| AUTHOR: Snyder, William
TITLE: Child Language
SUBTITLE: The Parametric Approach
PUBLISHER: Oxford University Press
Aviad Eilam, Department of Linguistics, University of Pennsylvania
This book deals with the acquisition of interlinguistic grammatical variation,
construed as differences in the abstract grammatical generalizations the child
posits, whether they be in the form of syntactic parameters, phonological
constraints, or other possible representations. Examining the acquisition of
specific constructions, primarily in spontaneous speech, it aims to demonstrate
the significance of research in this domain for understanding how grammatical
knowledge is organized.
Following a brief introduction to the book and its general organization in Ch.
1, Ch. 2 (The View from Syntactic Theory) provides an overview of two major
syntactic theories, the Principles and Parameters (P&P) framework and
Minimalism, and discusses how they approach the issue of crosslinguistic
variation. The former views the grammar of a given language as a list of
parameter settings, and thus can straightforwardly model such variation.
Unfortunately, theoretical work attempting to identify parameters has produced
mixed results at best, as Snyder notes, possibly due to the lack of sufficiently
detailed analyses of typologically distinct languages. Nonetheless, he proposes
to consider P&P from the perspective of acquisition. Assuming that certain
constructions are contingent on some grammatical properties, the order in which
these constructions are attested in the child's language arguably reveals the
way in which the properties are arranged, i.e., the form of the underlying
parameter. Specifically, the theory yields two primary acquisitional
predictions, labeled concurrent (1) and ordered (2) acquisition (p. 7):
(1) If the grammatical knowledge (including parameter settings and lexical
information) required for construction A, in a given language, is identical to
the knowledge required for construction B, then any child learning the language
is predicted to acquire A and B at the same time.
(2) If the grammatical knowledge (including parameter settings and lexical
information) required for construction A, in a given language, is a proper
subset of the knowledge required for construction B, then the age of acquisition
for A should always be less than or equal to the age of acquisition for B. (No
child should acquire B significantly earlier than A.)
These predictions will be tested repeatedly in the following chapters. A more
controversial proposal put forward in this section is Snyder's notion of
''grammatical conservatism'' (p.8):
''Children do not begin making productive use of a new grammatical construction
in their spontaneous speech until they have both determined that the
construction is permitted in the adult grammar, and identified the adult's
grammatical basis for it.''
This idea is essential for the use of acquisition data to derive parametric
hypotheses, since it rules out the possibility that the child's data reflects a
grammar different from that of the adult. Moreover, grammatical conservatism
affords the researcher ''clean'', unambiguous data, and a well-defined point in
the time course of acquisition from which to begin the analysis. Snyder mentions
a number of studies ostensibly supporting the hypothesis of grammatical
conservatism, to be discussed in later chapters.
Having surveyed P&P, the book turns to the architecture of the Minimalist
Program. In addition to a succinct description of its general characteristics,
this portion addresses the issue of crosslinguistic variation within the theory.
Snyder presents Chomsky's (1995) original proposal to reduce all variation to
features of lexical items, and examines the extent to which various analyses
conform to this idea. One analysis which follows Chomsky is Longobardi's (2001)
Referentiality Parameter, which derives crosslinguistic variation in the order
of determiners, nouns and adjectives from the strength/weakness of the N feature
in D. A second attempt at explaining crosslinguistic variation discussed by
Snyder is Bobaljik and Thráinsson's (1998) Split IP Parameter. This proposal
aims to account for the divide in the Germanic language family between languages
which display a certain group of properties, including verb raising out of vP,
distinct inflectional affixes for tense and agreement, and multiple positions
for subjects and objects, vs. those that do not. It posits that these properties
result from the existence of more functional projections between CP and vP in
the former group compared to the latter. In other words, the parameterization is
in the hierarchy of projections, and not in the information stored in a single
lexical item, unlike Longobardi's approach. A third and final account of
crosslinguistic variation is Bošković's (2004) DP parameter, as Snyder labels
it, which also does not strictly adhere to Chomsky's version of parameterization
within Minimalism. Specifically, it links Japanese-style scrambling to the lack
of a DP layer above NP, as well as the presence of overt case marking. The
acquisitional predictions of this parameter were tested in a study on Korean,
reviewed in Ch. 6.
Ch. 3 (The View from Phonological Theory) moves from syntax to phonology,
surveying two major phonological frameworks, Optimality Theory (OT) and
Government Phonology (GP). The general machinery of OT is first depicted and
applied to the example of crosslinguistic variation in syllable structure.
Snyder then focuses on the question of learnability, acknowledging that OT is
not genuinely compatible with the idea of grammatical conservatism: the ranking
of a pair of constraints commits the learner to a resultant ranking of said
constraints vis-à-vis other constraints, possibly conflicting with the target
grammar. In the latter case, the child is expected to make substantive errors,
or errors of comission, rather than just errors of omission, contra the
predictions of grammatical conservatism.
The manner in which GP treats the typology of syllable structure is also
discussed. This framework fares better with respect to Snyder's claim of
grammatical conservatism, since its inviolable constraints can be set
independently of one another. Furthermore, its parameters tend to have a
subset/superset character, allowing the child to be grammatically conservative
by selecting the subset value and thus to avoid possibly erroneous output, until
there is sufficient evidence for the superset value.
Ch. 4 (The View from Children's Spontaneous Speech) presents a detailed guide on
how data from spontaneous speech can be used in parametric research, focusing on
the English corpora available on CHILDES. To illustrate working with these
corpora, as well as examine the frequency of errors in child language, a section
is devoted to the case study of the English verb-particle construction. Snyder
goes through the stages involved in this type of corpus work, including
step-by-step directions for computerized searches using CLAN, the software
package tailored for CHILDES. Applying his methodology to the verb-particle
constructions produced by one English-speaking child, Snyder finds only three
errors of comission in the child's 10,233 utterances. He considers the dearth of
such mistakes support for the notion of grammatical conservatism. In addition,
the data reveals a ''geyser'' effect in the acquisition of high-frequency
constructions: infrequent use at first, increasing soon after, and an explosion
of examples within a few months.
While the findings reported in Ch. 4 did not require statistical inferences, in
most cases conclusions cannot be reached simply by eyeballing the data.
Accordingly, Ch. 5 (Statistical Methods for Longitudinal Studies) provides a
synopsis of statistical methods needed to test predictions of concurrent and
ordered acquisition. It describes the relevant statistical tests, and then
demonstrates their use in a study of the acquisition of the verb-particle
construction and compounds in English. The theoretical backdrop for this study
is The Compounding Parameter (TCP), which reduces the occurrence of
verb-particle constructions in a language to the availability of novel
endocentric root compounds. This proposal is corroborated by a typological
survey: if a language exhibits the former, it necessarily allows the latter, but
not vice versa. The acquisitional prediction, whereby compounding will be
acquired prior to, or at the same time as, verb-particle constructions, was also
borne out: among 19 English-learning children, compounding was never attested
later than verb-particle constructions. Moreover, this result held even when
mean length of utterance in morphemes, argued to reflect the children's
processing capability, was factored out.
Ch. 6 (Experimental and Statistical Methods for Cross-Sectional Studies) turns
to experimental methods which can be used to assess parametric hypotheses. The
chapter reviews the techniques of elicited production and truth value judgment,
examining studies in which they were employed and the conclusions one can draw
from them. Although some of these studies have been put forward as evidence that
children's grammars are not necessarily identical to those of adults, allowing,
for example, left-branch extraction in English, Snyder presents data from both
experimental studies and studies of spontaneous speech contesting these claims.
Furthermore, he maintains that the productions and interpretations of children
in these settings do not always reflect a commitment to the grammar underlying
them. In other words, the idea of grammatical conservatism is retained, albeit
with the caveat that it may be trumped in certain contexts, such as an
experimental task. Experimental methods are nevertheless beneficial for
parametric research, provided that they are used to test forms which are assumed
to be available to the child at some point.
Following a description of statistical tests which are appropriate for these
methods, Snyder illustrates their use in a study of the acquisition of case
marking and Japanese-style scrambling in Korean (Kang 2005). Given Bošković's
(2004) DP parameter, described in Ch. 2, an ordering effect was predicted,
namely, that case marking is acquired before or concurrently with Japanese-style
scrambling. Kang indeed found a contingency between these two grammatical
options; in fact, only one child exhibited knowledge of case marking but failed
on the items testing scrambling. While this result might suggest that case
marking is not only necessary, but also sufficient, for scrambling, the fact
that languages like Icelandic have the former but not the latter constitutes
evidence against such a hypothesis. Thus, Snyder concludes that the other
prerequisites for Japanese-style scrambling are acquired earlier than case marking.
Ch. 7 (Case Studies in the Parametric Approach) is devoted to three studies of
parametric acquisition using data from spontaneous speech, which Snyder
conducted over the past few years with various collaborators. The chapter first
describes work on the acquisition of syllable structure in Dutch, done within
the theoretical framework of GP, which was surveyed in Ch. 3. The availability
of s-initial consonant clusters in word-initial position is derived from the
positive setting of two parameters, the Branching Rhyme Parameter, which
allows/prohibits branching rhymes, and the Magic Empty Nucleus Parameter, which
allows/prohibits an empty nucleus before an s+C sequence. The positive setting
of the Branching Rhyme Parameter alone licenses consonant clusters in
non-word-initial position; in other words, a subset of the options allowed by
the positive setting of both parameters. In terms of acquisition, this
hypothesized configuration leads to a prediction of ordered acquisition:
branching rhymes will be acquired prior to or concurrently with s+C sequences in
word-initial position. The findings matched the prediction: no child learning
Dutch produced s+C sequences before producing branching rhymes. The fact that no
attested language allows s+C word-initial sequences but not branching rhymes
constitutes converging evidence for the hypothesis from the typological domain.
The second study deals with noun-drop (N-drop) in Spanish, which allows speakers
to realize a DP without a noun, as in _el azul_ 'the blue one' (cf. its
ungrammatical English counterpart *the blue). The goal was to test the relation,
postulated by many researchers in the generative tradition, between N-drop and
rich overt agreement morphology within the DP. While one Spanish-speaking child
produced N-drop as soon as she exhibited mastery of gender and number agreement
on determiners and adjectives, another did not; there was a significant gap
between the two. Accordingly, Snyder argues that rich agreement is necessary but
not sufficient for N-drop, and additional abstract properties are necessary to
license the latter.
The third and final case study in this chapter involves preposition-stranding
(p-stranding) vs. pied-piping in English and Spanish. This work aimed to
determine whether all parameters have a default value with which the child
begins the acquisition process, as proposed by Chomsky (2001). If this
hypothesis is correct, we expect to find learner errors when the target grammar
is the marked value for the parameter, and the initial default value permits
grammatical constructions banned by the target grammar. In the case at hand,
this derives two possible scenarios: (1) p-stranding is the default value, and
hence both English- and Spanish-speaking children should exhibit it initially;
(2) pied-piping is the default value, so that English-speaking children are
predicted to use this strategy before shifting to the greatly preferred (and
perhaps exclusive) option found in the adult grammar, p-stranding. The results
did not match either scenario: no Spanish-speaking child exhibited p-stranding,
while no English-speaking child used pied-piping. Moreover, a majority of the
English learners displayed p-stranding following an extended period in which
they used both prepositions and direct object wh-questions, indicating that
p-stranding was also not the default for them. Thus, Snyder reasons that not
every parameter has a default specification; in addition, the notion of
grammatical conservatism is supported by the absence of substantive errors in
the children's speech.
Having gone through a range of studies on parametric acquisition, in Ch. 8
(Conclusions: Grammatical Conservatism and Cross-Linguistic Variation) Snyder
returns to the fundamental issues raised in the course of the book and
summarizes his take on them. He first tackles the question of what the child is
acquiring in her first years of life, focusing in syntax on various answers
provided by researchers working within Minimalism. As for phonology, Snyder
revisits the problem OT poses for grammatical conservatism given the collateral
effects of setting a particular ranking for a pair of constraints, but he leaves
open the question of whether there are indeed specific cases in which OT
predicts errors of comission. The subsequent portion recapitulates the evidence
for grammatical conservatism, and also takes into account possible
counterexamples, beyond the results of experimental techniques which were
already addressed in Ch. 6. For example, Snyder views the optional infinitive
stage in children, involving the substitution of an infinitive for the finite
verb of a matrix clause, as a maturational phenomenon, rather than an
acquisition error. This claim is supported by the fact that children learning a
wide variety of languages go through this stage at approximately the same time,
but otherwise do not commit to an incorrect grammar (e.g., they place the
infinitive in the appropriate clause-final position in German). Nonetheless,
Snyder acknowledges that in some domains, such as inflectional morphology,
comission errors are more common than in syntax.
Ch. 8 also reviews the acquisitional predictions derived from parametric
hypotheses, adding a prediction of earliness to those noted earlier, i.e.,
concurrent and ordered acquisition. In this case, an underlying link between two
constructions, supplied by Universal Grammar (UG), allows the child to deduce
that a given construction, when lacking sufficient evidence in the language, is
possible based on evidence for the other. We thus predict that rare
constructions may be attested early on if they are linked to other, more
widespread forms, and indeed Snyder provides experimental evidence for such a
claim from Isobe's (2005) work on head internal relative clauses in Japanese.
Snyder subsequently situates his type of acquisitional research within the
broader field of work on parametric hypotheses, noting the strengths and
weaknesses of his and other methods, such as typological studies and second
language acquisition. Lastly, he considers the linguistic theory one must
postulate in light of grammatical conservatism. According to Snyder, it cannot
be of the type proposed by Yang's (2002) variational model, since the latter
posits that all possible UG-compatible grammars are available to the child, and
hence predicts comission errors. Rather, the ideal model would consist solely of
parameters with a subset/superset character, as mentioned in Ch. 3.
Unfortunately, many of the parameters proposed in the literature are not of this
form; possible solutions to this issue include splitting existing binary
parameters into two separate parameters, or assuming that some parameters
initially have an ''unset'' value, and grammatical conservatism would follow from
the child avoiding utterances that rely on the setting of these parameters.
This book is a clear and elegant summary of the author's research on children's
acquisition of grammatical knowledge from a crosslinguistic perspective.
Essential techniques for analyzing spontaneous speech data are discussed in
depth, complemented by a theoretical point of view which puts forth interesting
observations and claims, some of which merit additional consideration. In terms
of editing, the book is nearly flawless; I found only a small number of
Among the issues which warrant attention is grammatical conservatism, i.e., the
hypothesis that the child does not produce a given construction until she has
worked out its underlying grammar, necessarily identical to that postulated by
adults. In theoretical terms, it is not obvious that this idea is different from
the observation known as Baker's paradox (Pinker 1989): children avoid obvious
generalizations when these are erroneous under the adult grammar, despite the
fact that they otherwise generalize from their input. From an empirical point of
view, it is perhaps a tendency, or an ideal, rather than an exceptionless
principle. Furthermore, its validity could be construed, at least to some
extent, as an artifact of the methodology employed, namely, transcriptions of
spontaneous speech. In order to examine the idea, it seems useful to distinguish
between two components it includes, which are confounded in the above definition:
(1) Children begin producing a construction only when they have worked out a
possible grammatical basis for it.
(2) When children produce a construction, it is necessarily derived from the
same grammatical basis as that postulated by adults.
While the validity of (1) is widely accepted, given the absence of a ''trial and
error'' period among language learners (Crain & Lillo-Martin 1999; but see, for
example, Mills 1985 for a description of the variable word order found in the
initial stages of the acquisition of German), (2) is highly debatable. Even if
it is true that in the general case, the child is a competent learner,
eventually reaching the adult grammar (Kroch 2001), this does not entail that
the latter grammar is the one used at every stage in acquisition, nor that it is
always the one adopted in the end. Moreover, the fact that languages change
arguably shows that it is not.
Already in the early days of generative grammar, it was claimed that the grammar
of a child and her parents need not be identical (e.g., Andersen 1973). Actual
cases substantiating this claim are not abundant, as expected, but do exist; for
example, the production of the possessive construction in child Dutch, which is
at odds with grammatical conservatism not only due to the unique characteristics
found in the children's output, and hence putatively also in their underlying
intermediate grammar, but also because there exists variation among the children
(van Kampen & Corver 2006). Regardless of the empirical facts, the notion of
grammatical conservatism rules out the possibility that children could be the
source of language change. This seems particularly problematic for cases of
syntactic change triggered by exogenous causes, such as morphological change,
since both the morphological change and the more profound syntactic change would
have to take place among the adults in order to be carried over to the next
generation. The latter type of linguistic innovation, involving reorganization
of an underlying grammar, is highly unlikely to occur in adults given the nature
of language learning mechanisms following the critical period. As it stands,
grammatical conservatism is not compatible with current attempts to resolve the
logical problem of language change (Niyogi & Berwick 1995).
Why, then, is there so little data from acquisition corroborating the existence
of errors, and why does Snyder fail to find substantive errors in spontaneous
speech? First, as just mentioned, all researchers acknowledge that such errors
are the exception rather than the norm. Furthermore, the reflex in the output of
underlying grammatical differences may be barely detectable, or detectable only
when the grammar is applied to a specific set of utterances. Even if the errors
exist and are noticeable, transcriptions of spontaneous speech are surely not
the best source of evidence for them, since they are both partial and
error-prone: as Snyder himself says, ''transcribing child speech is a difficult
and uncertain task'' (p. 78). The most detailed documentations of speech involve
at most one hour per week of recordings from a child, and the people responsible
for their transcription are unavoidably biased, since they all possess the adult
grammar. In addition, as Snyder notes, the available spontaneous speech data
only allow us to examine the production of high-frequency constructions, so that
there may be plenty of errors in low-frequency constructions which we simply miss.
Andersen, H. 1973. Abductive and deductive change. _Language_ 49:765–793.
Bobaljik, J. and H. Thráinsson. 1998. Two heads aren't always better than one.
Bošković, Z. 2004. Topicalization, focalization, lexical insertion, and
scrambling. _Linguistic Inquiry_ 35:613–638.
Chomsky, N. 1995. _The Minimalist Program_. Cambridge, MA: MIT Press.
Chomsky, N. 2001. Beyond explanatory adequacy. _MIT Occasional Papers in
Crain, S. and D. Lillo-Martin. 1999. _An Introduction to Linguistic Theory and
Language Acquisition_. Oxford: Blackwell.
Isobe, M. 2005. _Language Variation and Child Language Acquisition: Laying
Ground for Evaluating Parametric Proposals_. Doctoral Dissertation, Keio
van Kampen, J., and N. Corver. 2006. Diversity of possessor marking in Dutch
child language and Dutch dialects. In _Variation in Sprachtheorie und
Spracherwerb: Proceedings of the 39th Linguistic Colloquium 2004_, ed. by M.
Vliegen. Berlin: Peter Lang, 385–398.
Kang, B. 2005. _Acquisition of Language Particular Properties under Impoverished
Input_. Doctoral Dissertation, University of Connecticut, Storrs.
Kroch, A. 2001. Syntactic change. In _The Handbook of Contemporary Syntactic
Theory_, ed. by M. Baltin & C. Collins. Oxford: Blackwell, 699–729.
Longobardi, G. 2001. The structure of DPs: Some principles, parameters and
problems. In _The Handbook of Contemporary Syntactic Theory_, ed. by M. Baltin &
C. Collins. Oxford: Blackwell, 562–603.
Mills, A. E. 1985. The acquisition of German. In _The Crosslinguistic Study of
Language Acquisition: Volume 1: The Data_, ed. by D. Slobin. Hillsdale, NJ:
Lawrence Erlbaum, 141–254.
Niyogi, P. and R. C. Berwick. 1995. The logical problem of language change.
AI-Memo 1516, Artificial Intelligence Laboratory, MIT.
Pinker, S. 1989. _Learnability and Cognition: The Acquisition of Argument
Structure_. Cambridge, MA: MIT Press.
Yang, C. 2002. _Knowledge and Learning in Natural Language_. Oxford: Oxford
ABOUT THE REVIEWER
Aviad Eilam is a third year Ph.D. student at the University of Pennsylvania. His
research interests include the semantics and acquisition of expletive negation
in Hebrew, the Yiddish component in the syntax and semantics of Hebrew, and
syntactic issues in Amharic: basic clause structure, left-dislocation and
*Thanks to Charles Yang for helpful discussion.