RUNNING HEAD: The development of verb argument structure
When do children develop knowledge of verb argument structure? Evidence from
verb bias effects in a structural priming task
Michelle Peter1a
Caroline F. Rowland1
Franklin Chang1
University of Liverpool
Ryan Blything2
University of Manchester
1Department
of Psychological Sciences, Institute of Psychology, Health & Society,
University of Liverpool, Eleanor Rathbone Building, Liverpool, L69 7ZA, UK.
2Max
Plank Child Study Centre, School of Psychological Sciences, University of
Manchester, Manchester, M13 9PL, UK.
a Corresponding
author: Michelle Peter, Department of Psychological Sciences,
Institute of Psychology, Health & Society, University of Liverpool, Eleanor Rathbone
Building, Liverpool, L69 7ZA, UK. Email: Michelle.Peter@liv.ac.uk. Telephone: +44
151 794 1401.
Abstract
The structural priming paradigm is a powerful way of constraining theories about
how children’s knowledge of verb-argument structure develops. Using a priming task
that capitalises on verb bias effects and on a phenomenon called prime surprisal, we
tested two accounts of syntactic development: one by Thothathiri & Snedeker
[Thothathiri, M. & Snedeker, J. (2008a). Syntactic priming during language
comprehension in three- and four-year-old children. Journal of Memory and
Language, 58, 188-213] that claims that children have abstract knowledge about
syntax before they learn the lexical links to these structures from verbs, and another
by Chang, Dell, and Bock [Chang, F., Dell, G. S., & Bock, K. (2006). Becoming
syntactic. Psychological Review, 113, 234-272] that proposes that children acquire
abstract and lexical knowledge simultaneously. We assessed structural priming and
the lexical boost across development. We also investigated whether the bias of the
target verb influences structural priming across development, and whether the
priming effect is stronger when there is a mismatch between prime verb bias and
prime structure (prime surprisal).Children (aged 3-4 years and 5-6 years old) and
adults heard and produced double-object dative (DOD) and prepositional-object
dative (PD) primes with DOD- and PD-biased verbs. All participants heard both the
same verb in the prime and target (e.g., gave-gave), and also a different verb (e.g.,
gave-sent). All age groups demonstrated significant evidence of structural priming,
but only adults showed increased priming when the prime and target shared a verb
(the lexical boost). Both children and adults produced more DOD responses with
DOD-biased verbs (e.g., give) than with PD-biased verbs (e.g., bring), and priming
was stronger when there was a mismatch between the prime verb’s bias and its
structure (prime surprisal). These results show that, like adults, children’s knowledge
of verb syntactic preferences influences their structure choice. We conclude that
early syntax is lexically-grounded and that the lexical boost is not a reliable measure
of verb-structure links.
Keywords: structural priming, syntactic development, verb bias, lexical boost, prime
surprisal.
When do children develop knowledge of verb argument structure? Evidence
from verb bias effects in a structural priming task
Structural priming is a powerful way of measuring what adults know about
syntactic structure. A number of priming studies have shown that, despite having the
ability to be linguistically creative, adult speakers tend to repeat the syntactic
structure of the sentences that they have recently encountered (Bock, 1986b). This
effect is not contingent on non-syntactic factors such as the prosody of sentences or
the thematic roles played by arguments, nor does it rely on the repetition of words
across sentences - although the priming effect is often larger when prime and target
sentences share a verb (the lexical boost; Pickering & Branigan, 1998). Thus, the
phenomenon of structural priming is widely interpreted as evidence that adults have
abstract representations of syntax that are stored independently of lexical items
(e.g., Bock & Loebell, 1990; Cleland & Pickering, 2006; Noppeney & Price, 2004).
Recently, the priming paradigm has helped us to learn more about children’s
syntactic representations. These studies have been particularly useful in informing
us about the way in which children’s knowledge of syntactic structure is similar to
(and differs from) that of adults. For example, Shimpi, Gámez, Huttenlocher, and
Vasilyeva (2007) found that children as young as three years old, who heard and
repeated passive prime sentences, were more likely to describe target pictures using
a passive rather than an active structure. Similar results were reported in a more
recent study by Messenger, Branigan, McLean, and Sorace (2012). Evidence of
abstract structural priming in children has also been demonstrated in language
comprehension; Thothathiri and Snedeker (2008a) reported that children’s
interpretation of target dative sentences was influenced by prior processing of dative
prime sentences that contained a different verb. The fact that these finding hold
even when prime and target do not share any lexical items, suggests that, like
adults, children as young as three years old possess verb-general abstract syntactic
representations that they employ during both production and comprehension. Thus,
the results from the child literature are fairly consistent: children appear to have
abstract knowledge of syntax from early on in development.
However, priming studies in the acquisition literature tend to focus only on this
question of what type of syntactic knowledge children have at the early stages of the
acquisition process. While this approach is a useful way of distinguishing between
early abstraction and item-specific theories (e.g., Fisher, 2002a; Pinker 1984; Valian,
1986 vs. Pine, Lieven & Rowland, 1998; Tomasello, 1992; 2000), it does not tell us
much about how children’s knowledge of syntactic structure develops. Given that
children cannot start with an adultlike knowledge of the particular syntactic structures
used in their language, all theories need to incorporate acquisition mechanisms that
enable children to interpret their own particular input to build knowledge of syntax.
Studying how children’s responses to priming tasks change over development allows
us to examine not only what knowledge children bring to the language learning task,
but also how acquisition mechanisms interact with the input to build mature linguistic
knowledge.
Recently, Rowland, Chang, Ambridge, Pine, and Lieven (2012) used a structural
priming task to investigate just this. Using a paradigm designed to test structural
priming across development, they reported significant structural priming effects in
both children (aged 3-4 years and 5-6 years) and adults. All age group produced
more double-object dative (DOD) responses after a DOD prime than after a
prepositional-object dative (PD) prime even when the prime and the target included
different verbs (e.g., give-send). From the fact that there were no significant
differences across development, the authors concluded that three year olds, like
older children and adults, have already built abstract syntactic representations of
English dative constructions.
However, when prime and target sentences shared a verb, the pattern of priming
differed across development; unlike adults, children did not show a lexical boost
effect. The adults demonstrated a substantially larger priming effect when there was
verb overlap compared to when the prime and target verbs were different (a
significant lexical boost of 34%). However, the effect was only marginal in the 5-6
year olds (10%) and non-existent in the 3-4 year olds. In other words, the repetition
of verbs across sentences did not act to enhance the priming effect for the youngest
age group in the same way that it did for adults.
These results have implications for our understanding about the relationship
between abstract and lexical knowledge across development. One possible
explanation is that the lexical boost is absent in the youngest children because
adults and children differ in the way in which they have linked the verb lexicon to
knowledge of syntactic structure. For the adults in Rowland et al.’s (2012) study, the
presence of the lexical boost suggests that syntactic structure choice is influenced,
not only by the activation of the prime structure, but by activation of the verb used in
the prime. This occurs because, in adult representations, there are connections
between verbs and syntactic structure in the form of argument structure links that
specify the syntactic structures in which verbs can occur (see Pickering & Branigan,
1998). However, in three-year olds, these verb-structure links may not exist, or, at
least, may not be strong enough yet to influence syntactic choices in a priming
paradigm. Thothathiri and Snedeker (2008a) have suggested that children may first
build abstract representations of syntactic structures (e.g., the dative structure), and
only later establish links between these representations and individual verbs,
building knowledge of the argument structure constraints of verbs slowly as
experience with each individual verb accumulates. On this view, children’s syntactic
representations are initially wholly abstract, and the lack of a lexical boost in the
youngest age group is due to an absence of any links between verbs and structures.
An alternative explanation, however, is that the lexical boost is the wrong tool to
make inferences about the relationship between verbs and syntactic structure. On
this model, the lexical boost does not stem from a lack of knowledge of verb
argument structure but from a limitation in explicit memory.. According to Chang,
Dell, and Bock (2006), structural priming occurs as a consequence of an error-based
learning mechanism. In their Dual-path model, which provides an account of
structural priming, the difference (or error) between predicted and actual sentences
is used to makes gradual weight changes in the connections between syntactic
representations. Because the model conceptualizes structural priming in terms of the
same learning mechanism that drives syntax acquisition, these weight changes
enable the model to learn syntactic structure whilst developing representations
based on the structural properties of verbs. In this way, Chang et al. predict that
verb-structure links develop in parallel with the development of syntactic knowledge.
Lexical boost effects, however, are too large to result from these types of changes;
they result, instead from the speaker’s explicit awareness of the repetition of lexical
items across prime and target sentences (Chang, Janciauskas, & Fitz, 2012). On this
model, lexical overlap simply acts as a cue in the retrieval of the explicit memory of
the prime structure and, given that explicit memory increases with age (Naito, 1990;
Sprondel, Kipp, & Mecklinger, 2011), the boost increases in line with the ability to
form, store, and retrieve explicit memories. Thus, the lexical boost is small (or even
absent) in young children not because children do not have verb-structure links, but
because they are less efficient at retrieving an explicit memory trace of the prime
sentence.
In sum, there are two possible reasons for why children and adults respond
differently in lexical boost paradigms; one that suggests that children’s knowledge of
verb-structure links has yet to develop, and one that suggest that it develops in
parallel with knowledge of syntactic structure, but that the lexical boost is the wrong
place to look for these links in children. In order to distinguish between these two
theories, and determine what children know about verb argument structure, we thus
need to use a different method to establish whether children’s and adults’ knowledge
of verb argument structure influences their syntactic structure choice. In the current
study, we did this using a modified structural priming method; one that capitalises on
verb bias effects in priming studies and on a phenomenon called prime surprisal.
Verb bias and prime surprisal
Although many dative verbs can occur in both prepositional and double object
datives (e.g., I gave him a cake/I gave a cake to him), most of these verbs will tend
to occur more often in one structure than another. For example, give tends to occur
more often in double object than prepositional dative structures (Campbell &
Tomasello, 2001; Gries & Stefanowitsch , 2004a). These syntactic preferences (or
preferred argument structure constraints) affect how adults behave in priming
studies. For example, in a corpus analysis of English dative verbs, Gries (2005)
found that target verbs strongly associated with one structure resisted being primed
into another structure. In another study by Coyle and Kaschak (2008), priming
effects were found to be larger when the target verb was not strongly associated with
one structure (i.e., when they were equi-biased). In other words, an adult’s
knowledge of a verb’s preferred argument structure (i.e., whether this verb occurs
more often in a DOD or a PD structure) influences how easily it is to prime that adult
to produce the verb in that structure.
Studying the effect of verb bias on priming provides us with a different way of
tapping into a participant’s knowledge of the links between verbs in the lexicon and
syntactic structure. We know that adults’ knowledge of a verb’s preferred argument
structure influences their structure choice during a priming task. If children have also
established these links (as predicted by Chang et al, 2006), then we should expect to
see evidence of target verb bias effects across development. However, if children
have not yet linked verbs with their argument structure preferences (as predicted by
Thothathiri and Snedeker, 2008a), then only adults, and perhaps older children,
should demonstrate target verb bias effects during a structural priming task.
In addition, related research has suggested that the identity of the prime verb
plays an important role in the size of the priming effect, such that priming is stronger
when the bias of the prime verb makes its occurrence in a particular prime structure
unexpected – an effect termed prime surprisal (Chang et al., 2006). For example,
Jaeger and Snider (2007) found that adults were more strongly primed when DODbiased prime verbs were presented in a PD prime structure and Bernolet and
Hartsuiker (2010) reported stronger priming when primes with PD-biased verbs were
presented in a DOD-structure in Dutch. In other words, the more unexpected (or
surprising) a verb is in a prime sentence, the more likely participants are to be
primed.
Not only do these results demonstrate that adult speakers store information about
verb syntactic preferences, but they also suggest that adults make predictions about
prime sentences based on their knowledge of these preferences. When these
predictions are not met (i.e., when a verb is presented in an unexpected structure),
prime surprisal works to boost the priming effect (Chang et al., 2006).We can exploit
this to assess children’s knowledge of verb-structure links. If, like adults, children
have created links between verbs and syntactic structure, then verb-structure
mismatches during a structural priming task should also lead to prime surprisal
effects in children. But, if they have not yet formed these verb-structure links, then
we should only expect to see evidence of prime surprisal effects in adults.
The current study
In this study, we capitalise on verb bias and prime surprisal effects to test two
accounts of how children develop knowledge of verb argument structure. One
possibility is that children have abstract knowledge about syntax before they learn
verb argument structure links between syntactic structures and individual verbs
(Thothathiri & Snedeker, 2008a). On this view, we should see abstract structural
priming across development but, because children have not yet created the links
between verbs and the structures in which these verbs can occur, only adults should
demonstrate evidence of lexical boost, target verb bias effects and prime surprisal.
Another possibility is that children acquire abstract and lexical knowledge
simultaneously, but that the lexical boost is not a reliable measure of verb-structure
links (Chang et al., 2006). This account also predicts abstract structural priming
across development and a lexical boost that increases with age. However, on this
account, we should see target verb bias effects and prime surprisal across
development since this account predicts that children acquire both abstract syntactic
representations and knowledge of verb argument structure simultaneously from early
on.
The current study used a modified version of the priming paradigm used by
Rowland et al. to test for structural and verb-specific priming effects in young
children (3-4 years), older children (5-6 years), and adults. Children and adults heard
and repeated DOD and PD prime sentences containing DOD- (give and show) and
PD-biased verbs (bring and send) before producing target sentences with these
verbs. First, we assessed whether we could replicate the findings of Rowland et al.
by examining structural priming and the lexical boost across development. To do
this, we observed whether participants produced more DOD- target sentences after
DOD primes than after PD primes when verbs in the prime in the target were
different (e.g., give-send), and then observed whether this priming effect increased
when the prime and target verb were the same (e.g., give-give). Second, we tested
whether children, like adults, show verb bias effects in priming tasks by exploring
whether the size of the priming effect was influenced by the bias of the target verb.
Third, we assessed whether children, like adults, demonstrate evidence of prime
surprisal. To do this we explored the priming effect was stronger when prime verb
bias and prime syntactic structure were mismatched (e.g., DOD-biased verb in a PD
structure).
Method
Participants
A total of 183 participants were tested. One hundred and twenty-three
monolingual English-speaking children were recruited from nurseries and schools in
the Liverpool area. Sixty-three of these children (32 females) were aged between
five and six years old (mean age 5;8, age range 5;0-6;11) and 55 children (33
females) were aged between three and four years old (mean age 4;0, age range 3;04;11). An additional five children from the 3-4 year old age group were tested but
produced eight or more (over half) non-target responses during the task and so were
excluded from the final analysis. A further 60 monolingual English-speaking adults
(42 females) were recruited from the University of Liverpool student participation
pool. Participants were tested individually in either their classroom/nursery or in the
language development laboratory at the University of Liverpool.
Design and Materials
Design. The study was presented in the form of a bingo game and used a 3 x
2 x 2 x 2 x 2 mixed design. Age (3-4 year olds/5-6 year old/adults) was the betweensubjects variable1. The four within-subjects variables were Prime Type (DOD and
PD), Verb Match (same verb and different verb), Prime Bias (DOD- and PD-biased)
and Target Bias (DOD- and PD-biased). The dependent variable was the proportion
1
Although descriptive analyses are presented by age-group, age was coded in months as a continuous variable
(3.5 years/5.5 years/20 years) since it is not possible to compare three categorical factors within a mixed effects
model.
of dative responses that were DODs – a ratio of DOD responses over the sum of
DOD and PD target responses.
Visual stimuli. Sixty-four video cartoon animations were created in Anime Studio
Pro and were presented in E-Prime 2.0. The cartoons included three pairs of donor
and recipient characters that are familiar to young British children and have proper
noun names: Tigger and Piglet, Dora (the Explorer) and Boots, and Bob (the Builder)
and Wendy. A further three pairs of donor and recipient characters were referred to
with determiner+noun NPs: the prince and the princess, the king and the queen, the
boy and the girl. Donor and recipient characters were always paired together (e.g.,
Wendy was always paired with Bob, and the prince was always paired with the
princess). A further five characters acted as objects and were referred to with nondefinite determiner+noun NPs: a cat, a baby, a fish, a puppy, a rabbit. All of the
characters were animate to prevent animacy contrasts between the object and
recipient influencing syntactic structure, since DOD sentences tend to occur with
animate recipients and inanimate objects.
Thirty-two of the animations depicted transfer actions that can be described
with dative sentences. Thirty-two others were used as fillers and depicted noncausal actions that can be described with an intransitive sentence. Eight of the
animations that were used as fillers were also used in a practice session. Each prime
picture was always paired with a target picture that included different characters from
those in the prime. Animations also depicted the direction of motion of transfer
actions from right-to-left equally as often as they were transferred from left-to-right to
control for the possibility that direction of transfer could influence structure choice.
Bingo cards were created to match each video cartoon animation. Four bingo
boards were created on which to place the cards during the ‘game’. Two of these
boards included a grid of four squares that were used in a practice session before
the actual experiment. The other two boards were used in the experiment and
contained nine squares.
Sentence stimuli. The four verbs used in this study – give, show, send, and
bring are dative verbs that have been shown to be familiar in both structures to
young children in the DOD and PD structure (Campbell & Tomasello, 2001; Gropen
et al., 1989).
Ninety-six different sentences and 32 verb stems were created to describe the
64 video cartoon animations. Thirty-two of these sentences described thirty-two
different cartoon animations (eight DOD sentences for each verb). These were used
as primes and depicted transfer actions using a DOD structure (e.g., Dora gave
Boots a rabbit). A further thirty-two prime sentences described the same transfer
actions but used the PD structure (e.g., Dora gave a rabbit to Boots). Thirty-two
target verb-stems (eight verb-stems for each verb) were created (e.g., the boy
brought...) in addition to thirty-two filler sentences which used an intransitive
structure (e.g., the princess jumped).
Each verb was presented eight times per participant: four times in the prime
sentence (twice in the PD structure and twice in the DOD structure) and four times in
the target verb-stem. Each participant was exposed to 16 prime-target pairs,
interspersed with 16 filler-filler pairs. Fillers were used to minimize priming effects
between pairs.
Overall each participant was presented with, and produced, 64 sentences in
total. No participant was asked to produce the same prime sentence twice and all
participants were exposed to an equal number of prime-target pairs from each of the
prime conditions.
We explored both lexically-specific and lexically-independent priming as a
within-subjects variable. Therefore participants were exposed to sentences in which
verbs were repeated across primes and targets and also sentences where the verbs
in primes and targets were different. Pairs of characters appeared equally as often in
prime and target sentences and to avoid lexical overlap (other than that of the verb),
characters in primes were always different to the characters in the targets with which
they were paired. Furthermore, primes that contained determiner noun phrases (e.g.,
the princess) were always followed by targets with proper noun phrases (e.g.,
Wendy) and vice versa. Sentences were always presented in the past tense to avoid
phonological overlap of the regular progressive (–ing) ending. Three of the verbs
were irregular (gave, sent, brought) to avoid repetition of the regular past tense (-ed)
ending.
To control for sentence-specific preferences, 12 counterbalance groups were
created to ensure that: 1) sentences that appeared as a DOD prime in one
counterbalance group appeared as a PD prime in another and vice versa; 2) all
characters in both prime and target sentences appeared equally as often with each
verb, and 3) each prime verb was paired with itself and the three remaining verbs
equally often in target sentences. All sentences across counterbalanced groups were
presented semi-randomly to ensure that participants could not predict the structure
of consecutive prime sentences. This also enabled us to ensure that characters
appearing in a filler-filler pair would not have appeared in the preceding target
sentence or in the following prime sentence.
Procedure
Children. The experiment used a confederate-scripting dialogue method
adapted from Pickering and Branigan (1998) by Rowland et al. (2012). The
experiment was conducted in the form of a bingo game in which the experimenter
and the child took it in turns to describe cartoon animations on a laptop computer to
a confederate. The experimenter introduced all of the characters involved in the task
to the child by showing them a selection of bingo cards on which these characters
appeared. Once happy that the child knew the names of the characters, the
experimenter explained the game to the child.
The experimenter and child sat in front of the computer side-by-side, while the
confederate sat opposite. The confederate was unable to see the cartoons on the
screen. The experimenter’s description of the cartoon on the left-hand side of the
screen served as a prime for the child. The child was asked to repeat the prime
sentence, addressing a hand puppet held by the confederate. The experimenter
explained that the puppet could not see the screen and needed the child to tell them
what was happening. The child was next instructed to produce a target sentence by
describing a cartoon animation on the right-hand side of the screen. A stemcompletion technique was used to ensure that the target sentence contained the
target verb but the child was instructed to produce the entire target sentence. For
example, after describing a prime animation using the verb gave (e.g., The king gave
a baby to the queen) and having the child repeat this sentence, the experimenter
would, then say “Wendy showed...”, encouraging the child to produce either, “Wendy
showed a rabbit to Bob” or “Wendy showed Bob a rabbit”. After each sentence the
confederate looked to see if he/she had the bingo card corresponding to that
cartoon. If he/she did, the correct bingo card was given to the experimenter or child
as appropriate. Each dative prime-target pair was immediately followed by an
intransitive filler-filler pair. The first person to fill the bingo grid with bingo cards was
the winner of the game and the experiment was designed so that the participant
always won.
Before running the experiment using the nine-squared bingo board, a practice
session using the four-squared board was carried out to ensure that the children
understood the task.
Adults. The procedure for adult participants was identical to the procedure for
child participants except that adults were informed that they were serving as controls
in an experiment designed for children. Additionally, they received explicit
instructions to repeat the prime and produce the entire target sentence and did not
have to direct their speech towards a hand puppet.
Coding
Target responses were recorded by the experimenter using the response
coding function of E-Prime 2.0. The experiment was also audio-taped, allowing the
first author to listen back to and then transcribe and code the data. Many of the
young children and some of the older children needed prompting from the
experimenter to produce the prime and the entire target sentence correctly. Some of
them, however, only produced partial target responses (e.g., they completed the
stem without including the target verb). In order to capture these partial target
responses, we employed three levels of coding: lax, intermediate, and strict. To
qualify for lax coding, the prime sentence had to be repeated correctly but the
participant might have received help to do this. In addition, the participant may not
have produced the target verb or the entire target utterance, but instead just
completed the target stem. To qualify for intermediate coding, the prime sentence
had to be repeated correctly with the participant needing minimal help to do this. In
addition, the entire target utterance was produced, but prompting to do this was
needed more than once. To qualify for strict coding, the prime and target sentence
had to be produced correctly with no more than one prompt. A target response was
considered a DOD if it contained the correct target verb followed by two noun
phrases, and a PD if it contained the correct target verb followed by a noun phrase
and a prepositional phrase headed by ‘to’. Responses coded as ‘other’ were those
where, a) the participant failed to repeat the prime correctly (even after help), b) the
participant produced the incorrect target verb and, c) the target sentence included
the preposition ‘at’ rather than ‘to’. Preliminary analysis revealed that all coding
schemes produced a virtually identical pattern of results, and so the following
analyses are reported only on the strictly-coded data.2.
2
Under the lax coding scheme, 3-4-year olds, 5-6-year olds, and adults produced 0.13%, 0.03%, and 0.01%
‘other’ responses respectively. Under the intermediate coding scheme, 3-4-year olds, 5-6-year olds, and adults
produced 0.22%, 0.05%, and 0.03% ‘other’ responses respectively. Under the strict coding scheme, 3-4-year
olds, 5-6-year olds, and adults produced 0.24%, 0.07%, and 0.03% ‘other’ responses respectively.
Results
The aims of the present study were: 1) to replicate Rowland et al. (2012) to assess
structural priming and the lexical boost across development, 2) to explore whether
the bias of the target verb influences structural priming across development, and 3)
to investigate whether a mismatch between prime verb bias and prime structure
influences structural priming (prime surprisal) across development.
1. Structural priming and the lexical boost across development
Our first aim was to investigate whether we would find abstract structural priming and
a lexical boost across all three age groups. Using a production-to-production priming
paradigm, we tested lexically-independent and lexically-specific priming. The mean
proportion of DOD responses after DOD and PD primes in the same verb and the
different verb conditions was calculated. Structural priming was demonstrated by a
greater proportion of DOD responses after DOD primes than after PD primes.
Evidence of a lexical boost would be demonstrated by a bigger priming effect in the
same verb condition than in the different verb condition.
% of dative target responses that were DODs
Fig. 1.
Mean proportion of DOD responses after DOD and PD primes when primes and targets
were different (different verb) and the same (same verb) (error bars indicate standard error)
0.8
0.7
0.6
0.5
After DOD
0.4
After PD
0.3
0.2
0.1
0
3-4
5-6
Different verb
Adults
3-4
5-6
Adults
Same verb
Figure 1 demonstrates the mean proportion of dative responses that were DODs
after PD and DOD primes for each age group and verb condition (same/different).
Table 1 reports the size of the priming effect for each age group and verb match
condition (calculated as a percentage of DOD responses produced after DOD primes
minus the percentage of DOD responses produced after PD primes) and also as
Cohen’s d. These data indicate that the structural priming effect was largest in the
adults (27%). Comparison of the same and different verb conditions shows that verb
overlap boosted the structural priming effect (by 23%) for the adults, but not at all for
the children.
Table 1.
Size of priming effect – calculated both as the percentage of DODs produced in each prime
condition (difference score) and effect size (Cohen’s d)
Age
3-4
5-6
Adults
Different verb
Same verb
Difference
Score
Standard
error
Cohen's
d
Difference
score
Standard
error
Cohen's
d
0.15
0.14
0.27
0.049
0.030
0.048
0.45
0.42
0.90
0.03
0.06
0.50
0.041
0.023
0.044
0.08
0.17
1.86
The results were analysed with a logit mixed effect (maximal) model as these models
are well suited to analyzing binary dependent measures and allow us to model both
random subject and item effects together (Baayen, Davidson, & Bates, 2008; Jaeger,
2008). The model included as fixed effects: a) Verb match condition (same/different
verb); b) Age (3-4 year olds/5-6 year olds/adults) and c) Prime type (DOD, PD). Age
was centred to reduce multi-collinearity (Neter, Wasserman, & Kutner, 1985) and
other factors were effect/sum coded (Wendorf, 2004). Participant was included as a
random effect. The dependent variable was the proportion of DOD structures
produced – a ratio of DOD responses over the sum of DOD and PD target
responses.
The results revealed a main effect of Prime Type; β = 1.26 (SE = 0.11), z =
11.51, p < .001, indicating that the participants produced more DOD responses after
DOD primes than after PD primes. There was also a main effect of Age; β = 0.11
(SE = 0.02), z = 6.37, p < .001, so the likelihood of producing DOD targets increased
with age. An interaction between Prime Type and Age; β = 0.09 (SE = 0.01), z =
6.24, p < .001, showed that, overall, the size of the priming effect increased with age.
There was also a main effect of Verb match condition; β = -0.21 (SE = 0.11), z = -
1.99 p < .01, indicating that more DOD responses were produced in the same verb
condition than in the different verb condition. There was no interaction between
Prime type and Verb match condition, suggesting that the priming effect did not
increase when verbs in the prime and target were the same. However, these results
should be considered in the light of the three-way interaction found between Prime
type, Age and Verb match condition, β = 0.12 (SE = 0.03), z = 4.18, p < .001. This
finding indicates that while, on the whole, there was no lexical boost, the priming
effect was greater when there was verb overlap for some of the age groups. To
explore the interaction further, we conducted three models on each age group.
1.1.
Analyses by age
Three separate logit mixed effect models were run on each age-group. The 3-4 year
olds demonstrated a main effect of Prime type only, β = 0.77 (SE = 0.21), z = 3.68, p
< .001, no effect of Verb match condition and no interaction. There was no significant
difference between the proportion of DOD responses after DOD primes produced by
3-4 year olds in the same (31%) and different (35%) verb condition. In other words,
they did not demonstrate a lexical boost. Likewise, the 5-6 year olds showed a main
effect of Prime type only, β = 0.85 (SE = 0.19), z = 4.54, p < .001, and no effect of
Verb match condition or interaction; they too showed no lexical boost. In contrast,
analysis of the adults’ data revealed a main effect of Prime type; β = 2.02 (SE =
0.16), z = 12.49, p < .001, Verb match condition; β = -0.32 (SE = 0.16), z = -2.07, p =
.04, and also a significant interaction between Prime type and Verb match condition;
β = 1.30 (SE = 0.31), z = 4.13, p < .001. This means that, unlike the children, the
adults produced significantly more DOD responses after DOD primes in the same
(76%) verb condition than in the different (70%) verb condition. Thus, there was a
large lexical boost effect in the adults.
2. The effect of target verb bias on structural priming across development
Our second aim was to investigate both how the target verb bias influenced the
participants’ choice of syntactic structure overall, and whether it influenced the size
of structural priming across development.
Fig. 2.
Mean proportion of DOD responses when prime and target verbs were and the same with
(a) DOD-biased target verbs and (b) PD-biased target verbs (error bars indicate standard
error)
(a)
% of dative target responses that were DODs
0.9
0.8
0.7
0.6
After DOD
0.5
After PD
0.4
0.3
0.2
0.1
0
3-4
5-6
Adults
Different verb, DOD-biased target
3-4
5-6
Adults
Same verb, DOD-biased target
% of dative target responses that were DODs
(b)
0.8
0.7
0.6
0.5
After DOD
0.4
After PD
0.3
0.2
0.1
0
3-4
5-6
Adults
Different verb, PD-biased target
3-4
5-6
Adults
Same verb, PD-biased target
As before, we ran a logit mixed effect model to analyse the data. This time the model
included as fixed effects: a) Verb match condition (same/different verb); b) Age (3-4
year olds/5-6 year olds/adults); c) Prime type (DOD, PD) and d) Target bias (DODbiased, PD-biased). Again, age was centred to reduce multi-collinearity. There were
no interactions between Prime type and Target bias, suggesting that the amount of
priming was not affected by the bias of the target verb. However, the bias of the
target verb did affect the participants’ choice of syntactic structure overall. The
results revealed a main effect of Target bias; β = -1.10 (SE = 0.11), z = -9.80, p <
.001, which indicates that more DOD responses were produced with DOD-biased
target verbs than with PD-biased target verbs overall. There was also an interaction
between Target bias and Age; β = -0.03 (SE = 0.01), z = -1.84, p = .07, caused by
the fact that the effect of target verb bias increased with age. There was a three-way
interaction between Target bias, Age and Verb match condition, β = 0.07 (SE =
0.03), z = 2.48, p = .01. This suggests that more DOD responses were produced
with DOD-biased target verbs than with PD-biased target verbs and this effect varied
across both age groups and verb conditions.
2.1.
Analyses by age
To analyse the three-way interaction further, three separate logit mixed effect models
were run on each age-group. The 3-4-year olds demonstrated a main effect of
Prime type, β = 1.02 (SE = 0.26), z = 3.86, p < .001, and a main effect of Target bias
only; β = -2.03 (SE = 0.40), z = -5.06, p < .001. So, while the size of the priming
effect was not influenced by the bias of the target verb, 3-4-year olds produced more
DOD responses with DOD-biased target verbs- than with PD-biased target verbs
overall. Likewise, 5-6-year olds showed a main effect of Prime type, β = 0.95 (SE =
0.20), z = 4.67, p < .001, and a main effect of Target bias; β = -1.44 (SE = 0.23), z =
-6.23, p < .001 (see Figure 3). They also showed a main effect of Verb match
condition; β = -0.44 (SE = 0.22), z = -2.02, p < .05, indicating that more DOD
responses were produced when there was no verb overlap (since this is not relevant
to our predictions, we do not consider it further). In the adults, there was a main
effect of Prime type, β = 2.53 (SE = 0.24), z = 10.62, p < .001, of Target bias; β = 1.59 (SE = 0.22), z = -7.38, p< .001, and of Verb match condition, β = -0.47 (SE =
0.19), z = -2.40, p < .01. There was also an interaction between Target bias and
Verb match condition, β = 1.44 (SE = 0.38), z = 3.83, p< .001 and between Prime
type and Verb match condition, β = 1.18 (SE = 0.45), z = 2.64, p< .001. However, as
with the children, there was no interaction between Target bias and Prime type,
indicating that Target verb bias affected the number of DODs produced overall, but
not the size of the priming effect. In sum, children and adults were more likely to
produce DODs with DOD-biased verbs (and thus more likely to produce PDs with PD
biased target verbs). However, the bias of the target verb did not affect the size of
the priming effect.
3.
The effect of prime verb bias on structural priming across development
Lastly, we investigated the effect of prime verb bias on structural priming.
Specifically, we explored whether children and adults are sensitive to prime surprisal
by testing whether the priming effect was bigger when there was a mismatch
between prime verb bias and prime structure.
Fig. 3.
Mean proportions of DOD responses when prime verb bias and prime structure were the
same (match), and when prime verb bias and prime structure were different (mismatch) with
(a) DOD-biased target verbs and (b) PD-biased target verbs (error bars indicate standard
error)
(a)
% of dative target responses that were DODs
1
0.8
0.6
After DOD
0.4
After PD
0.2
0
3-4
5-6
Adults
Match, DOD-biased target
3-4
5-6
Adults
Mismatch, DOD-biased target
(b)
% of dative target responses that were DODs
0.6
0.5
0.4
After DOD
0.3
After PD
0.2
0.1
0
3-4
5-6
Adults
Match, PD-biased target
3-4
5-6
Adults
Mismatch, PD-biased target
To analyse the data, two variables were created: ‘mismatch’, in which the prime verb
occurred in an unexpected structure (e.g. a DOD biased verb in a PD structure), and
‘match’, in which the prime verb occurred in a match structure (a DOD biased verb in
a DOD structure).
Using a logit mixed effect model, analyses were conducted on a subset of the
dataset that only included data from the different verb condition. Data from the same
verb condition was not included because the ‘match’ effect would be confounded by
a lexical boost effect. For example, in the ‘match’ condition, when the prime verb is
PD-biased, the target verb will always be a PD-biased verb (sent or brought) which
could increase the priming effect. The logit mixed effect model included as fixed
effects: a) Age (3-4 year olds/5-6 year olds/adults); b) Prime type (DOD, PD); c)
Prime bias match (match, mismatch), and d) Target bias (DOD-biased, PD-biased).
Again, age was centred to reduce multi-collinearity.
The results revealed main effects of Prime type; β = 1.61 (SE = 0.23), z = 6.98, p
< .001, Age; β = 0.14 (SE = 0.02), z = 6.26, p < .001, and Target bias; β = -1.43 (SE
= 0.21), z = -6.71, p < .001. There was also a significant interaction between Prime
type and Prime bias match; β = 0.88 (SE = 0.43), z = 2.06, p < .05, indicating that,
overall, structural priming was greater when the prime verb bias and prime structure
were mismatched (i.e., prime surprisal increased the priming effect). There was no
interaction between Prime type, Prime bias mismatch and Target bias, indicating that
the bias of the target verb did not influence prime surprisal. However, there was a
three-way interaction between Prime type, Age and Prime bias match; β = -0.09 (SE
= 0.04), z = -2.08, p = 0.037. Inspection of Figure 3 (a and b) shows that this threeway interaction resulted from the prime surprisal effect decreasing with increasing
age. In other words, a mismatch between prime bias and structure was less
surprising for adults than for children.
3.1.
Analyses by age
Three separate analyses were run to further explore the interaction between Prime
type, Age and Prime bias match. For the 3-4 year olds, there was a significant
interaction between Prime type and Prime bias match; β = 3.50 (SE = 1.45), z =
2.41, p < .05, indicating prime surprisal increased the size of the structural priming
effect for this age-group. There was also a significant interaction between Prime type
and Prime bias match for the 5-6 year olds; β = 2.28 (SE = 0.77), z = 2.98, p < .001,
so prime surprisal also increased the amount of structural priming for this age-group.
However, the adult data revealed only marginal interaction between Prime type and
Prime bias match; β = -1.41 (SE = 0.73), z = -1.94, p < .1, showing that the
strengthening effect of prime surprisal on structural priming decreased with age. In
other words, adults were less sensitive to prime surprisal than children.
Results summary
Both children and adults demonstrated significant structural priming; they all
produced more DOD sentences after DOD primes than after PD primes when prime
and target verbs were different. However, only adults showed increased priming
when prime and target sentences shared a verb. Thus, consistent with recent
findings (Rowland et al., 2012), the lexical boost increased with development. The
results also revealed verb bias effects across development: 1) both children and
adults produced more DOD responses with DOD-biased target verbs (e.g., give)
than with PD-biased target verbs (e.g., bring) and this effect was larger in adults than
in children, 2) priming was stronger when there was a mismatch between the prime
verb’s bias and its structure (prime surprisal), with this effect decreasing across
development.
Discussion
In this study, we used a priming paradigm to test for structural and verb-specific
priming effects in young children (3-4 years), older children (5-6 years), and adults.
Our aim was to test two accounts of how children develop knowledge of verb
argument structure. One is that children develop abstract knowledge about syntactic
structure first, and then learn the lexical links between these structures and verbs
that constitute verb argument structure knowledge (Thothathiri & Snedeker, 2008a).
On this view, we predict abstract structural priming across development but a
growing lexical boost with age. Crucially, we also predicted that only adults should
be affected by the argument structure preferences of particular verbs (target verb
bias) and by prime surprisal. An alternative account, however, proposes that children
acquire abstract and lexical knowledge simultaneously, but that the lexical boost is
not a reliable measure of verb-structure links (Chang et al., 2006). This account also
predicts abstract structural priming across development and a lexical boost that
increases with age. However, it also predicts that, once we see structural priming
effects, we should also see target verb bias and prime surprisal effects since children
acquire both abstract and verb-specific knowledge together from early on.
To test these accounts, we first assessed whether we could replicate the
findings of Rowland et al. (2012) that showed structural priming effects are present
across development, but that a lexical boost that grew with age. Second, we tested
whether children, like adults, are influenced by the argument structure bias of the
target verb. Third, we examined whether children, like adults, demonstrate evidence
of prime surprisal.
In our study we replicated the findings of Rowland et al. (2012). We found
evidence for structural priming across development, in that young children, older
children, and adults were significantly more likely to produce DOD sentences (e.g.,
Wendy gave Bob a rabbit) after hearing and repeating DOD primes (e.g., The boy
sent the girl a fish) than after PD primes (e.g., The boy sent a fish to the girl). These
results support Rowland et al.’s conclusion that, by the age of three, young children
have built some form of abstract syntactic representations for the English dative
constructions, which enable them to generalise across similarly-structured
sentences. Our findings also fit with other studies reporting evidence for abstract
priming in three-year-olds (e.g., Bencini & Valian, 2008; Messenger et al., 2012;
Shimpi et al., 2007). Like Rowland et al., we also observed a lexical boost that
increased with development: adults showed increased priming (23%) when the verb
in the prime matched the verb in the target, but 3-4-year old and 5-6-year old
children did not. In other words, although structural priming effects were apparent
across development from age three to adulthood, the lexical boost was only
apparent in adults.
Given that we replicated the results of Rowland et al. (2012), our second aim
was to compare the predictions of two differing explanations of how children develop
knowledge of verb argument structure. We did this in two ways. First we tested
whether participants’ responses were influenced by the bias of the target verb across
all three ages. We found that they were. Although the target verb bias did not affect
the size of the priming effect for either children or adults, all age-groups produced
more DOD responses with DOD-biased target verbs (give, show) than with PDbiased target verbs (bring, send). In other words, children as young as three years
were influenced by the alternation biases of the four dative verbs used in this study.
Second, we tested whether children and adults were influenced by prime
surprisal. Again, we found significant effects across all three age group. Priming
effects were stronger when there was a mismatch between the prime verb’s bias and
the prime structure (prime surprisal). In other words, the bias of the prime verb had
an effect on the size of the priming effect across all three ages.
Both of these findings are difficult to explain in terms of a theory that assumes
that the acquisition of abstract syntactic knowledge happens prior to, and in isolation
from, the acquisition of verb’s argument structure constraints (e.g., Thothathiri &
Snedeker, 2008a). Instead, our results the results support a model in which children
develop knowledge about syntactic structure at the same time as they learn about
the argument structure preferences of particular verbs.
This model is the connectionist model of syntactic processing and
development called the Dual-path model (Chang, Dell, & Bock, 2006). The model
has a dual-pathway architecture which separates conceptual-lexical and syntactic
learning into separate meaning and sequencing pathways. This architecture is
argued to be critical for learning isolable syntactic structures (Chang, 2002).
Syntactic knowledge is learned within the sequencing pathway through an errorbased learning algorithm that is applied to next-word prediction. For example, the
model might predict that after the word the, the next word should be boy. But, if the
next word is actually girl, then the mismatch between the prediction and the actual
next word (the error) is used to change the weights in the model so that it better
predicts girl after the. Since the weights are initially random, at first, the model might
not predict the next word correctly. With time, the model learns syntactic categories
which subsequently combine into syntactic structures that allow it to predict sets of
words (e.g., ‘the’ is followed by a noun: ‘boy’, ‘girl’, ‘cat’...). Because the model is
building syntactic knowledge from individual words, knowledge of which verbs can
occur in which structure (verb argument structure) develops at the same time as
knowledge of syntactic structure. In other words, the model develops knowledge
about verb argument structure using the same error-based learning mechanism that
it uses to develop knowledge of syntactic structure.
As a result, the model predicts that effects of target verb bias will be seen as
soon as we see evidence of structural priming, because children acquire syntactic
information about the argument preference of particular verb at the same time as
they are building up knowledge of syntactic structure. In addition, a mismatch
between the prime verb and the prime structure (e.g., a PD-biased verb in a DOD
biased structure) leads to more error in prediction and thus a greater weight change
in the model’s representation for that syntactic structure (e.g., the DOD). A greater
weight change means a greater likelihood of the model choosing the same syntactic
structure for its own subsequent production, which leads to an increased priming
effect.
Explaining the developmental pattern of target verb bias and prime surprisal
In sum, our findings support an account like that of Chang et al. (2006) in which it is
assumed that early syntactic representations are lexically-grounded. However, there
are two findings in our data that remain to be explained. First, the effect of target
verb bias increased over development: adults were more strongly influenced by the
bias of the target verb than children. Conversely, the effect of prime surprisal
decreased over development; adults were less influenced by the bias of the prime
verb than children. This apparent contradiction is difficult to explain because both
effects are assumed to stem from the same source; the participants’ knowledge of a
verb’s preferred argument structure, which both influences the choice in production
of the target and generates the expectation that creates primes surprisal.
We speculate here that both results could stem from the dynamics of the learning
process. The growth of the target verb bias over development can be explained by
the way that children use verb-structure pairs in the input to adjust their internal verbstructure links in their language network. If children learn verb biases over time, their
representations moment-by-moment represents the evidence they have so far
accumulated from the input. However, daily probabilities of verb-structure pairs do
not always match the long term probabilities. This means that early on in the
learning process, while the child is building up evidence, her representations are
likely to be less stable. For example, although the verb throw may have an overall
bias towards the PD frame, it is likely that there will be days in which children only
hear throw in the DOD. Early on in the learning process, these instances would
weaken the PD bias of her representation of the verb throw. Thus, learning target
verb biases is a gradual process of accumulating evidence over time which can be
temporarily derailed by random fluctuations in the input. This can explain why target
verb bias increases with development in this study.
The decrease in prime surprisal over development can be explained by changes
in the magnitude of the learning rate. Priming in the Chang et al. model is due to
error-based learning, which is influenced by a learning rate parameter. The error
between the predicted next word and the actual next word is the error signal (akin to
surprisal) and the influence of this error signal on the model’s representations is
modulated by the learning rate. Thus if the learning rate is higher early in
development, the effect of surprisal on priming will be larger earlier in development.
In other words, we might see greater prime surprisal early in development because
the child is learning at a faster rate than the adult.
There is independent motivation for using higher learning rates early on in
connectionist models (all of the versions of the Dual-path model use higher learning
rates early in development; Chang, 2002; 2009; Chang et al. 2006). In connectionist
models, weights are initially randomized, and a small learning rate means that
learning makes only small changes to these weights. The result is that models
learning a complex set of representations tend to fail to converge on the target
representations (e.g., sequences of abstract syntactic categories) within an
appropriate amount of time. Using large, initial learning rates and reducing them
over development allows the model to escape from these local minima and discover
the right representations. For example, Chang et al. (2006) started his model with a
learning rate of 0.25 for 10,000 patterns and then gradually lowered this learning rate
to 0.05 after another 30,000 patterns. In Chang et al.’s model, if we test priming
early on when the learning rate is 0.25, the effect of surprisal will be larger than if we
test it later on when the learning rate is 0.5. In this way, a decrease in learning rate
over development leads to a decrease in prime surprisal effects over development.
To clarify how these two effects could arise within a developmental account of
language acquisition, we developed a simple dynamic systems model of verb bias
acquisition. In this model, we assumed that children get 100 inputs for two verbs (A,
B), which were coded as 1 for PD and 0 for DOD (variable pd). Verb A occurs in PD
80% of the time and verb B occurs in PD 20% of the time. We then set the learning
rate parameter to start at 0.1 and reduce to 0.01 in 100 steps. The model is
governed by four equations. The first equation updates the PD bias for the verb V
that is inputted by adjusting the previous PD bias for that verb by the error between
the previous bias and the present input (PD = 1, DO = 0), multiplied by the learning
rate. The second equation updates the bias for the other verb to be unchanged from
the previous time step. The third equation set the initial bias to a uniform random
number between 0 and 1. The fourth equation sets the PDCHOICE variable to be
set to the bias of the verb that is being used at each point in time. These equations
were used to generate each random child:
1) BIASV,T = BIASV,T-1 + ( PDINPUTT - BIASV,T-1 ) * LRATET if verb matches V
2) BIASV,T = BIASV,T-1 if verb mismatches V
3) BIASV,0 = UniformRandom (0, 1)
4) PDCHOICET = BIASV,T if verb is V
To see how this model would explain the results in this study, we generated two
random children (child 1 and child 2). The initial bias (at time 0) is a random number
between 0 and 1 and hence child 2 has an initial bias of 0.85 for verb B, which is a
bias for the DOD. But since verb B occurs in the PD 20% on average, over
development, the bias gradually decreases until the child matches the adult’s
baseline bias. On the other hand, verb B in child 1 is already close to the adult level
(0.3) at the start of development. However, due to the random run of PDs early in
development, the bias is moved first towards PD before returning to the adult level.
Fig. 4.
Development of target verb bias for two verbs (A and B) over 100 exemplars for two children
Figure 4 shows that early verb bias might not match adult biases and that the
gradual learning of adult biases can explain the growth of the target bias in our
study. The figure also demonstrates effects of the learning rate. The learning rate
starts off high and reduces over development, which means that early verb biases
are affected more substantially by the identity of each prime structure (i.e., vertical
displacement is greater). For example, for Verb A, Child 2 starts off with a bias of
0.98 but, once the first DOD is experienced, the bias drops by 0.08 because the
learning rate is high. Later in development, biases change by only 0.01 because the
learning rate is lower. This explains why prime surprisal effects are larger earlier in
development.
The dynamic systems model helps us understand how a model that allows verbstructure representations to change over development could provide an explicit
account of the behavioural changes in this study. Although models which
incorporate changes in learning rate over development are generally motivated by
model-specific assumptions (e.g., avoiding local minima), we can link these
assumptions to more general issues with the literature on the critical/sensitive period.
For example, language learning ability is strongly associated with the age that
language learning starts, even when the amount of input is controlled (Johnson &
Newport, 1989; Lenneberg, 1967; Mayberry, 2007), which supports the view that the
learning rate may change over age. Furthermore, there are many different
phenomena in vision, auditory, and general cognition that have different sensitive
periods (e.g., Crawford, Harwerth, Smith, & von Noorden, 1996; Daw, 1994; Katz
and Shatz, 1996; Weisel & Hubel, 1965; Kral & Sharma, 2012), which suggests that
the learning rate changes may be governed by neural mechanisms that are present
across multiple brain areas (e.g., myelination, axon elaboration, synapse elimination,
Knudsen, 2004). Since a complete model of language must be able to explain better
learning abilities early in development, the same mechanisms that support that
phenomenon could also explain the larger surprisal effects early in development in
the present study.
Conclusion
Exploring the effect of verb syntactic preferences on priming in three age groups has
allowed us to begin to build a theory of verb-syntactic development in which the role
of the input as well as prior linguistic knowledge is considered. We observed
evidence of abstract structural priming across development and, although only adults
showed evidence of a lexical boost, both children and adults were sensitive to verb
bias effects in the form of target verb bias and prime surprisal effects. We suggest
that the lexical boost is not a reliable tool for making inferences about verb-syntactic
development and propose that children acquire abstract knowledge of structure
whilst also learning the lexical links to these structures from verbs. We also
tentatively suggest that this process is best explained by a model that uses an errorbased learning mechansim with a variable learning rate.
Acknowledgements
Financial support for this research was supported by a faculty-funded doctoral
studentship awarded to Michelle Peter by the University of Liverpool. We wish to
thank the schools and children who took part in this study, and to the students who
helped to collect the data. Thanks also go to the audiences at the AMLaP
conference (Riva del Garda, Italy) in September, 2012 and at BUCLD in November,
2012 for their helpful comments on an earlier version of this paper.
Appendix A
A1. Experimental items
Prime sentences before the slash are presented in the DOD structure; those after
the slash are in the PD structure. The four prime verbs are given in parentheses.
Target stems used the same combination of agent, verb, beneficiary, and theme as
prime sentences (e.g., the target stem, “Piglet GAVE...” was designed to elicit either
Piglet gave Tigger a puppy or Piglet gave a puppy to Tigger).
1. The king (GAVE/ SHOWED/ BROUGHT/ SENT) the queen a baby/ a baby to
the queen
2. The boy (GAVE/ SHOWED/ BROUGHT/ SENT) the girl a fish/ a fish to the girl
3. Wendy (GAVE/ SHOWED/ BROUGHT/ SENT) Bob a puppy/ a puppy to Bob
4. Dora (GAVE/ SHOWED/ BROUGHT/ SENT) Boots a rabbit/ a rabbit to Boots
5. The prince (GAVE/ SHOWED/ BROUGHT/ SENT) the princess a baby/ a
baby to the princess
6. The king (GAVE/ SHOWED/ BROUGHT/ SENT) the queen a cat/ a cat to the
queen
7. Piglet (GAVE/ SHOWED/ BROUGHT/ SENT) Tigger a puppy/ a puppy to
Tigger
8. Wendy (GAVE/ SHOWED/ BROUGHT/ SENT) Bob a rabbit/ a rabbit to Bob
A2. Filler items
1. Boots was flying
2. The princess jumped
3. Piglet and Tigger bounced
4. The king and queen waved
5. Tigger was washing
6. The prince was rocking
7. Piglet waved
8. The cat was swinging
9. Dora was flying
10. Bob was swinging
11. The princess and the cat were rocking
12. Dora and Boots waved
13. Bob was flying
14. The prince jumped
15. Tigger was rocking
16. The queen waved
17. The king and queen bounced
18. Piglet jumped
19. Wendy was flying
20. Dora was washing
21. The boy waved
22. Boots pointed at Dora
23. Wendy and Bob jumped
24. Dora was swinging
25. The girl waved
26. Wendy pointed at Bob
27. Boots was washing
28. Piglet was rocking
29. The cat bounced
30. Bob jumped
31. Boots waved at Dora and the baby
32. The king pointed at the queen
References
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modelling with
crossed random effects for subjects and items. Journal of Memory and
Language, 59, 390-412.
Bencini, G. M. L. & Valian, V. V. (2008). Abstract sentence representations in 3-year
olds: Evidence from language production and comprehension. Journal of
Memory and Language, 59, 97-113.
Bernolet, S. & Hartsuiker, R. J. (2010). Does verb bias modulate syntactic priming?
Cognition, 114, 455-461.
Bock, K. (1986b). Syntactic persistence in language production. Cognitive
Psychology, 18, 355-387.
Bock, K. & Loebell, H. (1990). Framing sentences. Cognition, 35, 1-39.
Campbell, A., & Tomasello, M. (2001). The acquisition of English dative
constructions. Applied Psycholinguistics, 22, 253-267.
Chang, F. (2002). Symbolically speaking: A connectionist model of sentence
production. Cognitive Science, 26, 609-651.
Chang, F. (2009). Learning to order words: A connectionist model heavy NP shift
and accessibility effects in Japanese and English. Journal of Memory and
Language, 61, 374-397.
Chang, F., Dell, G. S., & Bock, K. (2006). Becoming syntactic. Psychological Review,
113, 234-272.
Chang, F., Janciauskas, M., & Fitz, H. (2012). Language adaptation and learning:
Getting explicit about implicit learning. Language and Linguistics Compass, 6/5,
259-278.
Cleland, A. A. & Pickering, M. J. (2006). Do writing and speaking employ the same
syntactic representations? Journal of Memory and Language, 54, 185-198.
Coyle, J. M., & Kaschak, M. P. (2008). Patterns of ecperience with verbs affect longterm cumulative structural priming. Psychonomic Bulletin and Review, 15, 867970.
Crawford, M. L., Harwerth, R. S., Smith, E. L., & von Noorden, G. K. (1996). Loss of
stereopsis in monkeys following prismatic binocular dissociation during infancy.
Behavioural Brain Research, 79, 207-218.
Daw, N. W. (1994). Mechanisms of plasticity in the visual cortex. Investigative
Opthalmology and Visual Scicence, 35, 4168-4179.
Fisher, C. (2002a). The role of abstract syntactic knowledge in language acquisition:
a reply to Tomasello (2000). Cognition, 82, 259-278.
Gries, S. Th. (2005). Syntactic priming: A corpus-based approach. Journal of
Psycholinguistic Research, 34, 365-399.
Gries, S. Th., & Stefanowitsch, A. (2004a). Extending collostructional analysis: a
corpus-based perspective on ‘alternations’. International Journal of Corpus
Linguistics, 9, 97-129.
Gropen. J., Pinker, S., Hollander, M., Goldberg, R., & Wilson, R. (1989). The
learnability and acquisition of the dative alternation in English. Language, 65,
203-257.
Jaeger, T. F. (2008). Categorical data analysis: Away from ANOVAs (transformation
or not) and towards logit mixed models. Journal of Memory and Language, 59,
434-446.
Jaeger, T. F., & Snider, N. (2007). Implicit learning and syntactic persistence:
Surprisal and cumulativity. In L. Wolter and J.Thorson (Eds.), University of
Rochester Working Papers in the Language Sciences, 3(1), 26-44.
Johnson, J. S., & Newport, E. (1989). Critical period effects in second language
learning: The influence of maturational state on the acquisition of English as a
second language. Cognitive Psychology, 21, 60-99.
Katz, L. C., & Shatz, C. J. (1996). Synaptic activity and the construction of cortical
circuits. Science, 274, 1133–1138.
Knudsen, E. I. (2004) Sensitive periods in the development of the brain and
behavior. Journal of Cognitive Neuroscience, 16, 1412-1425.
Kral, A., & Sharma, A. (2012). Developmental neuroplasticity after cochlear
implantation. Trends in Neuroscience, 35, 111-122.
Lenneberg. E. H. (1967). Biological foundations of language. New York: Wiley.
Mayberry, R. I. (2007). When timing is everything: Age of first-language acquisition
effects on second-language learning. Applied Psycholinguistics, 28, 537–549
Messenger, K., Branigan, H. P., McLean, J. F., & Sorace, A. (2012). Is young
children’s passive syntax semantically constrained? Evidence from syntactic
priming. Journal of Memory and Language, 66, 568-586.
Naito, M. (1990). Repetition priming in children and adults: Age-related differences
between implicit and explicit memory. Journal of Experimental Child
Psychology, 50, 462–484.
Neter, J., Wasserman, W., & Kutner, M. H. (1985). Applied linear statistical models:
Regression, analysis of variance, and experimental designs. Homewood, IL:
Irwin.
Noppeney, U. & Price, C. J. (2004). An fMRI study of syntactic adaption. Journal of
Cognitive Neuroscience, 16, 702-713.
Pickering, M. J. & Branigan, H. P. (1998). The representation of verbs: Evidence
from syntactic priming in language production. Journal of Memory & Language,
39, 633-651.
Pine, J. M., Lieven, E. V. M., & Rowland, C. (1998). Comparing different models of
the English verb category. Linguistics, 36, 807-830.
Pinker, S. (1984). Language learnability and language development. Cambridge,
MA: Harvard University Press.
Rowland, C. F., Chang, F., Ambridge, B., Pine, J. M., & Lieven, E.V.M. (2012). The
development of abstract syntax: Evidence from structural priming and the
lexical boost. Cognition, 125, 49-63.
Shimpi, P. M., Gámez, P. B., Huttenlocher, J., & Vasilyeva, M. (2007). Syntactic
priming in 3- and 4-year-old children: Evidence for abstract representations of
transitive and dative forms. Developmental Psychology, 43, 1334-1346.
Sprondel, V., Kipp, K. H., & Mecklinger, A. (2011). Developmental changes in item
and source memory: Evidence from an ERP recognition memory study with
children, adolescents, and adults. Child Development, 82, 1638-1953.
Thothathiri, M. & Snedeker, J. (2008a). Syntactic priming during language
comprehension in three- and four-year-old children. Journal of Memory and
Language, 58, 188-213.
Tomasello, M. (1992). First verbs: a case study of early grammatical development.
New York: Cambridge University Press.
Tomasello, M. (2000). Do young children have adult syntactic competence?
Cognition, 74, 209-253.
Valian, V. (1986). Syntactic categories in the speech of young children.
Developmental Psychology, 22, 562-579.
Wiesel, T. N., & Hubel, D. H. (1965). Extent of recovery from the effects of visual
deprivation in kittens. Journal of Neurophysiology, 28, 1060–1072.