Implicit Statistical Learning Is Directly Associated With the
Acquisition of Syntax
Evan Kidd
The University of Manchester
This article reports on an individual differences study that investigated the role of implicit statistical learning
in the acquisition of syntax in children. One hundred children ages 4 years 5 months through 6 years 11
months completed a test of implicit statistical learning, a test of explicit declarative learning, and standardized
tests of verbal and nonverbal ability. They also completed a syntactic priming task, which provided a dynamic
index of children’s facility to detect and respond to changes in the input frequency of linguistic structure. The
results showed that implicit statistical learning ability was directly associated with the long-term maintenance
of the primed structure. The results constitute the first empirical demonstration of a direct association between
implicit statistical learning and syntactic acquisition in children.
Keywords: implicit statistical learning, language acquisition, syntax, priming
Work on infant speech perception has revealed an early domain-
general ability to detect statistical regularities (e.g., Saffran, 2003;
Saffran, Aslin, & Newport, 1996). This facility for implicit statis-
tical learning is assumed by many theories of language acquisition
to lay the foundations for the acquisition of syntax (e.g., Bannard,
Lieven, & Tomasello, 2009; Bates & MacWhinney, 1982; Chang,
Dell, & Bock, 2006; Chang, Lieven, & Tomasello, 2008; Kuhl,
2004; Yang, 2004).
1
Although there have been many empirical
demonstrations showing that children can detect statistical regu-
larities in both the visual and auditory modalities (see Go´mez &
Gerken, 1999, 2000; Romberg & Saffran, 2010; Saffran, Werker,
& Werner, 2006), there has been no empirical demonstration
showing that this ability is implicated in the acquisition of syntax
in natural languages. The current article reports on an individual
differences study that directly tested the role of implicit statistical
learning in the acquisition of syntax.
Statistics in Language Acquisition
In natural languages, both word tokens and syntactic structures
differ in their frequency of occurrence. It is uncontroversial to
argue that lexical frequency affects vocabulary acquisition; however,
the role of statistics and statistical learning in the acquisition of
grammar has been more controversial. This controversy stems from
the argument from the poverty of stimulus (Chomsky, 1980; Gold,
1967), which argues that the input is too impoverished for children to
acquire a grammar through the kind of induction assumed to drive
statistical learning. The nature of this debate has remained largely
unchanged since the beginning of the modern study of language
acquisition (e.g., Lidz, Waxman, & Freedman, 2003; for a reply, see
Akhtar, Callanan, Pullum, & Scholz, 2004; MacWhinney, 2004; and
for commentaries, see Pullum & Scholz, 2002). Although this debate
continues, an increasing number of published studies have shown
frequency effects in the acquisition of syntax.
Data from naturalistic studies have shown that input frequency
is a major determinant of the acquisition of a variety of syntactic
structures (e.g., Diessel, 2004; Goldberg, Casenhiser, & Sethura-
man, 2004, 2005; Huttenlocher, Waterfall, Vasilyeva, Vevea, &
Hedges, 2010; Rowland, 2007; Rowland & Pine, 2000; Theakston,
Lieven, Pine, & Rowland, 2002, 2004). Other studies have shown
that frequency estimates from corpora predict children’s linguistic
behavior in experiments. For instance, Matthews, Lieven, Theak-
ston, and Tomasello (2005) showed that verb frequency predicts
children’s tendency to accept or correct ungrammatical word or-
ders. Bannard and Matthews (2008) reported that 2- and 3-year-old
children were better able to repeat high-frequency four-word se-
quences (e.g., sit in your chair) than four-word sequences that
differed only in their final word (e.g., sit in your truck), a change
that significantly altered the frequency of the string and, hence, the
predictability of the final word in the sequence (see also Matthews
& Bannard, 2010).
These frequency effects scale up to more complex constructions.
Huttenlocher, Vasilyeva, Cymerman, and Levine (2002) reported
on a longitudinal study that showed that the complexity of input
that kindergarten children receive from both their parents and their
teachers predicted their subsequent syntactic knowledge (see also
Huttenlocher et al., 2010). Kidd, Lieven, and Tomasello (2006,
1
Following Conway et al. (2010), the terms statistical and implicit
learning are used interchangeably throughout this article, but most often the
compound implicit statistical learning is used.
This article was published Online First October 3, 2011.
Evan Kidd, School of Psychological Sciences, The University of Man-
chester, Manchester, United Kingdom.
This research was supported by the Nuffield Foundation (SGS/33866)
and a Charles La Trobe Research Fellowship. I thank Rachael King for
testing the children and thank Franklin Chang, Jarrad Lum, and Mark
Sabbagh for helpful suggestions.
Correspondence concerning this article should be addressed to Evan
Kidd, School of Psychological Sciences, Faculty of Medical and Human
Sciences, The University of Manchester, Oxford Road M13 9PL, Man-
chester, United Kingdom. E-mail: evan.j.kidd@manchester.ac.uk
Developmental Psychology © 2011 American Psychological Association
2012, Vol. 48, No. 1, 171–184 0012-1649/11/$12.00 DOI: 10.1037/a0025405
171
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
2010) have shown that 3- to 6-year-old English-speaking chil-
dren’s knowledge of finite sentential complement clauses (e.g., I
think she’s wearing a lovely dress) is predicted by the frequency
with which the main verb occurs in a complement construction
relative to its use in other syntactic environments. Similarly, Kidd,
Brandt, Lieven, and Tomasello (2007) and Brandt, Kidd, Lieven,
and Tomasello (2009) have shown that distributional frequency
information predicts English- and German-speaking children’s
production and comprehension of relative clauses. Finally, Casen-
hiser and Goldberg (2005) showed that skewed frequency distri-
butions, where one exemplar verb occurred many more times than
did other verbs, facilitated 5-year-old children’s learning of a novel
construction (for a longer review, see Lieven, 2010).
Therefore, there is a wealth of evidence to show that (a) infants
and children are capable of implicit statistical learning in multiple
modalities and (b) children are sensitive to frequency distributions
in natural languages. That is to say, it is clear that children are
adept at extracting statistical regularities from their input, and
frequency effects in children’s linguistic behavior suggest that this
skill is important for language acquisition. However, as yet there
has been no direct empirical demonstration that implicit statistical
learning is implicated in the acquisition of grammar, and little is
known about the mechanism and the processes that track fre-
quency distributions and, as such, might support the language-
learning process. I review some relevant research next.
I begin with the working hypothesis that language learning
involves both the implicit and explicit learning processes. Using
broad brushstrokes, these processes map onto putatively syntactic
and lexical processes, respectively, although the boundary between
syntax and the lexicon is almost certainly fuzzy (Bates & Good-
man, 1997). Current work on language acquisition that assumes
the dual action of these processes has concentrated on the role of
implicit statistical and explicit (or declarative) learning in typically
developing children and children with specific language impair-
ment (SLI). Implicit statistical learning is defined broadly here as
the largely or wholly unconscious process of inducing structure
from input following exposure to repeated exemplars (e.g., Per-
ruchet & Pacton, 2006).
Tomblin, Mainela-Arnold, and Zhang (2007) showed that ado-
lescents with a diagnosis of SLI were slower to learn an implicit
pattern than a control group without language impairment. They
used a Serial Reaction Time (SRT) task to test implicit learning.
The task is a test of visual statistical learning: A visual stimulus
occurs in a repeating pattern in one of four spatial locations on a
computer screen, and the participant’s task is to press a button
corresponding to the location of the stimulus as quickly as possi-
ble. Implicit learning is observed if response times (RTs) decrease
across multiple presentations of the pattern. Because the pattern is
typically too long a sequence to memorize explicitly (e.g., 10), any
learning is argued to be implicit (but see Jamieson & Mewhort,
2009). Tomblin et al. interpreted their findings to suggest that the
grammatical deficits observed in SLI are due to a deficit in implicit
learning. These findings were supported by research reported by
J. L. Evans, Saffran, and Robe-Torres (2009), who showed that
younger children with SLI performed poorly compared with con-
trols on statistical learning tests modeled on those used with
infants. Furthermore, these authors reported a significant correla-
tion between statistical learning and vocabulary in matched typi-
cally developing controls following 21-min of exposure to the
artificial language, suggesting that performance on statistical
learning tasks are sensitive to individual differences in develop-
mental populations.
2
Subsequent research has suggested that chil
-
dren with SLI also have compromised explicit learning abilities.
Lum, Gelgic, and Conti-Ramsden (2010) tested children with SLI
and matched controls on a version of the SRT task but also tested
them on a measure of explicit (or declarative) learning. They found
that children with SLI were impaired in both tasks. This raises the
possibility that the grammatical deficits seen in SLI are the result
of impairments in both implicit and explicit learning.
The only published research to have directly investigated the
relationship between implicit and explicit learning and natural
language acquisition is a study by Kidd and Kirjavainen (2011),
who investigated the acquisition of Finnish past tense morphology
in Finnish-speaking children ages 4 years 0 months to 6 years 6
months.
3
Past tense morphology has served as a test case in the
language sciences in the debate regarding the extent to which
language is best characterized as rule driven (e.g., Pinker, 1999) or,
following connectionist approaches, the product of associative
learning (e.g., Bybee, 1995; Rumelhart & McClelland, 1986).
Compared with English, Finnish morphology is very complex:
Any one lexeme can have in excess of 100 potential surface forms,
raising the possibility that only a powerful implicit learning mech-
anism could ever learn the language. In spite of this fact, Kidd and
Kirjavainen found that implicit statistical learning, as measured by
performance on an SRT task, did not predict the children’s per-
formance on a test of past tense elicitation of both real and novel
verbs. Instead, they showed that performance on an explicit learn-
ing task predicted the children’s vocabulary knowledge, which in
turn predicted the children’s morphological knowledge. That is,
explicit learning was indirectly associated with morphological
knowledge through its association with vocabulary. This was ar-
gued to be consistent with connectionist-style single route ap-
proaches to the language acquisition of the past tense.
4
Thus, there is still no direct and specific empirical demonstra-
tion that implicit statistical learning is implicated in the acquisition
of syntax. Although Kidd and Kirjavainen (2011) did not find an
association between implicit learning and the acquisition of past
tense morphology, they suggested that such an association was not
likely because morphological processes are better characterized as
lexical rather than syntactic. Instead, they predicted that implicit
2
The same association was observed only in children with SLI follow
-
ing 42 min of exposure.
3
Kaufman et al. (2010) reported significant associations between im
-
plicit learning in 17-year-olds and foreign language achievement scores
from 1 to 2 years earlier. Although this is an impressive result, it is unclear
what exactly this association means, given that foreign language achieve-
ment is assessed in many different ways (e.g., oral and written exams,
listening comprehension, written pieces).
4
This does not entail that implicit statistical learning is not implicated in
vocabulary acquisition. On the contrary, work on infant statistical learning
suggests that this skill drives segmentation, thus enabling infants to identify
words in the speech stream (see Romberg & Saffran, 2010). Other research
suggests that attention to cross-situational statistics aids word learning
(e.g., Smith & Yu, 2008; Yu, 2008; Yu & Smith, 2007). The past tense
results seem to suggest that the composition of complex morphological
forms is dependent on lexical knowledge, consistent with usage-based
approaches to language (e.g., Da˛browska, 2008).
172
KIDD
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
learning is likely to support the kind of sequencing required to
analyze linguistic structure at the supralexical level, that is, when
analyzing sequences of words. Several results in the adult literature
provide preliminary support for this assertion. Conway, Bauern-
schmidt, Huang, and Pisoni (2010) reported on three studies that
investigated the relationship between statistical implicit learning
and language processing in adults. In each experiment, they re-
ported a significant positive correlation between accuracy on an
implicit learning task and a task that tested participants’ ability to
predict upcoming words in a sentence under degraded input. The
correlations remained significant even after static measures of
language, nonverbal intelligence, working memory, and executive
control were partialed out. These correlations were intra- and
intermodal; visual implicit learning predicted auditory sentence
processing (Experiments 1 and 3), and auditory implicit learning
predicted audiovisual sentence processing (Experiment 2). Con-
way et al. interpreted their results to suggest that implicit statistical
learning supports the acquisition of knowledge about the predict-
ability of items in a sequence (see also Conway & Pisoni, 2008;
Conway, Pisoni, & Kronenberger, 2009). Such knowledge is of
great importance in language comprehension, where the integra-
tion of incoming speech into the current analysis is eased by high
predictability. Consistent with this interpretation, Misyak and
Christiansen (2007, in press) showed that implicit learning of
nonadjacent dependencies in an Artificial Grammar Learning
(AGL) task predicted adults’ processing of subject- and object-
relative clause constructions (see also Levy, 2008; Misyak, Chris-
tiansen, & Tomblin, 2010; for models of production, see Chang et
al., 2006; Jaeger, 2010). In a more recent study, Conway, Pisoni,
Anaya, Karpicke, and Henning (2011) reported a significant cor-
relation between implicit statistical learning and grammatical
knowledge as measured by a standardized test of language in
hearing impaired children.
Syntactic Priming
In the current study, I tested children’s ability to acquire gram-
matical knowledge using the syntactic priming technique. In this
method, an experimenter or interlocutor provides a description of
a picture using a target structure (e.g., a passive). The child is then
required to describe a different picture. If they use the same
structure that was used by the experimenter, they are said to have
been primed. The technique is useful because it is a controlled
laboratory-based method of exposing participants to variations in
input. Priming has been demonstrated in children in both produc-
tion and comprehension using a variety of structures (e.g., Bencini
& Valian, 2008; Huttenlocher, Vasilyeva, & Shimpi, 2004; Kidd,
in press; Savage, Lieven, Theakston, & Tomasello, 2003, 2006;
Thothathiri & Snedeker, 2008).
Just as language learning can be conceptualized as involving
both implicit and explicit learning processes, syntactic priming has
been categorized in the same manner. Evidence supporting a role
for explicit learning processes in structural priming comes from
the fact that priming has often been observed to be short-lived
(e.g., Branigan, Pickering, & Cleland, 1999), suggesting activation
yet rapid decay, and from the fact that priming effects are in-
creased when there is lexical overlap between a prime and target
sentence (the lexical boost; Cleland & Pickering, 2003; Hartsuiker,
Bernolet, Schoonbaert, Speybroeck, & Vanderelst, 2008; Kaschak
& Borreggine, 2008).
The role of implicit learning in priming derives from the argu-
ment that exposure to a syntactic structure increases the likelihood
that the same structure will be reused because the selection of
syntactic structure has been altered as a result of prior experience.
Ferreira and Bock (2006) stated that an explanation of syntactic
priming as implicit learning “is a reflection of a longer term
process of learning how syntactic constructions in a speaker’s
language map onto the features of meaning that they express” (p.
1013). This explanation has some empirical support. Bock and
Griffin (2000) showed that priming effects last over long time lags
when priming through production, and Bock, Dell, Chang, and
Onishi (2007) showed the same effect when priming through
comprehension (see also Nitschke, Kidd, & Serratrice, 2010).
Perhaps even more convincingly, developmental data from Savage
et al. (2006) demonstrated that priming can last up to a month after
exposure in 4-year-old children (see also Vasilyeva, Huttenlocher,
& Waterfall, 2006). Finally, Ferreira, Bock, Wilson, and Cohen
(2008) showed that people with anterograde amnesia, whose ex-
plicit learning ability is severely compromised but whose capacity
for implicit learning is left relatively intact, were primed to the
same extent as normally functioning adult controls but performed
significantly worse on a recognition component of the task that
aimed to test participants’ explicit (declarative) memory for syn-
tax.
The Current Study
The literature on language acquisition currently shows that (a)
infants are capable of detecting statistical regularities over toy
grammars and (b) children are sensitive to distributional properties
of natural languages. Recent work in the adult literature has shown
that implicit statistical learning predicts performance on language
processing tasks, but similar work in the developmental literature
that has concentrated on morphology has not reported the same
relationship (though see Conway et al., 2011). The current study
extends this work by investigating the dual role of implicit and
explicit learning in syntactic acquisition. One hundred children
ages 4 to 6 years were tested on measures of implicit (i.e., statis-
tical) and explicit learning and on a syntactic priming task. A
syntactic priming task was used because it provides a dynamic
assessment of linguistic performance that indexes learning in re-
sponse to changes in the input. As such, the study aimed to identify
associations between implicit statistical learning, explicit learning,
and changes in linguistic behavior in response to changes in input
frequency (i.e., syntactic priming). Following Conway et al. (2010)
and Misyak and Christiansen (2007, in press), it was hypothesized
that implicit statistical learning would be directly associated with
children’s tendency to be primed in the syntactic priming task.
Specifically, following the syntactic priming literature, I predicted
that implicit statistical learning would be directly associated with
long-term priming effects. In contrast, it was hypothesized that the
association between explicit learning and priming would be weak
or nonexistent, because the conditions under which the children
were primed in the syntactic priming task were not consistent with
explicit learning explanations for priming (i.e., there was no lexical
overlap). That is, I primed children at the level of syntax and did
not expect a major contribution from explicit, lexical processes.
173
STATISTICAL LEARNING AND THE ACQUISITION OF SYNTAX
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Method
Participants
One hundred children ages 4 years 5 months to 6 years 11
months (M ⫽ 5 years 7 months) were recruited from primary
schools drawn from a medium-sized city in Northern England,
United Kingdom. Children of this age range were recruited be-
cause their implicit and explicit learning systems are still devel-
oping (Arciuli & Simpson, 2011; Lum, Kidd, Davis, & Conti-
Ramsden, 2010), thereby maximizing the power of the individual
differences analysis. The children’s socioeconomic status was
mixed yet representative of the region, with a majority being
working-to-middle class. There were 44 boys and 56 girls. The
children spoke English as a first language and had no cognitive or
linguistic impairment. Because the children were native English
speakers, ethnicity was not recorded.
Materials
The children were tested individually on a battery of assess-
ments, including (a) a test of syntactic priming, (b) a test of
implicit learning, (c) a test of explicit learning, (d) a standardized
test of verbal ability, and (e) a standardized test of nonverbal
ability. The presentation of the tests was pseudorandomized to
avoid order effects. The children were tested over two sessions
lasting approximately 30 to 40 minutes each, which were 1 week
apart (⫾2 days). Each test is briefly described next.
Test of implicit statistical learning. Implicit learning was
tested using a version of the SRT task (Nissen & Bullemer, 1987),
adapted for use with young children by Lum, Kidd, et al. (2010).
In this task, participants implicitly learn a repeating 10-sequence
pattern. The sequence consists of a single visual stimulus that
moves between four spatial locations on a computer screen. The
only instruction provided to participants is to press one of four
buttons on a response panel that matches the location of the visual
stimulus. For example, if the stimulus appears in Location 1, the
participant is required to press Button 1 on the response panel. If
the stimulus appears in Location 2, the participant is required to
press Button 2, and so on. The repeating sequence was presented
to participants over four blocks; in the final block, the visual
stimulus appeared in a random order. Participants’ RTs were the
primary dependent variable of interest. In healthy children (e.g.,
Lum, Kidd, et al., 2010; Thomas & Nelson, 2001) and adults (e.g.,
Daselaar, Rombouts, Veltman, Raaijmakers, & Jonker, 2003), RTs
typically decrease as they progress through the blocks with the
repeating sequence and increase on the random block. This in-
crease is often referred to as a rebound effect and is used to
demonstrate whether implicit learning has occurred, because de-
creases in RTs observed over preceding blocks may represent
practice effects. SRT tasks are arguably the best way to measure
implicit learning. Unlike Artificial Grammar Learning tasks,
where participants are explicitly told to memorize strings, learning
in the SRT task is incidental, minimizing the role of explicit
learning (Destrebecqz & Cleeremans, 2001; Kaufman et al., 2010).
Further evidence supporting the assertion that the SRT task mea-
sures implicit learning comes from studies involving patients with
basal ganglia pathology. Research has shown that patients with
Huntington’s disease (Kim et al., 2004; Knopman & Nissen, 1991)
and Parkinson’s disease (Siegert, Taylor, Weatherall, & Abern-
ethy, 2006), who have compromised implicit learning abilities,
demonstrate a smaller rebound effect than controls.
In the current study, children completed the SRT task using a
Gravis Gamepad Pro, which was connected to a Dell Latitude
C620 laptop computer. The Gravis Gamepad Pro consists of four
buttons arranged in the shape of a diamond, which children oper-
ated using their right thumb. Lum, Kidd, et al. (2010) showed that
presenting the task in this way helped maintain children’s interest
in the task, because it is presented to children in the context of a
computer game, serving to pique the children’s interest in the task
across the five blocks of trials. The visual stimulus consisted of a
well-known cartoon character, which appeared in one of four
spatial locations presented on the computer monitor with a black
background. The spatial locations on the computer monitor were
marked by four boxes with white boarders. The arrangement of
these boxes was identical to the arrangement of the buttons on the
response pad (i.e., a diamond configuration; see Figure 1). During
testing, children sat approximately 40 cm away from the computer
screen. The white boxes subtended 6.4° ⫻ 6.4° of visual angle.
During testing, children were told that the cartoon character
would appear in one of four places and that their task was to press
the buttons that matched the character’s location. Ten practice
trials were presented to ensure the children understood the task. All
children obtained an accuracy level of at least 90% on the practice
trials. The test trials were then presented. The test trials consisted
of five blocks of 60 trials. Unbeknown to the children, on Blocks
1 through 4, the appearance of the visual stimulus followed a
10-item sequence [4, 2, 3, 1, 3, 2, 4, 3, 2, 1]. The presentation of
the stimulus in the fifth block was presented in a pseudorandom
order, with the following two constraints. First, in this block, the
visual stimulus appeared in each spatial location the same number
of times as in each of the preceding blocks. Second, the probability
of observing pairs of items within the repeating sequence was the
same. Introducing this constraint meant I was able to control for
the possibility that differences in performance between the repeat-
ing and pseudorandom blocks reflected the fact that children had
learned only paired associations between picture transitions. The
transitional probabilities between locations are shown in Table 1.
For the data analysis, I computed the children’s median RT for
each of the five blocks. Only RTs for correct responses were used.
Thus, each child’s performance on the SRT task was summarized
with five data points reflecting the median RT of each block (four
learning blocks and one random). Following previous research
(e.g., Knopman & Nissen, 1991), implicit learning was indexed by
Figure 1. Spatial locations of the repeating stimulus in the Serial Reac-
tion Time task.
174
KIDD
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
finding the difference in RTs between the patterned blocks and the
random block; a significant elevation in RT from the patterned
blocks to the random block suggests that the children learned the
implicit pattern. Following past research, I call this index the
rebound effect. For the individual differences analysis, the data
were z normalized to control for individual differences in motor
speed across the sample (see the Results section).
After completing the SRT task, the children were presented with
a recall task that assessed their awareness of the pattern. In past
research using SRT tasks, adults have typically been asked
whether they detected a pattern during the test trials. Those par-
ticipants who respond in the affirmative are asked to generate the
sequence. Adults who are able to consciously recall the pattern are
then excluded from the data analysis. Adhering strictly to this
protocol with children was considered problematic given that
children may provide a prosocial response; that is, they may
indicate that they did identify a pattern to please the experimenter.
In the recall task, children were not asked whether they recognized
a pattern. Instead, they were informed that there was a pattern and
were asked to recall it. Explicit knowledge of the pattern was
assessed using a single trial. Children were seated in front of the
computer screen. The cartoon character then appeared in Position
4, and the test administrator asked the child to indicate where the
character would appear next. The children indicated their re-
sponses by pointing to one of the three boxes on the screen. The
test administrator encouraged the children to provide 10 responses;
that is, the child was encouraged to generate the entire repeating
sequence. Credit was given for any two or more sequential posi-
tions recalled throughout the child-generated 10-location se-
quence, although credit was not given for repeated identical se-
quences (e.g., 4, 3, 4, 3, . . . was only counted as 2). Consistent
with past research with children in this age range (Lum, Kidd, et
al., 2010; Thomas & Nelson, 2001), none of the children were able
to recall the complete pattern. The average number of correct
sequential locations recalled was 3.23 (SD ⫽ 1.16), the mode was
2, and the range was 2–6.
5
The number of correct sequential
positions that the children recalled was not related to the size of
their rebound effect (r ⫽ .027, p ⫽ .793).
Measure of explicit learning. The Word Pairs subtest from
the Children’s Memory Scales (Cohen, 1997) was used to measure
explicit (declarative) learning. Research has shown that perfor-
mance on learning word pairs is impaired following pathology
associated with the left medial temporal lobe (Jones-Gotman,
1992), the major neural structure thought to subserve declarative
(i.e., explicit) memory (Squire, Stark, & Clark, 2004).
In this subtest, children are asked to learn a single list consisting
of 10 word pairs (e.g., nurse–fire). Children are given three trials
to learn the list. At the start of each trial, the list of word pairs is
presented orally. Following this initial presentation, children are
presented with the first word of the pair (e.g., nurse) and asked to
recall the second (e.g., fire). This procedure is followed for the
second and third presentation of the word pairs using a different
presentation order. At the conclusion of the third trial, there is
another recall task in which participants are asked to recall both
words in the pair. The Word Pairs test therefore produces two
indices that measure declarative memory. Children’s performance
on this task was described by summing the total number of correct
responses over the three trials. Their raw scores were used in the
analyses.
Test of verbal ability. Children’s vocabulary was tested with
the British Picture Vocabulary Scale (2nd ed.;Dunn, Dunn, Whet-
ton, & Burley, 1997). This scale is a published standardized test
that measures receptive vocabulary in children. In this test, chil-
dren are orally presented with a word. Children are asked to
identify the picture that matches the word from an array of four.
Children’s raw scores were used in the analyses.
Test of nonverbal IQ. The Raven’s Colored Progressive
Matrices (Raven, Raven, & Court 1998) was used to assess non-
verbal reasoning. In this test, children are presented with a series of
stimulus pictures of abstract patterns. In each picture there is a piece
missing; the child’s task is to choose the missing piece that matches
the pattern in the stimulus picture from an array of six possible
alternatives. The decision to include a measure of nonverbal reasoning
was motivated by the concern that potential associations found be-
tween the measure implicit statistical learning and syntactic acquisi-
tion might reflect an association with general intelligence. Children’s
raw scores were used in the analyses.
Syntactic priming task. The structural priming task was
designed to prime the English full be passive construction (e.g., the
guitar was played by the man). The passive construction was
chosen because (a) it has been the most successfully primed
construction in acquisition research, and (b) it is very low in
frequency in spoken language in its full be form and seems to be
mastered only after formal reading instruction (⬍1% in either its
full or truncated form in spoken English; Bencini & Valian, 2008).
This maximized the chances of observing a priming effect, because
children are unlikely to produce passives spontaneously, and in-
creased the chance that a proportional increase in the use of
passives could be attributed to learning.
Forty-two pictures depicting transitive scenes that could be
described with either an active or a passive construction were used.
Twelve of these were prime pictures, and 30 were test pictures that
were rotated throughout the different testing phases of the task.
The pictures all depicted scenes that contained different actions,
such that the children were less likely to use a verb from a prime
sentence in their descriptions of the pictures. The prime sentences
were based on those used in previous studies (Bencini & Valian,
2008; Huttenlocher et al., 2004; Savage et al., 2003; see Appendix).
They contained a mixture of animate and inanimate nouns. The
majority of target pictures contained an animate agent performing an
5
This number is likely to be slightly inflated, because it includes 14
children whose responses went totally in an anticlockwise direction, which
credited them with five consecutive correct locations. Only one child
recalled six consecutive correct locations.
Table 1
Transitional Probabilities Between Spatial Locations in the
Serial Reaction Time Task
Location
Transitional probability
1234
1 0 0 0.5 0.5
2 0.33 0 0.33 0.33
3 0.33 0.67 0 0
4 0 0.5 0.5 0
175
STATISTICAL LEARNING AND THE ACQUISITION OF SYNTAX
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
