they must
basically be considered as derived categories which emerge from the accommodation
of actual data according to general principles of representation provided
by Universal Grammar which may be interpreted as genetically fixed dispositions.
On the basis of the analysis of trajectories of song development in nightingales
Hultsch and Todt provide evidence that, in addition to interactional
variables and a predisposition to sensitive phases, the development of bird song
shares learning mechanisms with human language development like the hierarchical
organization of memory, the chunking of information into distinct units, e.g.,
songs vs. sentences, and the sensitivity to contextual factors. These similarities
have to be contrasted with the structural differences between bird song allowing
only for a limited number of meaningful elements, and human language which
provides the speaker with the possibility of an unlimited number of novel
meaningful utterances. |
楼主: 帆帆
